Kernighan Traces Unix, C Inventions to Shielded Bell Labs Monopoly

柯林汉：Unix与C语言的诞生源于贝尔实验室受保护的垄断地位 贝尔实验室之所以能孕育出Unix和C语言等计算领域的奠基性突破，是因为受法律保障的电话垄断收益使研究人员免受企业截止日期和绩效指标的干扰，赋予了科研人员长达数年不受约束的自主探索自由。 布莱恩·柯林汉（Brian Kernighan）回顾了他在贝尔实验室数十年的职业生涯，探讨了促成Unix、C语言和awk等奠基性技术在没有自上而下的截止日期要求下蓬勃发展的独特经济与组织条件。他讲述了与肯·汤普森（Ken Thompson）、丹尼斯·里奇（Dennis Ritchie）和理查德·汉明（Richard Hamming）等杰出人物共事的经历，并详细回忆了一些著名的历史轶事，例如为中央情报局（CIA）局长伪造拼写检查演示，以及佩戴假徽章通过安检等。转向他在普林斯顿大学的学术视角后，柯林汉探讨了当今大语言模型对计算机科学教育的影响、答疑时间的减少，以及人工智能代码验证在认识论层面的局限性。最后，他基于机缘巧合提出了职业建议，勉励年轻工程师培养广泛的兴趣，并积极抓住意料之外的机遇。 在AT&T受到法律保障的垄断利润资助下，贝尔实验室将全美业务收入的1%至2%直接投入到无固定期限的基础研究中，允许科学家在没有商业里程碑要求的情况下开展长达数年的探索性项目。 (6:34) 肯·汤普森在仅有8k内存的PDP-7上，用三周时间构建了最初的Unix操作系统、汇编器和文本编辑器，为丹尼斯·里奇的C语言凭借极小的运行时开销和硬件透明度取得行业主导地位奠定了基础。 (27:56) 普林斯顿大学的计算机科学专业入学人数大幅下滑，从2026届的200多人预计骤降至两年后的80至90人，原因是学生们转向硬件和机器人领域，并在遇到问题时依赖大语言模型而非参加教师的线下面对面答疑。 (55:51) 生成式AI代码生成引发了认识论上的无限倒退：尽管大语言模型在大约90%的情况下都能正常工作，但由于剩下的10%出错时缺乏明显破绽，导致不得不使用AI检查器来验证AI检查器，从而削弱了人工进行软件验证的技术技能。 (71:20)

In conversation with Ryan Peterman, computer scientist Brian Kernighan argues that foundational breakthroughs such as Unix and C emerged from a unique institutional ecosystem at Bell Labs, which combined monopoly-backed research freedom with practical, resource-constrained toolmaking. Looking at modern computing, he views artificial intelligence as another historical layer of abstraction, though one complicated by verification dilemmas and shifting academic incentives. The Economics and Culture of Bell Labs (6:34) Bell Labs operated as the research arm of AT&T, which held a regulated monopoly serving roughly 80% or more of the United States. In exchange for universal service commitments and rate caps, AT&T received a guaranteed return on capital, enabling it to funnel 1% to 2% of recurring revenues directly into unconstrained, fundamental research. Researchers worked without sprint cycles, corporate deadlines, or rigid project mandates, spending years pursuing open-ended questions. Quality was maintained not through layers of management, but through collegial annual reviews where technically active peers evaluated single-page summaries of work. Constraint-Driven Tool Design and Unix (27:56) The technologies developed at Bell Labs succeeded largely because their creators were their own primary users. Ken Thompson famously built an editor, an assembler, and the original Unix kernel in just three weeks on an obsolete PDP-7 minicomputer equipped with only 8k of 18-bit words. When Dennis Ritchie designed C around 1972, the target PDP-11 had only 24 kilobytes of memory and a 0.5-megabyte disk. This extreme resource scarcity compelled C to omit garbage collection and heavy runtime checks, creating a transparent, highly efficient language that mapped directly onto hardware instructions. AI as an Abstraction Layer and Validation Dilemmas (73:43) Kernighan contextualizes modern generative AI not as the death of programming, but as the latest step in computing's abstraction hierarchy, comparing it to the historic shift from handwritten assembly to Fortran and C. However, he highlights an epistemological dilemma inherent in large language model outputs. In his own experiments translating natural language into SQL, tools like Claude generated correct queries roughly 90% of the time, but the remaining 10% failed without obvious telltales. Because complex software cannot be fully vetted by hand, using automated tools to write and test code risks creating an infinite regress of unverified systems. Changing Educational Landscapes and CS Enrollments (55:51) At Princeton University, the rise of coding assistants and automated tools has catalyzed a dramatic shift in undergraduate career choices. While total university enrollment is fixed, Kernighan observes that computer science majors dropped from over 200 students in the class of 2026 to an anticipated 80 to 90 two years later. Seeking insulation from perceived AI displacement, students are instead pivoting to hardware-centric disciplines such as Electrical and Computer Engineering or Mechanical and Aerospace Engineering, drawn toward physical domains like robotics. Navigating Careers through Prepared Serendipity (86:15) Reflecting on his career path, Kernighan rejects the premise of deliberate career planning, comparing his trajectory to a marble descending a Galton board and bouncing unpredictably among pins. Drawing on Richard Hamming's advice to 'just say yes' to novel technical challenges, he maintains that while career collisions remain largely random, cultivating wide curiosity and learning fundamental concepts prepares an individual to capitalize on unexpected opportunities when they arise. While the corporate restructuring of AT&T ended the protected research sanctuary of Murray Hill, Kernighan suggests the rapid expansion of artificial intelligence may offer a comparable frontier of foundational problems, provided practitioners maintain the critical domain expertise needed to evaluate automated systems.

Brian Kernighan 在访谈中回顾了贝尔实验室以垄断利润支撑的基础研究文化与自下而上的工程创新，并结合 Unix、C 语言与 AWK 的诞生，探讨了计算抽象层的演进规律以及大语言模型对现代软件工程和计算机教育的深远冲击。 垄断经济学与贝尔实验室的无干预研究文化 (6:34) 贝尔实验室在 20 世纪下半叶催生了晶体管、信息论和 Unix 等里程碑式发明，其根本支撑在于 AT&T 当时受监管的公用事业垄断地位。AT&T 服务了美国约 80% 或更多的人口，作为承担普及通信义务并接受费率管制的交换，公司获得了法律保障的资本回报率。这使 AT&T 能将全国稳定收入的 1% 至 2% 直接拨给贝尔实验室，免除了来自季度财报和同业竞争的直接压力。管理层赋予技术人员极高自由度，允许研究人员在没有短期商业化指标的情况下探索两到三年，坚信顶尖头脑的聚合必将产生巨大的实用价值。 在日常治理上，研究部门几乎全员享有 Member of Technical Staff（MTS）职衔，日常缺乏条条框框，员工上下班弹性极大，并长期保留敞开办公室大门、在黑板前自发辩论的学术传统。研究人员往往抗拒晋升为管理层，认为脱离动手开发即是技术生命的终结。绩效评估摒弃了繁琐的细粒度追踪，仅要求员工每年提交单页自我总结（俗称“I am great”报告），由仍保持技术敏锐度的前线管理者评议其硬核贡献，从而有效规避了形式主义与平庸产出。 Unix与C语言：极端硬件约束下的效率与表达力 (27:56) Kernighan 盛赞 Ken Thompson 是其见过最顶尖的程序员，并以 Thompson 在 1969 年仅用三周独自在过时的 PDP-7（仅有 8k 18位字长内存）上用汇编语言手写编辑器、汇编器和操作系统为例，展现了实验室同仁惊人的工程执行力。此外，Doug McIlroy 奠定了宏处理器与 Unix 管道概念，并编写了 diff 和 sort 等核心工具。这些早期项目多源于自身工作痛点，如 Kernighan 与 Alfred Aho、Peter Weinberger 在 1976 至 1977 年联合打造的 AWK，便是融合了正则文本匹配与报表数值计算需求的产物。 针对 C 语言何以长期主导系统编程，Kernighan 分析其核心在于找到了低级效率与人类可读性之间的“黄金平衡点”。Dennis Ritchie 于 1972 年开发 C 语言时，目标硬件 PDP-11 仅有 24 KB 内存和 0.5 MB 磁盘，既要容纳操作系统、编译器又要留存文件缓冲区。极端严苛的硬件环境迫使 C 语言剔除垃圾回收及重型运行时安全检查，让程序员能直观地将语句映射到底层机器指令。C 与 Unix 形成高度共生关系，两者凭借简洁性与愉悦的开发体验被业界广泛采纳。 大模型冲击、无限递归验证与教育格局演变 (53:09) Kernighan 认为大语言模型（LLM）并非编程的终结，而是计算机抽象层向上递进的又一阶段，正如当年 Fortran 取代手写汇编、高级语言弱化手工 C 编程一样。然而，LLM 生成代码引发了棘手的认识论困境：例如在利用 Claude 生成 SQL 场景中，尽管大约 90% 的情况下输出完全可用，但剩余 10% 存在隐蔽且严重的错误，若使用者缺乏领域专业知识便极难排查；在大型软件工程中，由于人类无法手写覆盖所有测试，依赖 AI 生成测试用例又引发了“谁来检验检验者”的无限递归验证难题，且提示词生成剥离了手工排错调试的智力乐趣。 在普林斯顿大学的教学实践中，Kernighan 观察到 LLM 的普及正在重塑教育生态。即时问答工具使得学生访问教师答疑时间（office hours）的频率骤减，削弱了深度师生联结。同时，由于担忧纯软件岗位被自动化取代，本科生大量转投电气与计算机工程（ECE）或机械航空航天工程（MAE）等与物理硬件及机器人紧密相关的学科。普林斯顿 CS 专业规模预计将从 2026 届的 200 余人大幅锐减至两年后的 80 到 90 人，生源出现结构性向外分流。 早期黑客文化轶事与人生的高尔顿板隐喻 (63:45) 访谈还原了贝尔实验室早期的反叛黑客传统与安全生态。为抵制管理层强制佩戴员工胸牌的官僚举措，研究人员虚构了名为“Grace Emlen”（源自小精灵 gremlin）的虚拟员工并为其伪造胸卡，甚至将其列入官方人事档案；Kernighan 本人则常年佩戴米老鼠胸牌通行，连会见比尔·盖茨时也未被察觉。另外，早期 Unix 缺乏网络互联，安全防范极低，Kernighan 的个人密码一度被设置为极易击键的“/.,/.,”，数十年后老旧 secrets 文件的弱加密算法被现代算力破解，成为同行间的公开趣谈。 总结个人学术与著述生涯时，Kernighan 借用了高尔顿板（Galton board）的弹珠下落隐喻：个体的生涯走向充满随机碰撞与不可预测的跳跃，并非源自精密的前期规划。但他同时援引巴斯德的名言“机遇偏爱有准备的头脑”，并推崇 Richard Hamming“积极说好”（just say yes）的开放哲学。他建议年轻一代广泛涉猎、积极试错，在好奇心驱动的探索中赋予人生弹珠以正向的偏向力，从而在未来的不确定性中抓住偶发的战略机遇。 贝尔实验室以垄断租金孕育自下而上基础创新的制度环境虽已随企业拆分与资金碎片化消逝，但其推崇的极简工程主义依然深植于当代基础软件。面对大模型推动的高级抽象与教育分流，人类工程师的核心职责将进一步从手工代码编写转向对输出质量的高阶裁决与系统化验证。

[53:09] Brian Kernighan: The bandwidth when you're talking to somebody in person is enormously higher than when you're talking to Claude or whatever.

[0:00] And so in three weeks he builds himself [0:01] an editor, an assembler and an operating [0:04] system. And there you have it [0:05] >> by himself. [0:06] >> Yeah. Oh yeah, absolutely. [0:08] >> This is Brian Kernneahan, Exbell Labs [0:10] technical staff, co-author of the [0:12] original Cbook, and the man who [0:14] popularized Hello World and I asked him [0:16] all about what it was like working [0:18] there. [0:18] >> There were no deadlines. There were no [0:20] projects. There were no objectives, [0:22] goals, whatever set by other people. [0:25] >> There was a secrets file that was [0:27] committed to the repo. And so you and [0:30] all your colleagues passwords are [0:32] actually [0:33] >> Yeah. [0:33] >> So I saw yours is do you believe that [0:38] LMS are a net negative or a net [0:40] positive? [0:41] >> So who shall I insult today? [0:43] >> Here's the full episode. [0:49] >> Could you explain what Bell Labs was and [0:52] what is the story behind you getting a [0:54] job there? So Bell Labs uh was like the [0:59] uh research arm of AT&T. It was founded [1:04] in I think 1923 25 something like that. [1:08] AT&T at that point was growing. It had [1:10] absorbed a bunch of the local telephone [1:12] companies across the whole country and [1:14] they realized they just needed to [1:15] understand things better so that they [1:17] could improve service. So they set up a [1:19] research operation which was called Bell [1:21] Telephone Laboratories. It was in New [1:23] York on West Street. Um and people did [1:28] all kinds of things related to improving [1:31] telefan. [1:32] Um in 19 call it 63 something like that [1:37] early 60s they basically moved the whole [1:41] operation to New Jersey northern New [1:43] Jersey a place called Murray Hill which [1:47] near Summit and things like that. So [1:48] call it 25 miles west of New York City. [1:51] Um and that was a big operation. There [1:53] were probably 3,000 people in one [1:56] connected set of buildings doing um [1:59] research in all kinds of things. Uh [2:03] physics, semiconductor physics, devices [2:05] like that, chemistry, optical fiber [2:08] kinds of things, just all sorts of [2:10] research. And then uh more on the soft [2:12] end there was mathematics and [2:14] increasingly there was computers [2:16] computer related things as they [2:18] gradually came to the realization that [2:19] computing was going to be part of the [2:21] future in some way or another. In the [2:25] summer of 1966, [2:28] I was a grad student here at Princeton, [2:30] but I got an internship at MIT in [2:33] Cambridge working in project MAC, which [2:36] was basically the Multix project. And so [2:40] I used CTSS, that first original time [2:43] sharing system done by uh Corby and [2:47] friends to develop software for the [2:51] Multic system which was just just [2:53] starting up. And that was a lot of fun. [2:57] I enjoyed it. I'm not sure I ever did [2:59] anything that was very useful, but but [3:01] part of it was then the next year I got [3:05] an internship at Dell Labs in Murray [3:08] Hill. So I was still a graduate student [3:10] and that would be roughly the third year [3:13] here. I got this internship at Bell Labs [3:16] and that was a lot of fun and I worked [3:18] there. I worked actually with Doug Mroy [3:22] um for the summer came back here [3:25] continued to stall on doing my thesis [3:28] work which I guess is what one does. And [3:31] then I got another internship in the [3:33] summer of 1968 and there I worked with [3:35] Shan Lynn and he and I worked on a [3:37] problem which had been suggested by my [3:39] adviser here and this is graph [3:42] partitioning. Um and he and I worked on [3:45] it and we came up with an algorithm that [3:48] did a pretty decent job on part of it. I [3:50] came up with several special cases that [3:52] you could do really well and then I [3:55] wrote up the thesis really quickly and [3:57] escaped [3:59] [laughter] [3:59] >> escaped Princeton. escape from [4:01] Princeton. I took a whole week off and [4:03] went to Bell Labs because they had given [4:04] me a return offer which was very sweet [4:06] of them. Um, and I stayed there then [4:08] until roughly 2000 in the [snorts] group [4:13] that had done or was doing a variety of [4:15] things. [4:16] >> Did you consider other career [4:18] opportunities at the time? [4:19] >> You know, I think I didn't consider any [4:22] other option whatsoever. I I mean, I had [4:24] really enjoyed the the summers. The two [4:26] summers were great. The people were [4:28] really, really nice and interesting. [4:30] There was stuff to do. Uh, it was close. [4:32] I didn't have to go anywhere as it were. [4:35] And so, I literally never considered [4:39] anything else. I never had a job [4:40] interview, nothing like that. I just [4:42] went there. So, it was kind of like [4:44] falling down the fall line or something [4:46] like that. I think partly it was helped [4:49] along by some sort of old boy network [4:52] where somebody here at Princeton knew [4:54] somebody at Bell Labs because there's a [4:55] lot of back and forth between the two [4:58] institutions and so I'm sure that that [5:02] kind of thing helped. It also helped I [5:05] think at MIT that a guy who had had an [5:09] internship the year before me had done a [5:11] bangup job for them and so that probably [5:14] these folks from Princeton are maybe [5:16] okay. Um, so I'm grateful to him too. [5:20] >> Nowadays, Bell Labs is so legendary. But [5:24] at that time, what was the social [5:26] perception? If I were to have talked to [5:28] your peers and said, "Hey, what do you [5:30] think of Bell Labs?" What would they [5:31] have said at that time? [5:33] >> I suspect that it depended on the field, [5:35] but probably for most of them, they [5:36] would have thought, "Yeah, yeah, this is [5:37] a really good place." Because in the [5:39] physical sciences, it was obviously the [5:42] powerhouse in a lot of respects. That's [5:44] where your transistor came from, for [5:46] example. Um, and a lot of things related [5:49] to communications. That's where [5:50] information theory came from. It's where [5:53] error correcting codes came from. All [5:54] these things that were sort of critical [5:57] advances in technology had come from [5:59] Bell Labs. Bell Labs was a part of the [6:01] scientific community. People who went to [6:04] Bell Labs published in exactly the same [6:06] journals, went to the same conferences, [6:09] collaborated with people in the same [6:11] universities. And so it was really an [6:13] integral part. I think of the scientific [6:15] community at the time and that was you [6:18] know largely I will say the physical [6:21] sciences kind of thing but it obviously [6:23] sort of spread out into other things [6:26] that were peripherally related and [6:27] certainly the mathematicians were [6:29] absolute first rate as well tended to be [6:31] applied mathematicians rather than [6:33] theoretical ones. Everything you just [6:36] said there is in insane value for [6:40] humanity, the transistor and all of [6:42] those other things. Why is it that so [6:46] much of it came from one research lab [6:49] tied to a company that today is not [6:52] known for all these amazing innovations. [6:56] >> AT&T bankrolled all kinds of [6:59] contributions that Bell Labs made. And I [7:01] think the way it worked was that AT&T [7:03] provided telephone service to [7:05] essentially the whole country. I mean [7:06] maybe it was 80% but it was very [7:08] substantial fraction of the whole [7:09] country. [snorts] And that meant that [7:12] they had a pretty guaranteed revenue [7:15] stream. Now it was regulated. They were [7:17] a regulated public monopoly. That's the [7:19] the jargon term. And that meant that [7:21] their rates were set or restricted or at [7:25] least evaluated by lots and lots of [7:27] agencies, states and the federal [7:29] government. So there was a limit to what [7:31] they could do. The quid proquo was they [7:34] got sort of a guaranteed rate of return [7:37] on their investment in whatever they did [7:39] with the phone company. And the quid [7:41] proquo would be that they provided [7:43] telephone service to essentially [7:45] everybody. Even if you were way out in [7:48] the boonies or something like that, AT&T [7:49] would provide the service. and they were [7:52] able to rip off a small fraction that I [7:54] will make up a number and say 1 or 2% of [7:57] their pretty much guaranteed revenues to [8:00] fund a research operation that would let [8:03] them improve the service. And so that [8:05] was the the basic call it financial [8:08] arrangement. And the other thing is it [8:10] was very stable. you could sort of [8:12] predict there would be this steady [8:13] increase in um revenues [snorts] [8:17] but predictable in a way that would [8:19] guarantee that there would always be [8:21] money to do something. There was no [8:23] competition because AT&T had essentially [8:26] what amounted to this monopoly and so [8:28] that meant that they could take a very [8:30] long view of what they were doing. And [8:33] so somebody had an idea and wanted to [8:35] work on it for a year or two or three [8:38] fine go for it. And if you got a bunch [8:40] of people doing this, even if your idea [8:42] is dumb, my idea is dumb, somebody else [8:44] might have a good idea. And so if you [8:46] aggregate that, the output would be [8:50] probably something useful almost all the [8:52] time. [8:53] >> What gave this lab such recruiting pull [8:56] in your opinion? Like how did how are [8:58] all the best people coming to this [9:01] place? [9:02] >> Well, I think it depends. For me, it was [9:03] certainly happen stance, just blind [9:05] luck, being in the right place at the [9:07] right time, following people who had [9:08] done something that in some way prepared [9:10] the way. I think if you were a [9:13] physicist, getting your degree in [9:15] physics at pick your favorite good [9:16] university, uh where would you go to do [9:18] physics? Well, the opportunities would [9:20] fundamentally be government lab perhaps [9:24] uh or a place like maybe IBM or maybe [9:30] Bell Labs. and Bell Labs had a sterling [9:33] reputation, well-earned for doing [9:36] exceptional work in that area. So, so [9:40] that would be let's say anybody doing [9:42] physics, chemistry, materials, all of [9:43] that kind of stuff. Um, I think somewhat [9:46] the same would be true of anything [9:48] related to mathematics, especially the [9:50] mathematics of communication or anything [9:51] related to communication systems. That's [9:53] where Shannon was, that's where Hamming [9:55] was, uh, Bodh Nyquist, all of those [9:59] people before those. So that kind again [10:02] tremendous reputation for the place and [10:04] as I said they were part of the [10:05] scientific community. It wasn't like it [10:07] was off in Mars somewhere and it wasn't [10:09] like you went there and disappeared. You [10:11] went there and you were part of the [10:13] broader community. And so I think the [10:15] computing stuff came along and had some [10:17] of the same advantages. But it wasn't [10:20] clear who the heroes were at that point. [10:23] It was earlyish because remember I was I [10:25] went there in 1969. So this was [10:28] computers were still pretty young at [10:30] that point. [10:31] >> A lot of those names you mentioned, I [10:33] mean they were in my, you know, [10:37] textbooks when I was studying and so [10:39] they're they're kind of the heroes and [10:40] and everything. But so at that time [10:43] you're saying your perspective was they [10:45] were just kind of the lead on the team, [10:47] but they weren't the legendary computer [10:51] scientist. Well, that's it's hard to say [10:52] and it it's not even clear what a [10:54] computer scientist is, but you want a [10:55] story. When I went to Bell Labs for my [10:58] first internship, so call it the summer [11:00] of 1967 [11:02] and I was in an office that was sort of [11:06] a little smaller than the one we're [11:07] sitting in now, but as a summer intern, [11:09] for some accidental reason, I had an [11:11] office of my own. This never happens, [11:14] but it did happen. And so I'm sitting [11:16] there on the very first day at about, [11:18] you know, roughly [11:20] 11 o'clock in the morning wondering what [11:22] the hell do I do? And this older guy [11:24] came along, passed the door, and he [11:26] said, "Hi, I'm Dick. Let's go to lunch." [11:29] I said, "Okay." [11:31] So I went to lunch with Dick. I never [11:33] did get his name right. And so we had a [11:35] nice lunch conversation. I don't [11:36] remember anything about it. And then he [11:37] went off somewhere else. And I came back [11:40] and I went the next door to me and [11:43] looked at the name tag on the door. It [11:45] was Dick Hamming. Hamming had written [11:48] the book that I had used previous [11:50] semester for numerical analysis and of [11:54] course error correcting codes. And there [11:56] he was next door. Super nice guy. Um and [12:02] was he a computer scientist? Borderline. [12:05] But possibly he did win the third Turing [12:08] award. [12:09] uh won it was it that summer or the next [12:11] summer I'd have to go and look and so [12:13] there were people like that around but [12:16] in the very early days the computing [12:18] stuff was I think it hadn't developed [12:21] into something with its own academic [12:23] structure its own reward systems and all [12:25] those kinds of things [12:27] >> when I interviewed Leslie Lampport and I [12:30] forgot the his exact timelines but I [12:32] remember early in his education computer [12:35] science wasn't established and he [12:38] actually said it was not not as [12:42] respected as math and physics or maybe [12:44] because it just was a burgeoning that's [12:46] brand new. The same thing happened here. [12:49] I think I came here as a grad student in [12:51] 1964 and I came in electrical [12:54] engineering. I had thought to work on [12:55] computing because I had some notion that [12:57] it was interesting and I thought AI [13:00] I didn't know a thing. That's [13:01] unbelievably naive. Um, and I had sought [13:05] to work with a guy uh named Arthur [13:08] Samuels who had done some interesting [13:09] stuff in computer chess playing etc. And [13:12] I got here discovered I was the wrong [13:14] guy. Um, but when I got here the [13:18] electrical engineering department had a [13:19] bunch of very established, highly [13:22] respected people who were doing [13:23] traditional electrical engineering [13:25] things, information theory, [13:26] communications, solid state physics, [13:29] that kind of stuff. and a cohort of [13:31] young people who had interest in [13:34] computing. They had just gotten their [13:36] degrees from someplace and they were [13:38] very young, wanted to do computing [13:42] and I don't know whether any of them [13:43] ever got tenure because the older folks [13:45] didn't really gro the newer field. [13:49] John Hopcrush to pick a random example [13:52] was my technical advisor for a year. [13:55] Although it was quickly discovered I was [13:58] not a theoretician. [13:59] >> [laughter] [14:00] >> But you know, John went on to win a [14:02] touring award and was dean of [14:04] engineering at Cornell. There's all [14:05] kinds of stuff. Great guy. Uh but he [14:08] wouldn't have gotten tenure here at that [14:11] particular time because the the [14:13] computing science aspect things was too [14:15] new. [14:16] >> Bell Labs is kind of famous for very [14:20] hands-off management and letting the [14:22] engineers kind of do what they were [14:24] doing. What is a day in the life like of [14:27] a engineer working at Bell Labs? [14:30] >> Well, first thing I don't think any of [14:32] us would have called ourselves [14:33] engineers. Uh we were, you know, doing [14:36] whatever you did, but but I don't think [14:39] that engineer was the right word because [14:41] we weren't really engineering. [14:43] >> It was member of technical staff. [14:45] >> Member of technical staff, which was a [14:46] term that applied to everybody [14:48] essentially. Um, and a typical day would [14:54] be basically you'd come in at whatever [14:56] time, which could be anywhere from, you [14:58] know, 8 in the morning to 2 or 3 in the [15:01] afternoon maybe, [15:03] [snorts] and you would sit in your [15:05] office, private office, um, with the [15:08] door open, and you would think about [15:10] whatever it was you were thinking about. [15:12] Maybe you had been working on some kind [15:14] of program, or maybe you were trying to [15:15] prove a theorem. That's not me. Uh, and [15:19] people would wander by and you would [15:22] talk to them. They would come in. You'd [15:23] have discussions on the blackboard [15:25] perhaps of technical things or maybe [15:27] just armwaving discussions of something [15:28] else. You'd walk up and down the very [15:31] long main corridor to the coffee machine [15:33] at the other end. You would run into [15:36] random people. There would be technical [15:38] talks from time to time, maybe one every [15:41] week, two, something like that, [15:42] depending. Uh, sometimes from internal [15:44] people, sometimes from external people. [15:47] Um, so in a sense, maybe not too [15:50] different from what you would find in a [15:52] university, except there were no [15:53] students. And so that meant you could do [15:56] it full-time. [15:57] The the trade-off was that without the [16:00] students, you could do more of your own [16:02] thing and you were doing it with your [16:04] peers. [16:06] So, you know, I would argue with [16:08] somebody, Ken [snorts] Thompson or [16:10] Dennis Richie or whatever like that who [16:12] were my peers. I didn't have to work [16:14] with students. But of course the flip [16:16] side is that I didn't get any of the [16:17] benefit that you get with bright young [16:19] new people talking about it except that [16:22] as a newcomer there I was supposedly the [16:24] bright new young person. [16:26] Uh so mainly the lack of students which [16:28] I think is a mixed blessing [16:31] >> in industry there's often deadlines [16:35] managers and all of that stuff right [16:38] what what's stopping a researcher from [16:41] coming in at three leaving at 5 not [16:44] having any output [16:46] >> there was no immediate stop to that [16:48] period could there were no deadlines [16:50] there were no projects there were no [16:54] objectives goals whatever set by other [16:56] The mechanism for deciding whether you [16:59] were doing anything was that once a year [17:02] you had to write down on one side of one [17:05] piece of paper what you had done [17:07] [snorts] and we called it the I am great [17:09] report and it just summarized what you'd [17:12] done. Um and then the first and second [17:14] level of management would sit in a room [17:17] with the I am great reports from their [17:19] people. So this would be maybe 30 at the [17:22] bottom, half a dozen, one like that [17:25] little tree and they would discuss the [17:28] merit of the the contributions of the [17:31] work that was being done by the people [17:35] and they would come up with some sort of [17:36] ordering which says you're better than I [17:39] am or vice versa. how would we compare [17:41] to our peers and that would give them a [17:43] sort of roughly a rank orderering or at [17:45] least equivalence classes of people and [17:49] if somebody hadn't done anything for a [17:50] couple of years that would be noticed [17:53] but if somebody' gone off on something [17:54] and it didn't pay off very well or it [17:56] was still in progress but it looked okay [17:58] or whatever that was okay that's the [18:01] long view that I mentioned before it's [18:03] not like what have you done for us the [18:05] last week or even quarter what maybe [18:08] have you done over the last two three [18:10] ers and it was being assessed by people [18:13] who were not driven by quarterly [18:15] numbers. They were the same people as we [18:18] were except that they had moved into [18:20] management but they were still [18:21] technically active and they were very [18:23] technically knowledgeable and then the [18:25] process was repeated a couple of levels [18:27] up until they had any effect or ranking [18:29] of everybody. So that was kind of the [18:32] call abstract merit you know you know [18:34] I'm better than you are. And then there [18:35] was the concrete merit phase where we've [18:37] got this much money how much do you get? [18:38] how much do I get? Uh, [18:41] >> oh, so there were bonuses or tied to [18:44] >> It wasn't bonuses or anything like that. [18:46] It was simply there's a pot of money. [18:48] This is the raise pool and it will be [18:50] over all x% and so you will get x minus [18:53] epsilon. I'll get x plus epsilon or [18:55] vice. [18:57] Oh, and that was once a year and [19:01] that was the whole story. [19:03] But it was it was very deliberative. The [19:06] management was colleial. [19:08] >> [clears throat] [19:08] >> If you and I were managers and we had [19:10] our people, it wasn't like my people are [19:12] better than your people. It's rather, [19:13] hey, you forgot to mention that, you [19:15] know, Susie did this really great thing. [19:17] Let's don't forget Susie kind of thing. [19:20] So, it was quite colleial and you were [19:22] be evaluated by people who understood [19:24] the work and whose goals and aspirations [19:27] and so on were the same as yours, you [19:28] know, do good stuff. [19:30] >> It sounds like you mentioned that their [19:32] technical is like a big part of this [19:35] whole system. [19:36] Why is that important? [19:38] >> If somebody is doing something in a [19:41] technical area, they have created a new [19:43] program, they've proved a new theorem, [19:44] they developed a new process, whatever. [19:47] If you are not technical enough to [19:49] understand what they're talking about, [19:52] it's hard for you to tell the [19:54] from the actual useful content, right? [19:56] So, I could not you personally, but I [19:59] could blow smoke pretty well for people [20:01] who were I don't who shall I insult? [20:04] MBAs. Okay, something like that. That's [20:06] obviously a slander, but uh but it's [20:10] easier to blow smoke at somebody who is [20:12] not sort of knowledgeable in your own [20:15] field, but the people that you were t [20:17] who were evaluating your work were [20:19] exceptionally knowledgeable in the whole [20:22] field. And so they had a pretty good eye [20:26] for what was good stuff and what wasn't [20:29] yet good stuff or would perhaps never be [20:31] good stuff. these days this um member of [20:34] technical staff title is it's it's [20:38] almost has its own brand where it's it's [20:42] a well it's a nod towards Bell Labs and [20:44] everything that that represented there [20:46] but a lot of what it is signaling is [20:49] that titles don't matter that we're all [20:52] the same and we're all just focused on [20:53] the technical work did titles matter or [20:56] what was the culture at that time [20:58] >> titles didn't matter um everybody was on [21:01] a firstname basis. There were no doctor, [21:04] professor, whatever kind of titles at [21:06] all. Um, everybody was technically a [21:10] member of staff, member of technical [21:13] staff. I and I say everybody, this is of [21:14] course an oversimplification. there were [21:16] a variety of supporting people who you [21:19] know lab technicians associate members [21:21] of technical staff or whatever but I'm [21:22] talking about the MTS level and um uh so [21:26] they would be loosely equivalent to PIs [21:29] in the modern academic or principal [21:31] investigator whatever [21:33] and in computing I never would never [21:37] have had somebody who helped me with my [21:39] stuff whereas if I were a physicist I [21:41] might have some number of people who [21:43] would help me do experiments look after [21:45] the you know mechanics of you know [21:49] depositing stuff on stuff all those [21:51] kinds of things in a physical world [21:54] but everybody was technically a member [21:56] of technical staff in the management but [21:58] then there were department heads who [22:00] manage first level management and then [22:02] there were directors who were second [22:04] level management and then there was [22:06] executive director and then there was [22:07] vice president but it was very flat uh [22:10] in that sense most people did not want [22:12] to be promoted because that [22:15] would be seen as the end of their [22:17] personal [22:18] technical career that it would move them [22:20] away from the kind of work that they [22:22] really enjoyed doing. So being promoted [22:24] was something to be avoided, not to be [22:28] sought. Uh and I think that went [22:33] for quite a long time. Now things did [22:35] change, but I'm talking about the golden [22:37] era of call it the late 60s,7s, early [22:40] 80s, something like that. [22:42] Did you and your peers ever think about [22:45] the notion of career development in some [22:48] way or it was not even something you [22:51] cared about? [22:52] >> Career development for myself as an [22:54] individual I never thought about at all. [22:56] I mean it was I was having a good time [22:57] and I was surrounded by lots of good [22:59] people. Uh career development for uh the [23:02] people for example when I was in first [23:04] level management I was a department head [23:06] for some years. I don't remember much [23:08] about how explicit it was, but there was [23:11] some theory that you should actually be [23:13] looking after the interests of the [23:15] people in your department. I think the [23:18] main way that that would be reflected is [23:21] understanding the work they were doing, [23:23] really trying to understand what it was [23:24] they were doing so that you could [23:26] represent it fairly at things like merit [23:28] review. Also, so you could see [23:30] opportunities to, you know, oh gee, did [23:32] you know Dave was working on this stuff [23:36] so that you could sort of cross [23:37] fertilize across the way um to try and [23:41] give them feedback sort of to say, you [23:42] know, are you sure this is going to work [23:45] or keep an eye on this? Sort of guide [23:48] them towards future success. Uh but not [23:51] telling them what to do. There was never [23:53] any this is what you have to do. there [23:55] was no no uh there just wasn't any it [23:59] wasn't guidance at all at most there was [24:01] a gentle gravitational pull towards [24:03] stuff that would be useful or [24:04] interesting. [24:05] >> So you mentioned you were ahead at some [24:07] point and you managed some reports you [24:09] never had a report that was um saying I [24:12] want to grow in some way or they're kind [24:14] of hustling you know kind of pushing you [24:18] some way. [24:19] >> There were a handful of people I could [24:20] remember who wanted to be in management. [24:23] Okay, that was unusual, but there was [24:25] nothing wrong with it. It's just that [24:27] being in research management because the [24:30] structure was so flat [24:33] that it would take probably quite a [24:36] while before you would move up. If you [24:38] really wanted to be in management, you [24:40] probably wanted to move into a different [24:42] organization, development organization, [24:44] where the rules of engagement and so on [24:47] were much different. And there were [24:48] several I shouldn't say several but at [24:50] least a handful of people who were [24:52] interested in that kind of advancement [24:54] and they went into development type jobs [24:58] which are perfectly fine just differed [25:00] in a lot of ways and there they had the [25:02] much more traditional responsibilities [25:05] of management and including looking [25:07] after their people but I think the [25:09] research operation was much more a bunch [25:13] of cats and you can't really hurt them. [25:15] [laughter] [25:16] So what's the story behind you becoming [25:18] a manager in that kind of environment [25:20] because seems like it's something [25:22] everyone's avoiding. [25:24] >> I had no desire to do it. Um and I was [25:27] blandished a couple of times. Wouldn't [25:30] you like to be a manager, but it was in [25:32] some other organization even within [25:33] research and I always thought the the [25:35] group that I was in was special. [25:37] Probably was. Uh and so I had no [25:40] interest in being manager in some other [25:43] part of it. And so I basically declined [25:44] with thanks. Um but then it was would [25:48] you like to be a manager in your own [25:50] organization? And I wasn't very [25:52] enthusiastic but okay. And they got lots [25:55] of ways they can you know it's [25:58] inevitable. Why not now or alternately [26:02] you know if you don't do it now it might [26:03] never come but [laughter] you're [26:05] completely contradictory [26:07] [gasps] or if if you don't do it [26:09] somebody else not as good is going to do [26:10] it you know. all of the usual stuff and [26:12] I finally succumbed I think in about 81. [26:15] So I'd been there at that point, call it [26:17] 12 years for real. So that's the first [26:20] promotion and [26:22] and it was okay but it wasn't as much [26:25] fun I think because now I was [26:26] responsible for you know roughly 10 [26:28] people had to understand their work [26:31] which was all over the map. Had to be [26:33] able to talk to other people about it. [26:35] Had to go to more meetings. [26:39] So, it was okay, but it was not my [26:41] favorite part of the job really. [26:44] >> How did you know your group was so [26:46] special? Because when I look at it, it [26:48] it it is in hindsight. I mean, I saw [26:50] Bejarn was reporting to you and of [26:53] course there's Dennis Richie nearby and [26:55] Ken Thompson and you also mentioned [26:58] other people like Mroy and Hamming. So, [27:01] how how did you know though at that time [27:03] that it was such a special group? [27:06] I think a lot of it was not so much [27:08] knowing that the work they were doing [27:10] was going to prove to be prove to be in [27:12] the future important, but that they were [27:14] just really clearly smart and and seem [27:17] to be doing things that were interesting [27:19] and relevant. Most of us used the work [27:22] that our peers did. Certainly the all of [27:25] the things related to Unix, we just used [27:27] that. That was our computing [27:28] environment. The languages that were [27:30] developed, the tools that were developed [27:32] and so on, we used them. And so you [27:34] could get a sense of who's doing stuff [27:36] that actually matters to me. And and so [27:39] I think that was part of it. Uh [27:43] extrapolating long-term, [27:46] you know, effects. We're not very good [27:48] at that. Or I certainly was and figuring [27:50] out who were the ones who were going to [27:51] win touring awards. No, [27:55] >> not really. [27:56] >> You worked with so many legendary [27:59] people. Was there anyone that I guess or [28:02] any story of someone who kind of [28:05] impressed you the most or you thought, [28:06] "Wow, their their technical chops are [28:08] just untouchable." [28:10] >> There are several like that. Um Ken [28:13] Thompson's the obvious one. I think [28:16] certainly the best programmer I've ever [28:17] met, unparalleled. [28:20] And the story is the first version of [28:23] Unix where he's got this little old PDP7 [28:28] which is an obsolete machine has [28:29] something like 8k 18 bit words so it's [28:33] you know tiny little machine um and he's [28:38] been fooling around with something to [28:40] drive the disc on it and he's thinking [28:42] about you know how do you schedule disc [28:44] operations so that it'll go fast and [28:45] stuff like that. Um, and his wife takes [28:48] their young son off to California for [28:51] three weeks and he thinks I could do [28:53] something here. And so in three weeks he [28:55] builds himself an editor, an assembler [28:57] and an operating system. And there you [28:58] have it [snorts] [29:00] [laughter] [29:00] >> by himself. [29:01] >> Yeah. Oh yeah, absolutely. [29:03] >> When I used to write, you go to, you [29:05] know, Stack Overflow or you kind of like [29:07] look up how to write the code and then [29:10] you write it. How how does he do it [29:14] without all that internet resource? [29:17] >> Because he's a boatload smarter than you [29:18] and I put together. [laughter] [29:21] And this is before any of that stuff [29:22] existed, right? This was 1969. [29:25] And uh so he's probably writing the code [29:28] on in assembly language, bear in mind, [29:30] not not any high level language. And [29:32] he's doing it probably on some other [29:34] machine just because it has some kind of [29:36] text editor. And then he's making paper [29:38] tapes that he carries to this tiny [29:40] little machine and loads them and sees [29:43] what happens. [29:44] >> Oh my god. [29:45] >> Yeah. So, it's going back into the mists [29:47] of time. But three weeks and he's got a [29:49] working operating system. [29:51] >> So, I think that is a good example of [29:55] somebody who's just offscale good at [29:57] what they do. And there were a lot of [29:59] people like that at the labs. Um, you [30:03] know, I mentioned Doug Mroy a number of [30:05] times. He was [30:07] my first internship. He was notionally [30:09] my boss and he was interested in storage [30:11] allocators, you know, ML kind of thing [30:14] and uh so he set me to working on [30:18] something and I almost instantaneously [30:19] went somewhere else [laughter] [30:22] and he didn't care. So I spent the [30:24] summer doing something unrelated to that [30:26] and it seemed like it was fine because I [30:28] got a re come back next summer kind of [30:30] offer. But he was the kind of guy who [30:33] was exceptionally good at pretty much [30:35] anything he wanted to do. I mean, Ken [30:37] Thompson says Doug's the smartest person [30:39] he ever met. [laughter] So, this is [30:44] [snorts] we're multiple levels above me [30:46] at this point. Um, but Doug was [30:48] extremely good at uh assessing [30:52] experimenting with criticizing other [30:54] people's work in a very constructive [30:57] way. So he was a second read or the [30:59] outside reader on my PhD thesis and he [31:02] greatly improved my writing and the [31:04] content of the thesis and he continued [31:06] to do that essentially for his whole [31:08] life. He's still doing it and he [31:11] contributed a bunch of useful software [31:13] of his own Unix sort program Doug Unix [31:17] diff program compares to files Doug [31:20] bunch of things of of that sort. Um he [31:23] did language design and uh development [31:26] of his own. He wrote one of the very [31:28] first PL1 compilers. Uh he invented the [31:32] idea of macros and it's just on and on [31:34] and on and on. Um [31:37] and he was also the department head who [31:40] sort of had Ken Thompson in his [31:42] department. Now Doug wouldn't tell Ken [31:44] what to do obviously but appreciated the [31:48] work and was an incredibly effective [31:50] technical critic of that kind of work. [31:53] So, you know, gee, wouldn't it be nice [31:55] if or you thought about that way or [31:58] geez, this sucks, whatever, uh, to make [32:00] stuff good. And that's a role that he [32:02] played for the entire time that he was [32:04] at Bell Labs. I I first met him in ' 67, [32:07] I guess, and he was until he retired [32:09] probably to mid 90s, something like [32:11] that. [32:12] >> This this reminds me when you mentioned [32:15] department head somewhere in my [32:16] research, it might have been Hamming. [32:20] There was someone that you worked with [32:22] who was a head of the department with no [32:25] reports. [32:26] >> That's Hamming. [32:27] >> Yeah. Yeah. I I never understood that [32:29] when I first met Dick Hamming and you [32:31] know he was a department head. Um you [32:33] could tell because he had carpet on his [32:34] floor. I guess he showed up in the phone [32:37] book that way too. But um [32:40] and I couldn't figure out, you know, [32:42] because I had this naive view that you [32:43] know there was status associated with [32:45] being a department head and it was [32:46] totally wrong. But um but he had worked [32:50] very hard to get himself a department [32:52] head. So he had whatever status or clout [32:55] or whatever went with it for useful [32:58] purposes but no people reporting to him [33:02] and and so I think that worked out very [33:04] well because in you know in any [33:06] organization if you're not at the right [33:07] level you may get ignored and so it was [33:09] harder presumably for something for [33:11] people to ignore a department head than [33:13] an ordinary MTS. [snorts] So he got the [33:17] clout without any of the work of [33:19] >> without the dog work. Yeah. Right. [33:21] Administering. I mean he he'd put in his [33:23] dues because he he came to the labs I [33:25] think in 46 [33:27] and did a lot of early stuff with [33:30] computing. He's one of the very first [33:31] people to realize that computer act use [33:34] of computers was going to be a very [33:37] important part of all kinds of research [33:39] in the future. and he made this [33:41] prediction that that at some point more [33:44] than half of the the research work would [33:47] be done by computing not but in physical [33:49] labs and of course they laughed at him [33:51] [snorts] [33:53] in my research I somewhere on YouTube [33:56] there's this video posted by AT&T the [34:00] corporation itself and it's this I guess [34:03] a film titled Unix making computers [34:07] easier to use [34:08] >> Oh yeah yeah AT&T did a lot of those PR [34:11] films. [34:12] >> Yeah. And there's a And you're you're in [34:14] the video. You're one of the first [34:15] people speaking and you have your legs [34:17] up on some desk and you're kind of [34:20] explaining as what what is that? [34:23] >> It's like any PR film. It's a trying to [34:25] convince people who don't know anything [34:27] about you that you're a great whatever [34:29] institution that you're contributing to [34:31] the public good or whatever. And so they [34:34] hired a professional filmmaker kind of [34:38] thing. Eric Belove, I think his name [34:40] was. It was a long time ago. And he came [34:42] around and and he would do sort of what [34:44] we're doing. He'd have a conversation [34:45] about something and try and get whoever [34:49] he was talking to to elaborate on what [34:51] they were doing and then he would splice [34:53] it together into something that was [34:55] feel-good for AT&T. [34:58] And at the time this was done, which I [35:00] would guess is [35:03] 70 late 70s, early 80s, something like [35:05] that. [35:06] uh Unix was kind of just starting to [35:09] appear on the outside. It was a sort of [35:11] a visible interesting thing. And so part [35:14] of the thing is AT&T is using computers [35:16] to improve people's lives or our [35:19] business operations or whatever what is [35:21] Unix. So my job in these this film at [35:25] least and probably several others was to [35:27] sort of try to explain to a very [35:30] non-technical audience what is this [35:32] thing and what's it good for and why is [35:35] it new and interesting or something like [35:37] that. [35:38] >> I skimmed through the comments on this [35:40] video. It has a lot of views actually. [35:42] And um the comments in there, one of the [35:45] ones that has some of the most up votes [35:47] is it's saying that everyone in the film [35:50] is, you know, not to be derogatory, but [35:52] kind of like computer nerd, you know. [35:54] >> Well, we were still, [35:56] >> but but they highlight you as being a [35:59] very clear communicator and a bit [36:02] exceptional in your ability to convey [36:05] the technical stuff you're you're [36:07] talking about. And so I was curious like [36:09] if you have any advice on how to [36:13] communicate technical work but in such a [36:15] clear way. [36:17] >> Part of it is probably just putting [36:19] yourself to some extent into the other [36:21] person's shoes. They don't know anything [36:23] about what you're talking about. So how [36:24] can you explain it with the right [36:26] combination of metaphors and examples [36:28] and everything else that gets it across [36:32] to them? [36:34] And part of that, of course, it's nice [36:36] to have feedback. So it's easier to do [36:37] that if you're in a teaching kind of [36:39] setting where you can see whether [36:40] they're getting you or not. And if [36:42] they're not, you can adapt. It's harder [36:44] when you're doing a film because you you [36:46] know the audience is later if at all. Uh [36:49] but there there were a fair number of [36:52] those things. We used to have [36:55] dog and pony show as a image that's [36:58] probably obsolete at this point, but [37:00] AT&T or Bell Labs would bring through [37:02] distinguished people [37:05] and on a tour of Bell Labs and they [37:08] would get demos of various thing, you [37:11] know, the usual stuff, demos of what [37:13] interesting work was going on at Bell [37:14] Labs and they would go through variety [37:16] of places, labs, so on. and Mike Lek and [37:20] I were on a regular kind of rotation for [37:24] Unix demos where we would get some [37:27] exalted being accompanied by some [37:29] appropriate level of Bell Labs [37:31] management and others [37:34] to [37:36] sit on a couch in a little room and I [37:38] would show them the wonders of Unix in [37:40] some way or other. Well, what are the [37:41] wonders of Unix? I don't know. But one [37:43] of them was that it's easy to create [37:45] programs by combining programs. the idea [37:48] of pipes where I can take the output of [37:50] this program into the input of this [37:52] program and maybe string several [37:54] together and I had a standard demo of [37:56] that uh which was looking for spelling [38:01] mistakes in a document because what you [38:03] could do is here's a document so what we [38:07] collect the words in the document get [38:08] them into the lowercase whatever sort [38:10] them neatly and then compare them to [38:12] another document which is a dictionary [38:14] and if you find a word that's in your [38:16] document but it's not in the dictionary [38:17] It might be a spelling mistake and it's [38:19] a one one long pipeline, six or seven [38:23] programs. So I could do this and people [38:25] would sort of understand it. This is [38:27] before WordPress software really came [38:29] along. Um so I do this demo and people [38:32] would be impressed. [snorts] The high [38:34] point of this [38:37] one day one of those visitors was a guy [38:39] named William Colby. William Colby was [38:42] the director of the CIA [snorts] [38:44] and he was accompanied by a guy who [38:47] literally had a briefcase chained to his [38:49] arm [snorts] [38:51] and Colby was a accompanied by Bill [38:53] Baker who at that time I think was [38:55] probably president of Bell Labs but he [38:57] was also the head of the foreign [38:59] intelligence advisory board. So he was a [39:01] serious spook in his own right and so [39:04] they sat there. Now I had done this [39:06] spell demo so many times before that I [39:08] knew it took too long because you know [39:11] old machines slow etc. So what I did is [39:13] the night before I ran it into a file [39:17] and then what I showed Colby was a shell [39:19] script that said sleep two and then cat [39:23] previously [39:24] and so Colby was very impressed with [39:26] this. I guess [39:29] it does tell you you don't believe what [39:30] you see in demos. [laughter] [39:33] Wh what's in that briefcase? Is it a [39:35] bomb or a gun? [39:37] >> It's got to be nuclear codes, right? I [39:39] mean, I have no idea. [39:40] >> Interesting. [39:42] [laughter] [39:43] >> The gentleman was briefcase was not [39:45] identified. [39:47] >> One of the most famous books that you [39:49] co-authored was the one on the C [39:50] programming language. And since C kind [39:53] of came from Bell Labs and you were so [39:56] close to it, um, you know, why was C so [40:00] influential in your opinion like now [40:02] looking back? [40:03] I think it hit a sweet spot in some [40:06] sense of efficiency and expressiveness. [40:11] Um, [40:12] if you go back to when C was first [40:14] created, let's just call it 1972ish or [40:17] something like that, machines were [40:18] pretty wimpy. They didn't have much [40:20] memory. The first machine that C ran on [40:22] had, I think, 24K bytes and that was the [40:26] whole thing. So, you had to write a [40:28] compiler. Oh, you had an operating [40:29] system. You needed a bit space for [40:31] files. what do you do? Um the disk I [40:34] think was half a megabyte. Uh so so it's [40:39] a little tricky. Um so C was [40:43] simple enough that you could see the [40:46] translation into the underlying actual [40:50] instructions of the particular machine [40:52] the PDB11. At that point um language is [40:55] simple. You could see the compilation. [40:57] You could imagine building a compiler [40:59] for it straightforwardly enough. [41:01] It wasn't a big language. At the same [41:03] time, it really was expressive. I mean, [41:05] you could say what you needed to say [41:07] very, very effectively in the language. [41:10] And I think it's hard to find a spot [41:13] like that. And so, one of the reasons I [41:15] think C has succeeded as long as it has [41:18] is that it still has that efficiency and [41:23] uh [41:24] oh, I know expressiveness is I think the [41:27] right word. Certainly there are lots of [41:29] other languages which for various [41:30] purposes are better choice at this [41:32] point. Well, because there are plenty of [41:35] things that C doesn't do couldn't do at [41:38] the time because you didn't have enough [41:39] resources, right? You couldn't do a lot [41:42] with uh you know fancy runtime stuff. [41:45] You didn't want to do checking because [41:47] checking would take code and you come [41:48] on, you've only got 74k bytes for your [41:51] program. Now what? So an awful lot of [41:54] things were uh cut that way. The other [41:55] thing is the original target for C was [41:58] to build the operating system and so [42:00] there's class set of things that you [42:02] just don't need if you're building an [42:04] operating system. Um so I think it was [42:08] for its time a really really really good [42:11] match and then of course Unix and C was [42:14] this sort of symbiosis. They went [42:16] together. So Unix was just so much [42:18] better than any of the commercial [42:21] operating systems at the time. all just [42:24] so much easier to use, so much more [42:26] productive and so much more fun that [42:28] people used it in preference and then [42:30] they would write programs in C and they [42:32] was all interesting tools and so all of [42:34] this was kind of a virtuous cycle. [42:37] >> OpenAI, [snorts] Enthropic and Cursor [42:39] all used this product to make their [42:40] lives easier. Work OS is an API platform [42:43] that adds enterprise features like SSO, [42:46] skim, and arbback to your app. These are [42:48] things that you need to sell to other [42:49] companies but aren't the main focus of [42:51] what you're building. And now they've [42:53] added more support for AI agents. For [42:55] instance, O for agents gives each agent [42:58] its own identity with short-lived scope [43:00] tokens instead of a longived API key. So [43:02] if something goes wrong, you can revoke [43:04] one session and everything under it [43:06] dies. Another feature they have is [43:07] airlock, which is an authorization [43:09] layer. Every action agent takes is [43:11] checked against rules that you can write [43:13] in plain English. So it's not rubber [43:15] stamping everything and it's not running [43:17] unsupervised either. And lastly, they [43:19] have pipes which connects your agents to [43:21] over 300 third party tools without you [43:23] ever needing to manage a token. So if [43:25] you're building SAS or AI agents and [43:27] selling to enterprises, work OS fills [43:30] all these gaps in a few lines of code. [43:32] Check them out at works.com to learn [43:34] more and get started and I appreciate [43:36] them for sponsoring my show. Yeah, [43:39] [snorts] when I learned I guess how to [43:42] use Unix in college, there's all these [43:45] these uh utilities like you mentioned [43:47] diff that Mroy and then there's a [43:50] variety of other ones. GP and of course [43:52] there's O I remember I I I learned that [43:57] a long time ago. [43:58] >> Oh, good. [43:58] >> Yeah. [laughter] [43:59] Um yeah, what's the story behind that [44:01] and and why did you build it? So this is [44:05] back in the mid70ish [44:08] time, 767 [44:10] I guess. Um [44:13] Al Aho and I had been friends for a long [44:15] time. He went here at Princeton. He and [44:17] I were undergrads at Toronto, although [44:19] he was always a year ahead of me. Um and [44:22] and he was very interested in regular [44:24] expressions, you know, patterns of text [44:26] that you could use to recognize [44:28] interesting stuff. I had been fooling [44:31] around pretty unsuccessfully with the [44:33] idea of some kind of text editor that [44:36] would let me manipulate numbers as well [44:38] as text, you know, something like that. [44:41] And Peter Weinberger, who I don't [44:43] remember exactly how we met him, um was [44:47] interested in database kinds of things. [44:51] And so we sort of talked back and forth [44:54] and came up with this language which [44:56] combined some of those interests. the [44:58] regular expression stuff from Al um the [45:01] databasey flavor and report generation [45:04] like RPG uh from Peter and the sort of [45:07] wouldn't it be nice if strings and [45:09] numbers were kind of treated were both [45:12] first class citizens of whatever we were [45:14] doing. So we thought about it for a [45:17] while and then if a memory serves we [45:21] came up with a sort of language design [45:23] and then Peter went off one weekend and [45:24] built it. [45:27] not Peter's extremely good programmer [45:30] too. Um, but he had the advantage of [45:33] tools at that point because he could [45:35] define the grammar with Yak, the [45:37] compiler compiler that Steve Johnson had [45:39] built and the lexical part with Lex, the [45:43] lexical analyzer generator that Mike Le [45:45] had built. And so from that, building a [45:49] language was not as much work as it [45:51] would have been otherwise. There was [45:52] some just more of it was kind of built [45:55] for you. Um, and Al had the regular [45:58] expression code already because that [46:00] came from Egret. And so, [46:03] you know, you take this on one very [46:04] bright good programmer over a long [46:07] weekend or so and you've got a language. [46:09] Oh, and then from there we just sort of [46:12] evolved it slowly and then uh Alan Peter [46:15] dropped away from actual active [46:17] maintenance probably 5 10 years later [46:19] and I sort of inherited it. A lot of [46:22] these Unix tools that everyone takes for [46:23] granted, is that usually the story [46:26] behind them, which is you guys are all [46:28] using Unix and then you just realize, [46:31] oh, I I want to be able to pipe this to [46:33] something that diffs or I want to be [46:35] able to find something and uh and then [46:38] it you guys just build it. Is that [46:40] usually how these come about? [46:42] >> I think a lot of them come from that. [46:43] It's not from hypothesizing what [46:45] somebody else might find useful. It's [46:47] more like my what do I need? it's kind [46:49] of bothering me at this point that I [46:51] could improve if I just had a better [46:53] tool or a language or something. And and [46:56] I think that's true on almost all of the [46:59] little tools uh that you mentioned. It's [47:03] you know gee [47:05] I don't want to do this by hand or [47:07] existing programs aren't quite right. [47:09] Let me build something new and be very [47:11] easy. Oh for example make over which [47:15] Steve Feldman did. You know, somebody [47:17] told me, "Oh, I was I [47:20] my program didn't work. I can't figure [47:22] out why it didn't work." Oh, it's [47:23] because I forgot to recompile one of the [47:25] files. And so, could you in some way [47:28] automate that? And that's another one of [47:30] these weekend projects that uh Stu did. [47:34] And then, boy, [snorts] [47:37] within a few days, he's got a bunch of [47:39] users. And then all of the mistakes he [47:40] made in his first design he can't fix. [47:44] So, and I think all of these have that [47:47] same problem. Certainly, there are [47:48] things in awkward or in hindsight. Oh my [47:50] god, we should have done that better. [47:53] >> A lot of the the namesakes for these, [47:56] they either explain themselves like [47:58] diff. I think it's pretty clear. And [48:00] then a is the initials of the authors, [48:03] but GRAP is a really famous tool. But [48:06] what's the the story behind the namesake [48:08] of that? [48:09] >> Oh, yeah. graph. Ken Thompson tells it [48:11] very well, but the gist of it is that in [48:14] the text editor, the ED text editor, [48:18] which was the first real editor on the [48:20] Unix system, um there is a command [48:22] called G, which says do the [48:26] this command on all of the lines that [48:28] match a particular regular expression. [48:30] And the syntax is in fact G SE whatever [48:33] the regular expression is SLP print all [48:36] the lines that match this pattern. and [48:39] hence grap. [48:41] >> We talked a lot about Bell Labs and you [48:44] know you mentioned kind of the golden [48:47] years. When you look back on Bell Labs [48:50] and your time there, what is it that you [48:52] miss most? [48:54] >> Well, I miss the people. Although I'm [48:56] still in contact with a fair number of [48:58] them one way or another. Um but the [49:00] people were it was just an amazing [49:02] collection of really good nice people, [49:04] bright, fun to work with, lots of good [49:06] ideas and so on. So you miss that. One [49:09] thing that probably is not reproducible [49:11] in a direct sense is the um environment [49:13] where there just all kinds of new things [49:15] to do. So it was an era of lowhanging [49:18] fruit if you like. I suspect there's [49:21] similar lowhanging fruit, but I'm not in [49:22] the right orchard at this point. I just [49:24] don't know enough about it. But you [49:26] know, maybe some of the AI stuff that's [49:28] going on is is the moral equivalent at [49:30] this point. Um so I think those are the [49:33] things that are perhaps the most things [49:36] that I miss. Uh, Bell Labs gradually [49:40] became more of a conventional company as [49:43] AT&T went through troubled times and [49:46] split itself and split itself and so all [49:48] the fish products were smaller and it [49:50] meant that the Bell Labs, the big [49:52] cohesive bunch of people working on all [49:54] kinds of things became smaller and more [49:56] fragmented and didn't have as much money [49:58] and therefore were forced to do the [49:59] kinds of things you don't want to do [50:01] like raise money. So that part I don't [50:03] miss at all. In one of the classes that [50:06] you taught, there's a video somewhere [50:09] buried deep in YouTube where you are um [50:13] trimming your beard in the lecture. [50:16] >> Do you remember that? And what what's [50:17] the reason for um you know grooming [50:20] during the lecture? [50:20] >> I remember that one. That is actually [50:22] you should ask David. He was in that [50:24] class. Um that was at Harvard. That was [50:27] the first lecture of CS50 the year I [50:30] taught it at the beginning. So they'd be [50:32] September of 1996 I think. Um I had [50:37] never taught a big class before. Um the [50:40] class had been taught by Margot Seltzer [50:42] who at the time was in computer science [50:44] there and she was taking a sbatical and [50:46] she asked if I would be interested in [50:47] teaching her a course and I thought only [50:49] chance I'll ever get to teach at [50:50] Harvard. Why not? [50:53] But Margot had a stickick that she did [50:55] at the beginning of her first lecture [50:57] which was trying to explain to people [50:59] why it's hard to tell a computer what to [51:02] do. And so she would get the audience in [51:05] her class to direct her in making a [51:08] peanut butter and jelly sandwich. And [51:10] you know, it'd be very easy to get the [51:11] jelly on the desk and the peanut butter [51:13] on your head and you know, stuff if the [51:14] instructions weren't clear. But I'm not [51:17] a peanut butter and jelly kind of [51:18] person. And I thought, what's the [51:20] equivalent? And then it occurred to me I [51:22] have this ratty beard. [51:24] So, I brought in a a collection of [51:28] equipment suitable for trimming a beard, [51:31] ranging from a pair of scissors to [51:33] electric hedge trimmers. And I got the [51:36] audience to try and guide me through [51:38] trimming my beard with this stuff. [51:41] And I don't remember, I think actually [51:44] the hedge trimmers were plugged in, but [51:45] it was want to be very careful here. [51:49] >> So, you mentioned that was your first [51:51] big class. [51:52] >> Yes. 457. [51:55] >> Was public speaking nerve-wracking at [51:57] that time for such a large group? [52:01] No, I I think originally when I was like [52:03] a grad student or something, public [52:05] speaking was very challenging, but I [52:09] when one of the things I did first at [52:11] Bell Labs and one of those I think [52:14] [snorts] maybe even intern years, I [52:17] don't remember now, was um Vic Fatsky [52:20] who was across the hall from me said, [52:21] "We've got a bunch of physics guys who [52:23] need to learn forran. Would you mind [52:24] teaching them?" And so [snorts] I did [52:26] that. And so the the teaching experience [52:29] got me over this fear of public speaking [52:33] totally completely. Uh and so I had I [52:37] had done a bit of adjunct teaching uh [52:40] while I was at Bell Labs earlier before [52:42] the Harvard gig that was by far the [52:44] biggest like at least more than an order [52:46] of magnitude the biggest but at that [52:48] point it's not the size of the audience [52:51] so much it becomes it's a theater [52:53] performance and [52:56] David M can tell you that more [52:57] accurately because he's really good at [52:59] it. [53:01] So if um someone who is a student who's [53:04] scared of public speaking asks you for [53:06] advice, what would you tell them? [53:09] >> I think doing it getting into an [53:12] environment where you have to talk to [53:14] people and in at least small group to [53:17] begin with and then maybe bigger one is [53:19] okay. Um and I think there's no no [53:23] better way than actually just doing it, [53:26] getting out there and forcing. So, so [53:28] I'm past the stage where I worry about [53:30] that sort of thing. I guess conference [53:32] presentations are sort of like that too, [53:34] but but teaching's better because it's a [53:37] more intimate setting for the most [53:39] passing over 400 people, but you know, [53:41] it's it's a more intimate setting. And [53:43] so, um, you got a better way to judge [53:45] whether you're content [53:48] or whether they're understanding you or [53:50] not. And if you think of them as friends [53:52] or something like that and then you blur [53:53] a little your vision so you're not [53:55] really looking directly. It looks like [53:57] you're looking at them but you're not. [53:59] Or you focus on one person in the [54:01] audience who seems to be resonating. All [54:03] of these things I think help. [54:05] Um but after a while it just becomes [54:08] practice and then you don't don't worry [54:10] about it. I know you still lecture here [54:13] at Princeton from time to time and it [54:16] seems like more recently the education [54:20] may be changing because of some of this [54:22] LLM AI stuff. I mean unprecedented [54:25] access to answers I guess. Do you [54:28] believe that LMS are a net negative or a [54:32] net positive when you look at education? [54:36] >> I think it's too early to tell. I can [54:38] see plenty of elements of both sides of [54:40] that. [54:41] Um, one concrete example that I have [54:45] heard reported by all kinds of people [54:47] around here is that students don't come [54:50] to office hours. [54:53] They're presumably getting whatever it [54:55] is from their favorite large language [54:57] model and so they don't come in and talk [55:00] to faculty or the preceptor or grad [55:03] students or whatever it is. They just [55:04] don't do that very much. Um, and I think [55:08] that one's a net loss. [55:10] the bandwidth when you're talking to [55:12] somebody in person is enormously higher [55:14] than when you're talking to Claude or [55:15] whatever. Um, and the other thing is [55:19] that the human contact that you get from [55:21] this is useful in all kinds of ways. Not [55:25] the least of which is that if you come [55:26] in and talk to some faculty member, that [55:28] person now know who's you are and they [55:30] can say, "Oh, did you know here's an [55:32] opportunity? Here's a somebody looking [55:34] for a job, whatever." So, they're [55:36] missing the kind. [55:38] it's not it's not transactional that [55:40] they're missing that human connection [55:43] that helps [55:45] in the future. So So I'm sorry to see [55:47] the the office hours go away. Um [55:51] I think a weird [55:56] I don't know why you call it an upside [55:57] or a downside of this in computer [56:00] science in particular. Enrollments are [56:02] down. So, we have many fewer computer [56:04] science majors than we did a few years [56:06] ago and that means the classes are [56:08] smaller. Not entirely because many of [56:10] those people just went to other [56:11] departments, but they're still taking [56:13] our courses, but they're somewhat [56:14] smaller and that means that there's more [56:16] contact with the students and I think [56:18] that's good. So, [56:21] >> what percent down would you say? our [56:24] numbers are I don't know where they're [56:26] rep I think they're more extreme for [56:28] some unknown reason but I think our [56:33] the class that just graduated in uh this [56:36] the summer spring of 2026 was a little [56:41] over 200 computer science majors. The [56:45] class that will graduate two years later [56:47] is probably going to be less than half [56:49] that. [56:50] >> Less than half. [56:51] >> Yeah. [56:52] >> Wow. [56:52] >> Maybe even down to a third. We don't [56:54] know because there's a staggered way in [56:55] which majors appear but but I would [56:59] guess that if it was 200 it'll be sort [57:01] of 80ish 90 pure guess work but quite a [57:05] bit more. Those people have gone in here [57:09] to electrical and computer engineering [57:11] and mechanical and aerospace engineering [57:13] because those I think are more concrete [57:16] hardware oriented less likely to be [57:19] replaced by large language models. [57:21] [snorts] So the kids are voting with [57:22] their feet. I think uh to some extent [57:26] that's totally understandable. I think [57:28] they may be overreacting, but I don't [57:31] know. And if you just read the popular [57:34] press, my god, we don't need programmers [57:36] anymore. [57:38] I think that's fortunately [57:39] oversimplified, but but I don't it's not [57:43] my problem in the sense that I have a [57:45] job and they quite rightly are worried [57:47] about what they're going to do when they [57:48] get out of here. [57:50] >> All right. So the the total number of [57:52] students that go to Princeton is roughly [57:54] constant. [57:55] >> Oh, it's con it's totally constant. [57:56] >> Okay. But the proportion that goes to [57:58] the computer [57:58] >> different different fields is different. [58:01] And I don't know I I know those [58:04] particular numbers for where did where [58:07] does it look like computer science [58:09] people went over the last you know [58:12] voting with their feet right now and it [58:14] looks like mostly to electrical computer [58:16] engineer mechanical aerospace for the [58:18] robotics aspect. [58:20] I saw you're an extra on the film A [58:23] Beautiful Mind. It's a great movie. And [58:26] before the interview, I was trying I was [58:28] like kind of combing through the the [58:30] movie trying to find you, you know. How [58:32] did that come about? And what's it like [58:34] to be a part of such a big movie? [58:38] [snorts] [58:39] >> You know, that that was enormous fun. [58:42] They did a casting call uh for that [58:44] because parts big parts of it were going [58:46] to be shot at Princeton. Um, and this [58:50] was just shortly after my wife and I had [58:52] moved to Princeton, so I don't remember [58:54] the date, but call it 2001ish. [58:57] Um, so they had a casting call and so a [59:00] bunch of people showed up in one of the [59:02] big classrooms and we heard the story of [59:04] what was going to happen and then um, [59:07] some guy wandered around and said, "You [59:10] and he didn't pick my wife, which I [59:12] thought was nuts, but he pointed at me." [59:15] Okay, so this was the casting call. And [59:18] then later on I got uh invitation please [59:22] come and we're going to do some shooting [59:24] of the Nobel Prize ceremony. [59:27] So you'll find me there. I can tell you [59:28] where to find me. U the reason you don't [59:30] recognize me is first nobody is wearing [59:32] glasses because they would reflect. And [59:35] secondly I am wearing black tie [59:38] something which I have never worn in my [59:40] life because [clears throat] this is a [59:41] Nobel Prize ceremony. Um, and it was [59:46] shot in the Newark Performing Arts [59:48] Center, so a place I had never been [59:51] before. And it took about 22 hours to do [59:55] that. But because you got there at sort [59:58] of 5 in the morning, you had to be [59:59] dressed properly. Um, and then there [60:02] were shooting and shooting and shooting [60:03] and shooting and lots of dead spots [60:05] while I, you know, rearranged the [60:06] cameras and so on. Um, but it was really [60:10] interesting and I got paid. Not equity [60:12] rates, but I got paid. I got something. [60:14] I would have been better off to flip [60:15] burgers at McDonald's for the [laughter] [60:18] [snorts] uh and then the movie came out [60:20] and and it was fine. But the wonderful [60:22] thing is that I got an airish bacon [60:24] number of five. [60:25] >> Eridish bacon. So, okay. What those [60:28] numbers represent how far you are from [60:31] those people works [60:33] >> in in in their related works. So, my my [60:35] airish number is three because I can go [60:37] through several people a couple of paths [60:39] to in steps, three steps. But my Bacon [60:42] number is two because I was in this film [60:45] with somebody who was in a film with was [60:48] Kevin Bacon. [60:50] So that [snorts] anyway that's the the [60:51] the gist of it. I actually met John Nash [60:54] a number of times after that. Um [60:57] >> the not the actor, the real [60:59] >> the real John Nash. Yeah. Yeah. He was [61:01] >> you know he was kind of let's call it [61:04] introverted or something like that but [61:05] perfectly fine. The scary story for me, [61:09] I was giving a talk, I don't remember [61:12] how it came about, but I was giving a [61:13] talk to an internal group in the math [61:15] department about something that I had [61:18] worked on. I'd forgotten what it was. [61:20] And I was, you know, into this. I had [61:22] been going 10 minutes or something. And [61:24] I was on a roll and I'm not scared of [61:25] audiences or anything. I was having a [61:27] good time. And John Nash walked in [61:32] and I could just feel everything [61:33] tightening up inside. [laughter] [61:34] And then he asked some question. I don't [61:36] remember what the question was or [61:37] anything like that, but [sighs] I got [61:40] over it. [61:42] But but that that's one of the few times [61:44] where public speaking is kind of a bit [61:46] of a threat. Another interesting factoid [61:50] that I wasn't sure if this is true, but [61:53] so in the the git repo for the Go [61:57] programming language, actually the first [62:00] commit in it is from you in 1972 [62:04] supposedly, [62:06] but the you know the programming [62:08] language started in 2008. So [62:10] >> yeah, right. I have no idea what this [62:12] is. [62:13] >> Okay. You've never seen [62:14] >> this is news to me. [62:16] >> Okay. Yeah, it's supposedly I mean I [62:19] found it says first commit to the go get [62:23] ret. [62:23] >> Oh, is it a hello world example? That's [62:25] the B program. [62:26] >> Yeah. Yeah. Yeah. It's it's in B for [62:28] some reason, [62:29] >> right? [62:30] >> So, uh [62:32] >> and this is followed up by commit 1974 [62:34] to convert it to C and then another one [62:37] to convert to NC. So, maybe Robert [62:41] copied your hello world repo and used it [62:44] to start going. I never used git at that [62:46] point. I don't think I started using git [62:48] until maybe 10 years ago. [62:50] >> Interesting. [62:52] So then what did you use was there the [62:54] notion of version control back then [62:56] because somehow [62:57] >> version control came along in the [62:59] mid70s. [63:01] There was something done by the [63:04] programmers work people I think called [63:06] SECS the source code control system. [63:10] uh and that was widely used inside the [63:13] labs and I think outside and then there [63:14] was RCS by Robert Titchy which was a [63:19] variant of the same thing revision [63:21] control system and then there was [63:24] CVS was it concurrent version system or [63:28] something like that and then some other [63:29] version of it and then get and and [63:33] others scattered around. So, it's an old [63:35] idea, but in 1972, it did not exist. I [63:38] mean, Unix didn't exist in 1972. That B [63:40] program ran on something else. [snorts] [63:43] >> Interesting. Okay. Yeah. I guess it'll [63:45] just remain a mystery. Maybe maybe [63:47] Robert can uh answer or something if I [63:50] ever interview him. In your memoir, [63:55] there's a thank you to Grace Emlin. Mhm. [63:59] >> And then I dug a little more and I also [64:01] found this um Bell Labs alumni tracker [64:05] and it kind of shows where all the [64:07] alumni go and Grace Emlin it said uh [64:11] relocated witness protection program and [64:14] so curious what what that is about. [64:18] There was this period of time, [64:22] forgotten the dates, when Bell Labs [64:24] decided that employees had to have they [64:28] had to wear their badges so that if you [64:30] were walking through the hall, people [64:32] would know that you were an employee as [64:33] opposed to a tourist of some sort or [64:35] other. Um, and people were not very [64:38] enthused about wearing badges. And so [64:40] there was a bunch of kind of quiet [64:44] insurrection about it. And one of the [64:47] bits of insurrection was the creation of [64:49] badges that were fakes. And so, uh, [64:53] Grace Emlan was a creation of [65:00] I think it was probably Fred Gramp who [65:03] did it. There's this Gremlin thing. And [65:05] so, Grmlin Grace. Um, and we managed to [65:09] get Grace into a variety of official [65:11] records for the company for a while. and [65:13] and she uh this little gremlike thing, [65:18] you know, this high sat in the Unix room [65:20] with her badge on, but there were other [65:23] people who had weird badge. I had a [65:24] badge with Mickey Mouse on it. And so I [65:27] probably still have it in my desk, but I [65:29] will try to get it. Um that I used to [65:32] wear all the time. I mean, it was I [65:34] don't wanted to. Nobody looked at the [65:36] badges. They just see a blur and there [65:38] was a badge, so it was okay. So I wore [65:39] this badge all over the place. I wore it [65:41] when I met Bill Gates at one point. I [65:44] don't think he noticed either. [65:47] [snorts] And so, you know, people didn't [65:49] take kindly to bureaucratic stuff, even [65:52] though probably badges were a fine idea. [65:56] >> Was the badging system digital at that [66:00] point? [66:00] >> No, that was just paper. Pieces of [66:02] paper. Oh, [66:03] >> okay. [snorts] [66:03] >> Or they were pieces of plastic, but you [66:05] could [laughter] put your own paper in [66:07] front of it. another thing. So I found [66:10] somewhere on the internet in some very [66:12] old repository [66:14] there was a a secrets file that was [66:17] committed to the the repo and everything [66:21] in there is encrypted but it was [66:23] encrypted by the old dees security [66:27] standard. So it's possible to brute [66:29] force unencrypted with uh modern day and [66:33] so you and all your colleagues's [66:36] passwords are actually [laughter] [66:38] >> so I saw yours is I mean I'm assuming [66:42] it's okay. Yeah, it's just slash period [66:45] comma slash period comma. [66:46] >> Really easy to type quickly. Yeah. No, [66:48] it's funny. I hadn't seen that. I got [66:50] mail from Eric Schmidt saying, "Gee, did [66:52] you see this?" [laughter] [66:54] Sent to several of us. Um, and so it was [66:56] intriguing to see what the passwords [66:58] were. I I I guess you're probably right. [67:00] It probably was either dees or some [67:03] World War II machine like a Haggelan [67:04] machine or whatever that the crypt [67:06] program was based on. In our defense, [67:09] this was at a time when the machines [67:11] weren't [67:12] and the passwords were kind of why would [67:14] you need them anyway sort of thing. Uh, [67:17] and so, you know, we didn't pay much [67:19] attention to them. Um, and I guess the [67:22] other thing is the password file, this [67:25] was I think a tape that had was from [67:28] Berkeley and so I don't know somebody [67:30] must have copied the password file from [67:32] the research Unix things at Bell Labs [67:35] over to some Berkeley system. Don't know [67:37] how it got there. [67:39] >> Coming to the end, um, yeah, I know you [67:42] you wrote a ton of books in your career, [67:45] very prolific. I mean, there's the Cbook [67:48] that was just one of many, though. [67:51] What's your motivation for writing all [67:53] these books? [67:55] >> I find it fun. I think it goes back to [67:57] something you asked about earlier. [67:59] Trying to explain stuff to people who [68:01] are not experts in it. In some way, I [68:04] find that attempt to communicate [68:06] rewarding, worthwhile. Um, it's a way to [68:10] have influence at a distance perhaps. [68:12] So, teaching is kind of retail. You're [68:14] in the same room with a bunch of people. [68:16] Writing a book is wholesale if you like. [68:20] you're you're not you won't meet most of [68:22] the people who wait might look at a book [68:25] but with luck they will also learn [68:26] something from it. Um [68:30] partly it's a way to learn the stuff [68:31] yourself and [68:34] I guess partly it's just that once [68:36] you've done one [68:38] you do more. I remember a conversation [68:40] with Don Kuth eons ago where uh he said [68:44] that there are people who have written [68:46] no books, there are people who have [68:47] written one book and there are people [68:48] who have written many books and so I [68:50] guess I'm in that third category sort [68:52] of. [68:52] >> Yeah. I for me it's kind of fun and so [68:54] I'm still doing it. This is your chance [68:56] to do a commercial announcement. [68:58] [laughter] [68:59] >> Yeah. What's your latest book actually? [69:01] >> Yeah. So I I I did a book um over the [69:04] last year or so, a history book of [69:06] important ideas in computing and will [69:08] appear from MIT Press in January, end of [69:12] [snorts] commercial announcement. [69:14] >> Is is that the memoir? Oh, no, this [69:15] hasn't come out yet. [69:16] >> Yeah, this hasn't come out yet. Over [69:17] there underneath your camera is a the [69:19] cover of it. Uh [69:21] >> oh, shifts. [69:22] >> Yeah, right. [69:23] >> Okay. [69:24] >> It's it's available on Amazon, but it [69:27] won't appear until January or something [69:28] like that. But the idea is that there's [69:31] a bunch of interesting things that have [69:32] happened in computing where something [69:34] there was a change in a different [69:36] context. You'd call it a paradigm shift [69:38] like the Thomas Kern idea, but it's not [69:39] really quite but you know the shift from [69:41] analog to digital big different. I mean [69:44] all the camera stuff you're doing here [69:45] would not be feasible with film, right? [69:49] Uh and so there's the stored program [69:51] computer the idea that you can build one [69:53] general p device that will kind of do [69:56] anything you want uh just by giving it [69:58] different instructions and so that's a [70:00] big deal uh there's all the integrated [70:03] circuitry stuff that the evolution [70:05] started with the transistor and to IC's [70:09] today all of that stuff again all what [70:11] you're doing here wouldn't be possible [70:13] all our computers wouldn't be um [70:16] programming languages you know if we [70:18] were still writing [70:19] assembly language, it would be pretty [70:21] bad. But the evolution of programming [70:23] language is making it ever easier to [70:26] tell computers what we want to do. And [70:28] so go on like that. Pick another. And [70:30] then the the the final uh thing that's a [70:32] big big paradigm shift that's different [70:35] I think from most of the others is that [70:38] we're in the middle of it and it's [70:39] incredibly visible and that's the AI [70:42] large language model stuff which is [70:44] changing incredibly rapidly and is [70:47] affecting everybody and everybody's [70:49] noticing it which is different from most [70:51] of the other things that we've talked [70:52] about. So um took my wife who is not a [70:57] computer person is interested in [71:00] learning more about what's going on with [71:01] artificial intelligence large language [71:03] models explain this what's that mean [71:05] what what's the future going to be um [71:07] and so that paradigm shift I think is [71:10] really really really uncertain and we [71:12] are very much in the middle of it and [71:13] everybody knows about it so [snorts] [71:15] anyway now I've given you the synopsis [71:17] of the book go buy it [71:18] >> yeah [laughter] [71:20] >> the um you know with with this this AI I [71:24] and you know all these LMS one of the [71:26] big things that we've seen is that they [71:28] can write all the code actually like a [71:30] lot of my peers genuinely [71:33] could say 100% of the code in the last [71:36] so [snorts] many months was written by [71:40] uh an LLM and so a lot of people who [71:43] really love computing or love writing [71:46] code it's actually kind of sad for them [71:49] to see you know these things become [71:52] obsolete like typing things out and [71:54] stuff. I was curious in your [71:57] career in computing there must have been [71:59] points where something became obsolete [72:01] maybe like the punch cards or I don't [72:03] know something like that. And I guess [72:05] how did you feel at that time? [72:07] >> Yeah. Well, I don't miss punch cards in [72:09] the slightest. [laughter] [72:11] No way. I I was thinking, you know, my [72:14] my thesis was done on my PhD thesis was [72:17] done on punch cards and that was three [72:19] boxes of cards and I realized that that [72:20] was 30 pounds of stuff I had to carry [72:22] around for my thesis. [72:25] So that's so there are certain things [72:27] like that I don't miss. Um there are [72:29] other things that you miss the kind of [72:32] call it the intellectual challenge or [72:34] the puzzle solving or the whatever where [72:36] you write a piece of code. So right at [72:38] the moment I'm teaching class that [72:40] involves C programming. So it's back to [72:43] early days and you know I haven't [72:46] written much C for a longish time and [72:48] it's kind of fun to get back into it and [72:50] sort of rethink and relearn you know oh [72:53] that's what pointers are all about kind [72:55] of thing. Um so I suspect most people [72:58] don't miss that sort of thing because [72:59] the memory management aspects are a [73:02] pain. On the other hand, the challenge [73:05] and the fun of figuring it out and [73:07] getting it to work and so on is there. [73:10] Now, the same fun and challenge and [73:11] getting it to work is going to be true [73:13] if you're writing in Python. Is it there [73:16] if you prompt your favorite [73:19] claude or whatever to generate it for [73:22] you? I don't think so. I don't think [73:24] that at all. And and even once the [73:27] program is generated for you, you still [73:29] have to see whether it works and whether [73:31] it's the most effective way to do what [73:33] you want and is it really doing what you [73:35] want or should you have wanted something [73:36] different and all of those sorts of [73:38] things. And that's in in an environment [73:40] where you know what you're doing. I've [73:43] been using uh mostly claude but uh [73:47] others as well to generate SQL because [73:51] I'm not an ex SQL expert and I have a a [73:54] colleague over at the Institute for [73:56] Advanced Studies who has a database. [73:58] She's not a programmer at all but she [74:00] wants to query her database effectively. [74:02] So I need something to generate SQL from [74:06] natural language queries. So I want to [74:08] say give me all of the whatevers that [74:10] have this property and get the large [74:13] language model to generate the SQL and [74:15] then run it through the database. How do [74:18] I know whether it gave me the right [74:19] query? Because I'm not an SQL expert and [74:22] we if we already knew the answer, we [74:24] wouldn't need it. And so I'm wrestling [74:27] with that. And my observation is [74:32] 90% of the time it generates a perfectly [74:34] fine query and 10% of the time it [74:36] doesn't. gives you a totally wrong [74:38] answer and I don't know which is which. [74:40] It sounds like what you're saying is you [74:42] need some domain expertise otherwise you [74:44] can't be the judge of if it's [74:46] >> you need domain expertise probably it's [74:48] the same domain expertise you always [74:50] needed um you need ways of validating [74:53] what you do because you can't be [74:55] sufficiently expert that you can test a [74:57] big program all by yourself so you need [74:58] something AI again to help you with [75:01] generating the test but then you need [75:02] something to evaluate whether the tests [75:05] were properly generated and there's this [75:07] infinite regress of you know check the [75:11] checker who checked the checker kind of [75:13] thing. Um now I'm oversimplifying and [75:16] I'm not a high-powered user and I know [75:18] plenty of people who as you described [75:20] her are you know essentially everything [75:22] they used to do they have outsourced to [75:25] something like a large language model [75:28] and I don't know what they're giving up [75:30] by doing that in one sense they're [75:32] getting a lot out of it they're probably [75:34] 10 times as productive as they used to [75:36] be in terms of code to get some [75:38] functionality [75:40] and at the same time they're probably [75:41] losing 10% % of their brain cells are [75:46] failing to exercise them properly. And I [75:48] I don't I just don't know it. It's early [75:52] and things whatever we say today in a [75:55] week or a month will not probably be [75:57] correct anymore. The place where this [76:00] shows up right around here at least is [76:02] kids worrying about what's going to [76:04] happen when I get out of here. I wanted [76:06] to be a software engineer. [76:08] Does that job still exist? And the [76:10] answer is maybe. But it's different than [76:13] it would have been two or three years [76:15] ago. And by the time they graduate in a [76:17] year, it' probably be different yet [76:18] again. And [76:21] question is how do we on the faculty [76:24] side of things, let's say, help them [76:26] manage this? What do we try to teach [76:28] them? How do we get them to have the [76:32] right habits of thought or whatever so [76:34] that they are able to go, you know, [76:38] thrive in this brave new world? And the [76:41] short answer is I have no idea either. [76:43] >> And uh I don't think anywhere does. [76:47] >> In the world you describe, which is [76:49] today's world, the models are not [76:51] perfect. There's a lot of subtleties [76:53] where they get things wrong. But in a [76:56] world where they are perfect and they're [76:59] kind of doing everything, what just like [77:02] out of curiosity, what happens to a [77:04] computer science major? Well, I mean, [77:06] first you you have a premise there which [77:08] is that these things will become perfect [77:10] and I suspect they won't become perfect [77:12] although they may become a lot better [77:13] than they are right now. So there's [77:15] probably still going to be need for [77:16] somebody to assess their perfection. Um [77:20] I think the nature of what people do [77:22] with them will change in the same way we [77:24] don't have people writing assembly [77:26] language very much anymore. We trust our [77:28] compilers and in fact we don't write C [77:30] very much. if you're at higher level [77:31] languages and so this may be another [77:35] step up the call it the abstraction [77:37] chain or something like that but you're [77:39] still going to need people to guide it [77:40] and you know what's the right sensible [77:43] direction to go the other thing is that [77:45] people will be able to build more [77:47] interesting things because of this so I [77:50] have a student from a couple of years [77:52] ago graduated a couple of years ago who [77:54] is doing a very very successful [77:57] interesting company using artificial [78:00] intelligence large language models to do [78:02] translation of natural languages mostly [78:05] for user interator based kinds of [78:06] things. Um and I think his company is [78:09] going to be very successful and it's [78:12] interesting work and as the large [78:15] language models improve it just means [78:17] that he can do a better job at what he [78:20] is already doing and offering to his [78:23] customers. So I think that kind of thing [78:24] will probably continue to happen. [78:27] >> When when C came out were people [78:29] mourning assembly language. [78:32] >> I think it would be before that. I think [78:34] it was when forrren came along. If you [78:37] look at what John Bakus said when he was [78:40] and his group at IBM were working on [78:42] forran this would be roughly 56 [78:46] 57 that kind of era I guess. Um [78:52] he was very afraid that people would not [78:56] accept this new tool because it would [78:58] not generate code that was as good as [79:01] people could generate that the assembly [79:03] language came out or the program [79:05] wouldn't be as efficient. And so in the [79:08] design of forran he quite consciously [79:10] stayed away from elegance and and so on [79:14] and aimed towards efficiency of the [79:16] translation and and so it was a genuine [79:18] concern and at some point along the way [79:21] their program their compiler started [79:23] finding optimizations that people didn't [79:26] and that's the point where he figured [79:27] maybe this is going to fly now after [79:30] that I suspect people did mourn you know [79:33] the skill and the whatever that goes [79:35] into assembly language age program, but [79:37] for most people, nah, they just wanted [79:40] to get the results. And forrren was a [79:41] boatload better than assembly language [79:45] in multiple directions. It was easier to [79:47] learn. If you were a physicist or a [79:49] chemist, you could learn forran where [79:51] you never learn, you know, I know 7090 [79:53] assembly language. And furthermore, if [79:55] you got it working once, you could put [79:56] it on multiple different machines. [79:59] when you look across your whole career [80:01] because you you've worked in a bunch of [80:04] different programming languages and [80:06] you've taught a bunch of different ones. [80:08] Are there any that stick out to you as [80:10] your favorite or your least favorite? [80:14] [clears throat] I if I think about what [80:16] I do on a day-to-day basis, if I write a [80:17] program, it tends to be a small one and [80:19] it's an o [80:22] kind of silly but okay. At least I know [80:24] it. Um, if [snorts] I were to write [80:26] bigger ones, I would almost surely write [80:28] them in Python because for many things [80:30] that's the right compromise. Uh, it's [80:33] certainly efficient enough for anything [80:35] I do. Um, and it's totally expressive [80:37] for the kinds of things I might want to [80:39] do. I will occasionally go back to C [80:42] because I've got some older program and [80:43] maybe I want to do something with it or [80:45] as I've said, I'm teaching it at this [80:46] point. So, so those are kind of the [80:49] languages that I wind up using. Not very [80:51] many. Um, are there languages I wouldn't [80:53] use? So, who shall I insult today? [80:56] [laughter] [80:58] I used to in my class, I haven't done [81:00] this for a long time, but I used to [81:02] basically write the same program in as [81:05] many languages as I could find. Um, it [81:08] was a simple program. You would take [81:10] text input lines of arbitrary lengths [81:13] and put them out so that the lines were [81:15] all at most 60 characters or something. [81:18] So, it's a simple-minded version of the [81:20] reflow that you would see in a in a [81:22] browser. And I would write it in every [81:24] language I could lay my hands on. I had [81:26] two dozen or more of them and I'd show [81:29] them in class. And the question is which [81:30] ones ran faster and which ones were in [81:32] some sense small or easy to understand [81:34] or whatever. And lock was pretty fast [81:37] and it was the smallest always. It was [81:39] about 15 lines of code to do this job. [81:42] And C was always the fastest but it was [81:45] harder because you had to worry about [81:46] memory management a bit and it was [81:48] clunky to do strings. Uh, and the other [81:51] scripting languages like Pearl and [81:53] Python and PHP and Yian, they were [81:56] variously sort of okay. Pearl was always [81:59] the faster one. Um, and some of the high [82:02] powered languages that were supposed to [82:03] be better were actually slower. Java was [82:05] slower than the scripting languages. [82:07] [snorts] Kind of weird. C++ wasn't as [82:10] fast as I thought it would be. Um, and [82:12] at the far end of it was Haskell. I and [82:16] it took me weeks to write the Haskell [82:18] program whereas Lua a language I had [82:20] never touched took me about half an hour [82:23] you know look at the manual oh I see [82:25] done [82:27] whereas Haskell took forever and then I [82:29] showed it to some Haskell expert he said [82:30] oh but you have to import this character [82:32] library that does this and you're and [82:34] then it ran faster [82:36] >> what made Haskell take so long [82:38] >> it's a different way of thinking and I [82:41] just didn't gro it I really didn't and [82:44] and I think part of it was the the [82:47] problem is you're taking input in one [82:50] form and async not asynchronously but [82:53] not at the same cycle. You're trying to [82:55] produce output. So your lines are [82:56] varying lengths and you want to produce [82:58] lines at the same length and that [82:59] mismatch of input and output kind of [83:02] frequency didn't fit naturally with [83:06] Haskell as I understood it. Now this is [83:09] not a knock on Haskell. I mean there are [83:10] people who love it but I'm not one of [83:12] them. [83:14] I saw somewhere on the internet that uh [83:17] somewhat recently in the vast the last [83:20] few years you added uh Unicode to O the [83:24] the repo of O that you're kind of [83:27] contributing to and it made me think [83:29] what motivates you to continue to write [83:32] code. [83:33] >> Well, it's fun. It's this intellectual [83:35] challenge that we talked about and I [83:37] think part of it was that for O in [83:40] particular I wanted to actually be able [83:42] to process Unicode because an awful lot [83:44] of text in the world is in Unicode at [83:46] this point where by unic code I mean [83:48] UTF8. So I would um I wanted O to work [83:53] properly on that kind of stuff and it [83:55] didn't. So that was part of the change. [83:58] The other thing I did at the same time [83:59] was to make it work with CSV files [84:01] because I had data that was in CSV files [84:04] and it lock was very clunky for [84:07] processing that and so added a mechanism [84:11] so you could say hey do CSV properly and [84:14] the input mechanism I think both of [84:16] those were triggered probably by working [84:18] on a second edition of the O book and [84:21] that was done in you know just a couple [84:23] of years ago and it this day and age it [84:25] would be really embarrassing if you [84:26] couldn't handle unic [84:28] For young people, a lot of times they're [84:31] they're trying to discover, [84:34] I guess, their passion or the the thing [84:36] they want to work on. It sounds like you [84:38] found it pretty quickly in programming [84:40] and you're naturally enjoying just the [84:43] act of it. If you had a student that [84:45] comes to you or a young person and they [84:47] ask you, "How do I find that thing I [84:50] should be working on that I should be [84:51] passionate about?" What would you say? [84:54] >> It's a good question. I think probably [84:56] what you tell them is to go do as many [84:58] different things as you can in the hope [85:00] that something will resonate. I mean, [85:01] how do you find the perfect partner for [85:03] your life or whatever? You don't sit in [85:05] your room and think about it. You go to [85:07] parties or whatever or maybe you use [85:09] some app like Tinder or what I don't [85:12] know. Uh [85:14] but but you do things that get you out [85:17] in a place where you might discover the [85:19] sort of thing that you're hoping to [85:20] find. It's an imperfect process [85:22] obviously. Uh but I think if you don't [85:25] do things, if you don't try different [85:27] things, then you may not stumble into [85:30] it. Um long time ago, I think it was [85:34] Dick Hamming said to me that, you know, [85:37] it was Nancy Reagan who said just say [85:39] no. This was the don't do drugs kind of [85:42] thing. And Hamming said just say yes. [85:46] not to drugs presumably [85:48] but but the idea that if somebody comes [85:51] along with something and you think I [85:52] don't know let me try it and applied [85:55] with some common sense that seems like a [85:58] reasonable way to handle a lot of [86:00] different things it's hard to know and [86:02] what works for one person isn't [86:04] necessarily going to work at all for [86:05] anybody else and so you know giving [86:08] people career advice or life advice you [86:12] want to be careful if you get it from me [86:15] >> and And yeah, last question for you is [86:18] um you know with all the experience that [86:20] you have now and going through your [86:22] career and everything we discussed, if [86:24] you could go back in time and give [86:26] yourself advice knowing everything you [86:29] know now, [86:30] >> what would you say? [86:31] >> I think if I look at it, a lot of what I [86:35] have done, you've seen a Golton board. [86:37] It's this thing where you drop a marble, [86:38] there's there's an array of nails and [86:41] you drop a marble in at the top and it [86:43] comes out somewhere at the bottom. And [86:44] if you repeat it enough, you get a [86:46] normal distribution of distribution of [86:48] marbles across depending on which way it [86:50] bounds. And so I'm like a a ball dropped [86:53] into a golen board and you know and I [86:57] come out somewhere at the bottom. But [86:58] the decisions which nail when I hit a [87:00] nail, which way do I go? I have no idea. [87:03] And a lot of it is serendipity. However, [87:06] one one I think genuinely useful [87:08] observation uh a lot of this stuff comes [87:10] from talking with Dick Hamming. Um but [87:12] but it was pester who said chance favors [87:15] the prepared mind. So the more things [87:17] you have thought about, you've wondered [87:19] about, done a bit with, met people on [87:21] whatever, the more likely is some one of [87:24] those will actually lead you in the [87:26] right direction [87:27] >> on that board. You're saying prepared [87:30] mind kind of bias those outcomes a [87:32] little [87:33] >> towards a potentially better outcome. [87:35] Yeah, I think so. [87:37] >> Well, yeah, that's all my questions. [87:39] Thank you so much for your time. Really [87:40] appreciate it. [87:40] >> My pleasure. Hey, thank you for watching [87:43] this podcast. If you liked it and you [87:44] want to see the show grow, please [87:46] support with a comment or a like. Also, [87:49] if you have any recommendations for [87:51] people you want me to bring on, please [87:53] drop a comment. Guests like Barbara [87:55] Liskoff, Mike Stonereaker, Mark Brooker, [87:58] these were all people that I brought on [88:00] because someone left a comment. On [88:02] another note, aside from the podcast, [88:04] I'm working on building the ergonomic [88:06] keyboard that I wish existed. Here's a [88:08] glance at the prototype. It's a split [88:10] keyboard, so there's two sides. This [88:13] isn't the case. But yeah, we launched on [88:15] Kickstarter and we hit our goal within [88:17] eight hours of launching. I really [88:18] appreciate it if you were one of the [88:20] people who grabbed one of the early [88:21] units. Um, we're now working on the long [88:24] journey of building the tooling now. And [88:26] so, if you still want to pick one up, [88:27] I've left the late pledges open on [88:29] Kickstarter, so you can grab one there. [88:32] I'll put a link in the description. [88:34] Thank you again for watching the podcast [88:36] and I'll see you in the next