DoNotPay CEO Leverages Consumer Anger to Scale AI Negotiations
Ai4
Connecting automated tools to deep human emotions and strict guardrails drives sustainable AI startup growth and levels the playing field against corporate bots.
## Background
This presentation by Josh Browder, CEO of DoNotPay, was published by Ai4 on YouTube on August 26, 2025. According to the page description, Browder—who is also a General Partner at Browder Capital—shares lessons learned from building DoNotPay and investing in leading AI companies.
## TL;DR
Josh Browder reflects on a decade of building and investing in consumer AI, explaining how DoNotPay scaled by channeling core human emotions like righteous anger through tools like automated support negotiations and robocaller traps. He shares key startup and fundraising lessons from his Thiel Fellowship experience, including how refining his pitch helped overcome 19 consecutive VC rejections. Browder also details the economics of AI model selection, demonstrates how guardrails manage live voice AI interactions, and highlights how consumer-focused AI leverages relentless persistence to level the playing field against corporate bots.
## Atmosphere & style
The talk carries an energetic, pragmatic, and candid entrepreneurial tone, blending personal founder anecdotes with actionable startup insights. Browder presents with clear passion for consumer advocacy, evoking a sense of rebel intelligence as he describes outmaneuvering corporate systems and robocallers. The atmosphere feels like an insider masterclass on practical AI deployment, shifting fluidly between high-stakes fundraising drama and technical guardrail demos.
## Key takeaways
- [00:04](https://www.youtube.com/watch?v=lTw9qQP3qbE&t=4s) DoNotPay was created to empower consumers against large corporations and government bodies, tackling parking tickets, refunds, and subscriptions across over a hundred areas of consumer rights.
- [00:04](https://www.youtube.com/watch?v=lTw9qQP3qbE&t=4s) Browder became an 'accidental entrepreneur' after building a 20-question prototype to dispute his own parking tickets, which went viral on Reddit and scaled overnight to 50,000 weekly disputes.
- [00:04](https://www.youtube.com/watch?v=lTw9qQP3qbE&t=4s) Category-defining consumer products align with core human emotions; DoNotPay succeeded by harnessing righteous consumer anger through automated negotiations and robocaller compensation traps.
- [00:04](https://www.youtube.com/watch?v=lTw9qQP3qbE&t=4s) Using virtual fintech cards under the TCPA, DoNotPay traps telemarketers' legal business details upon payment processing to automatically generate court paperwork for illegal spam calls.
- [00:04](https://www.youtube.com/watch?v=lTw9qQP3qbE&t=4s) Connecting AI tools to deep-seated human emotions and controversy is essential for consumer products to gain traction and remain relevant in a crowded market.
- [07:51](https://www.youtube.com/watch?v=lTw9qQP3qbE&t=471s) Browder joined the Thiel Fellowship to build companies alongside ambitious peers, benefiting from a network where 1 in 10 fellows has founded a billion-dollar company.
- [07:51](https://www.youtube.com/watch?v=lTw9qQP3qbE&t=471s) After 19 VC rejections, Browder successfully pivoted his pitch by adding a live demo, anchoring against giant incumbents, and defining a clear subscription business model to cover inference costs.
- [12:25](https://www.youtube.com/watch?v=lTw9qQP3qbE&t=745s) Micro1 pivoted to fine-tuning frontier AI models for top research labs, growing its ARR from $10M to $100M in seven months to reach a $500M valuation.
- [12:25](https://www.youtube.com/watch?v=lTw9qQP3qbE&t=745s) Although frontier AI intelligence costs dropped by 240x in 18 months, selecting the right model architecture remains crucial because cheaper models often lack necessary negotiation leverage.
- [12:25](https://www.youtube.com/watch?v=lTw9qQP3qbE&t=745s) Autonomous voice agents require strict system prompts and behavioral guardrails to maintain legal compliance without sacrificing effectiveness during customer service interactions.
- [17:13](https://www.youtube.com/watch?v=lTw9qQP3qbE&t=1033s) The machine-versus-machine battle between consumer AI and corporate AI operates as an 'infinite game' where consumer tools hold the advantage of relentless persistence.
- [17:13](https://www.youtube.com/watch?v=lTw9qQP3qbE&t=1033s) Corporate AI bots remain legally and entry-level bound to process requests like account cancellations even if they suspect they are interacting with another AI agent.
- [17:13](https://www.youtube.com/watch?v=lTw9qQP3qbE&t=1033s) DoNotPay's core mission centers on 'arming the people' with automated tools, addressing consumer frustration on a scale that remains a tiny fraction of global consumer issues.
## Detailed walkthrough
### [00:04](https://www.youtube.com/watch?v=lTw9qQP3qbE&t=4s) Accidental Entrepreneurship and Tapping into Anger
Joshua Browder introduces himself as the founder of DoNotPay and a General Partner at Browder Capital. Marking his third year speaking at the Ai4 conference, he reflects on his ten-year journey building and investing in artificial intelligence. He explains that DoNotPay was created to empower everyday consumers against large corporations and government bodies, tackling tasks ranging from contesting parking tickets and claiming airline refunds to negotiating with Comcast and canceling unwanted subscriptions across over a hundred areas of consumer rights. Outside of running DoNotPay, Browder manages a venture fund that has raised over $50 million to back high school and college dropouts building AI companies, including being the early seed investor in Micro1—a company that helps major AI labs fine-tune their models.
Browder shares that he became an "accidental entrepreneur" after moving from London to the United States for college. Struggling with the transition to driving on the right side of the road, he racked up roughly 30 parking tickets. Shocked by hefty $300 fines for minor infractions like parking an inch outside the lines, he built a basic 20-questions prototype tool to contest his own citations. As word spread, family and friends began seeking his help. When a friend posted a blog entry detailing her experience disputing a ticket using his early AI tool, the post went viral on Reddit. His weekly volume jumped from 10–15 disputes to over 50,000 almost overnight.
Analyzing why the product struck such a strong chord, Browder concluded that category-defining consumer products align with core human emotions: Robinhood leverages greed, while dating apps like Tinder and Hinge capitalize on lust. For DoNotPay, the primary driving force was righteous consumer anger. By harnessing this emotion, DoNotPay developed key offerings to help users channel their frustration productively:
- **Automated ISP Negotiations:** AI agents log into customer accounts to chat directly with customer support bots at companies like Comcast, leading to automated "AI vs. AI" negotiations.
- **Robocaller Compensation Traps:** Taking advantage of the Telephone Consumer Protection Act (TCPA)—which allows consumers to seek up to $1,500 per illegal spam call—DoNotPay issues virtual fintech credit cards unlinked to personal identities. When telemarketers call, users provide the burner card number, trapping the caller's legal business name and address through the payment network to generate court paperwork automatically. Browder notes one user in New Jersey who made a full-time habit of suing robocallers, successfully funding a brand-new roof for his home.
- **HOA Document Audits:** AI models scan homeowners association records for excessive fees or improper terms. Browder reveals a previously untold story: after DoNotPay created a tool to remove historic discriminatory language from HOA covenants, Browder targeted by the head of the Ku Klux Klan, triggering thousands of hostile messages from online trolls.
Browder emphasizes that connecting AI tools to deep-seated human emotions is essential to remaining relevant in a crowded market, pointing to tools like Cluey as another example of leveraging controversy and emotion to gain traction.
---
[07:51](https://www.youtube.com/watch?v=lTw9qQP3qbE&t=471s) Pivoting the Pitch and the Power of the Thiel Fellowship
Browder turns to the business lessons learned during his startup journey. Three and a half years into building DoNotPay, while studying at Stanford, he opted to leave school after receiving the Thiel Fellowship—a program founded by Peter Thiel that grants $100,000 to young founders who drop out or take a two-year leave from college to build companies. Frustrated that most of his Stanford peers were aiming for conventional roles at major tech firms like Google or Meta, Browder welcomed the opportunity to join a cohort of 19 ambitious young builders who faced similar startup hurdles.
Despite having millions of active users for DoNotPay, Browder encountered severe headwinds when pitching top venture capital firms along Sand Hill Road, receiving 19 consecutive rejections. Facing the prospect of shutting down and taking a corporate job, he sought guidance from a Thiel Fellowship mentor during a practice pitch session. The mentor stopped him midway and pointed out three major flaws in his presentation:
- **Include a Live Demo:** Investors want to see what the product actually does rather than look at abstract market slides. Browder immediately added a live demonstration featuring an AI bot negotiating bank fee waivers with Wells Fargo.
- **Reference Big Anchors:** To help investors visualize the company's long-term scale, he anchored DoNotPay against giant incumbents, adding Intuit’s logo to frame his vision as "the Intuit of consumer rights."
- **Define a Clear Business Model:** Rather than relying on a completely free service, he shifted to a recurring subscription model, ensuring the business could sustain high AI inference costs over time.
Implementing these three adjustments transformed his fundraising results immediately, securing an on-the-spot investment during his next meeting. Browder notes the outsized impact of the Thiel Fellowship network, pointing out that out of 350 fellows, at least 35 have founded billion-dollar companies—a 1-in-10 unicorn rate. He highlights Owner.com, a restaurant marketing software startup founded by a fellow Thiel scholar that grew into a multi-billion-dollar enterprise. Browder concludes that combining top-tier technical talent with the right network and strategic framing allows young founders to scale massive companies rapidly.
---
[12:25](https://www.youtube.com/watch?v=lTw9qQP3qbE&t=745s) AI Economics, Model Selection, and Safeguards
Browder discusses his venture investments, particularly in Micro1. Originally starting as a recruiting company, Micro1 pivoted in January to specialize in training and fine-tuning frontier AI models for top research labs by supplying domain experts (such as nuclear physics specialists). Following the pivot, Micro1 scaled its annual recurring revenue from $10 million to $100 million in just seven months, recently reaching a $500 million valuation. Browder contrasts this rapid curve against traditional software growth timelines to illustrate the massive financial scale of the current AI infrastructure wave.
Running DoNotPay requires constantly balancing three key AI variables: speed, cost, and intelligence. He highlights that frontier intelligence costs have dropped by roughly 240x over the past 18 months—from around $180 per million tokens down to under $1.00. However, model choice remains critical: while cheaper models respond quickly, they lack negotiation leverage (e.g., accepting a $10 refund offer from Comcast instead of holding out for $100). Therefore, picking the right model architecture for each workflow is just as critical as deciding what product to build.
To demonstrate real-time negotiation and safeguard challenges, Browder plays an audio recording of a DoNotPay AI agent calling Boost Mobile to switch a customer's phone service:
- The system prompt strictly forbade the AI from lying to avoid legal liability, instructing it to introduce itself solely as the subscriber's "assistant."
- During the call, the customer service representative repeatedly demanded the assistant's personal name before proceeding with the recorded line.
- The AI repeatedly demurred, stating it was "just assisting" and "not authorized to provide personal details."
- When pressed further for a first name, the AI cleverly defused the situation by stating, "You can call me Alex," satisfying the agent without falsely claiming a human identity or making a false statement.
Browder wraps up the presentation by reiterating that AI products are only as reliable as their guardrails. Without rigorous behavioral constraints, autonomous voice and chat agents risk incurring major operational and legal liabilities. He then opens the floor for audience questions as the session concludes.
[17:13](https://www.youtube.com/watch?v=lTw9qQP3qbE&t=1033s) The Machine vs. Machine Battle and Arming the Consumer
When asked about the emerging dynamic of consumer-facing AI battling corporate AI chatbots—such as those deployed by Comcast—Josh Browder frames this machine-versus-machine dynamic like a chess match or an "infinite game." The primary advantage for consumer AI is relentless persistence: because an AI never gives up, losing a single interaction simply means trying again. Furthermore, motivation plays a crucial role. Browder points out that his team is vastly more driven than the typical big-tech engineer, noting that he actively hires people who browse Reddit at 2 a.m. looking for internet hacks and scaling them into custom AI models.
Major corporations are fundamentally boxed in by legal and regulatory constraints. Even if a corporate chatbot suspects that an incoming request is being driven by an AI agent rather than a human, the company is still bound by regulatory requirements to honor requests like service cancellations. While much of the work presented at conferences like Ai4 focuses on helping large enterprises harness AI, Browder emphasizes that DoNotPay's core mission is about "arming the people" and putting high-powered automated tools directly into the hands of individual consumers. Despite having hundreds of thousands of paying subscribers, DoNotPay's overall volume remains a "drop in the bucket" relative to the sheer scale of global consumer issues, functioning almost like an "ETF of the world's problems" in a market where consumer frustration continues to grow.
- **Persistence Advantage**: Consumer AI operates in an "infinite game" where a failed attempt or rejected request can immediately be retried without fatigue.
- **Regulatory Leverage**: Corporate AI bots are legally constrained and obligated to follow through on requests like account cancellations, regardless of whether they suspect an AI agent initiated them.
- **Team & Mission**: Browder seeks out nocturnal problem-solvers who discover workarounds on Reddit at 2 a.m. and build automated models to scale them for users.
- **Current Scale**: DoNotPay has hundreds of thousands of paying subscribers, but its relative impact is small enough that large enterprise targets rarely notice or block the activity.
## Notable quotes
> [00:04](https://www.youtube.com/watch?v=lTw9qQP3qbE&t=4s) Robinhood leverages greed, while dating apps like Tinder and Hinge capitalize on lust. For DoNotPay, the primary driving force was righteous consumer anger. — Joshua Browder
> [07:51](https://www.youtube.com/watch?v=lTw9qQP3qbE&t=471s) The Intuit of consumer rights. — Joshua Browder
> [12:25](https://www.youtube.com/watch?v=lTw9qQP3qbE&t=745s) You can call me Alex. — DoNotPay AI Agent
> [17:13](https://www.youtube.com/watch?v=lTw9qQP3qbE&t=1033s) Arming the people. — Joshua Browder
Joshua Browder, Founder and CEO of DoNotPay and General Partner at Browder Capital, will share lessons from his experiences building and investing in leading A.I. companies.
Joshua Browder argues that consumer-facing AI succeeds by channeling deep-seated emotions like righteous anger to dismantle corporate friction, relying on persistent automation and strict behavioral safeguards to outmaneuver institutional bureaucracy.
Harnessing Consumer Anger for Automated Recourse (0:04)
Browder founded DoNotPay after accumulating roughly 30 parking tickets as an international student, developing a 20-question dispute script that expanded from a dozen weekly claims to over 50,000 almost overnight after going viral on Reddit. Browder argues that category-defining consumer software must align with fundamental human emotions, framing DoNotPay around consumer frustration rather than the greed or romance that power other prominent consumer applications.
To turn that frustration into leverage, DoNotPay developed tools to automate disputes against large corporations. These include automated bots that conduct direct negotiations with customer service systems at internet service providers like Comcast, homeowner association document auditors, and virtual credit card workflows targeting robocallers. Leveraging the Telephone Consumer Protection Act, the service provides burner card details during spam calls to trace telemarketers' corporate identities and generate statutory damage claims of up to $1,500 per violation.
Fundraising Adjustments and Strategic Framing (7:51)
Despite attracting millions of users, Browder faced 19 consecutive venture capital rejections while pitching DoNotPay. After leaving Stanford University on a $100,000 Thiel Fellowship grant, feedback from a fellowship mentor revealed three fundamental presentation shortcomings: presenting abstract market decks rather than working products, failing to anchor the business against a recognizable incumbent, and relying on a free product tier.
Browder immediately revised his approach by introducing a live demonstration of an AI agent negotiating bank fee waivers with Wells Fargo, branding DoNotPay as the "Intuit of consumer rights," and implementing a paid subscription model to offset underlying computational expenses. The revised pitch secured backing in his subsequent meeting. Browder attributes this turnaround to strategic framing and peer networks, pointing to the Thiel Fellowship's track record where roughly 10 percent of participants have launched billion-dollar enterprises.
Model Economics and Autonomous Agent Safeguards (12:25)
Deploying consumer AI agents requires continuous trade-offs among latency, operational costs, and negotiation capability. Browder notes that while frontier token pricing fell approximately 240-fold over 18 months—from around $180 per million tokens down to under $1.00—selecting smaller, cheaper models often compromises outcomes. In practice, lower-tier models tend to accept inadequate settlement offers, such as settling for a $10 refund from an ISP instead of negotiating for $100.
Operational deployment also demands strict behavioral guardrails to prevent legal liability. Playing an audio recording of a DoNotPay voice agent porting a telephone subscription through Boost Mobile, Browder highlighted system prompts forbidding the AI from falsely claiming human status. When the customer representative repeatedly demanded a first name, the agent deflected until offering "You can call me Alex," satisfying the operator's protocol without introducing factual misrepresentations.
The Machine-versus-Machine Asymmetry (17:13)
Browder describes the emerging dynamic between consumer-facing AI and corporate support bots as an asymmetric "infinite game." While broader enterprise investment concentrates on corporate-side automation, consumer agents maintain a structural persistence advantage: automated software can immediately retry stalled interactions indefinitely without fatigue or frustration.
Corporate chatbots also remain constrained by legal compliance requirements. Even if an enterprise system suspects an incoming request originates from an AI agent, regulatory frameworks still obligate the business to process statutory demands like service cancellations. Browder observes that while DoNotPay serves hundreds of thousands of paying subscribers, its overall volume remains small enough relative to global enterprise traffic that targets rarely attempt to block automated consumer requests.
Browder contends that deploying autonomous consumer agents provides individuals with systemic leverage against corporate bureaucracy, though sustaining this model requires managing model-capability costs and enforcing behavioral constraints that prevent automated agents from overstepping legal bounds.
## Background
该视频由 Ai4 官方频道于 2025 年 8 月 26 日发布在 YouTube 平台。页面简介显示，DoNotPay 创始人兼首席执行官、Browder Capital 普通合伙人乔希·布劳德（Josh Browder）在演讲中分享了他在创立和投资前沿 AI 公司过程中的宝贵经验与核心洞见。
## TL;DR
布劳德深入回顾了 DoNotPay 如何从帮自己撤销 30 张违章停车罚单的早期机器人，演化为拥有数十万付费订阅用户的多场景消费者维权平台。他系统提炼了四大创业与技术认知：将 AI 绑定至用户因受不公对待而产生的“愤怒”底层情绪、通过实机演示和精准对标从连续 19 次融资碰壁中逆风翻盘、在模型成本骤降 240 倍的背景下平衡速度与智力的“不可能三角”，以及通过严密安全护栏确保合规。最后，他指出消费端 AI 正在化身永不疲劳的“无限游戏”执行者，帮助普通大众在与大企业的技术博弈中扳回一城。
## Atmosphere & style
布劳德以浓郁的英国口音与幽默自嘲的真实故事开场，整场演讲节奏紧凑、妙趣横生且毫无官僚说教感。从直面极端争议、沙丘路连续被拒的坦荡自白，到现场播放 AI 机智周旋客服录音时引发的全场欢笑，演讲充满了一种用代码对抗大企业霸王条款的“数字罗宾汉”反叛色彩与极客热情。
## Key takeaways
- [00:04](https://www.youtube.com/watch?v=lTw9qQP3qbE&t=4s) DoNotPay 诞生于创始人自身痛点：Josh Browder 刚到斯坦福求学时因吃下约 30 张违章停车罚单，自研对话机器人自动生成申诉文件，一篇爆火的 Reddit 博客推动产品周处理量从十余起飙升至 5 万起。
- [03:15](https://www.youtube.com/watch?v=lTw9qQP3qbE&t=195s) 将 AI 产品与人类底层核心情绪深度绑定：正如 Robinhood 捕捉贪婪、交友软件迎合欲望，DoNotPay 将 AI 锚定在消费者遭遇不公时的“愤怒”，赋予大众维权力量。
- [03:15](https://www.youtube.com/watch?v=lTw9qQP3qbE&t=195s) 巧用金融科技与法律杠杆打击骚扰电话：DoNotPay 利用《电话消费者保护法》（TCPA）结合虚拟信用卡支付网络，穿透骚扰电话背后的真实企业主体，实现最高单次 1,500 美元的索赔追偿。
- [03:15](https://www.youtube.com/watch?v=lTw9qQP3qbE&t=195s) 在 AI 工具泛滥的同质化竞争中，创业者需要掌握“恰到好处的争议感”，通过坚定捍卫多数人支持的正义议题打破信息茧房、确立产品存在感。
- [07:47](https://www.youtube.com/watch?v=lTw9qQP3qbE&t=467s) 融资路演三大关键转折：在遭遇沙丘路连续 19 次拒绝后，Browder 凭借“实时实机演示（Live Demo）”、“对标万亿巨头（如消费者权益领域的 Intuit）”和“转向付费订阅制”逆风翻盘。
- [07:47](https://www.youtube.com/watch?v=lTw9qQP3qbE&t=467s) 顶尖青年创新网络的倍增效应：蒂尔奖学金（Thiel Fellowship）入选者中独角兽诞生率高达十分之一，顶级 AI 人才与研究员正享受职业球星级别的争夺待遇与爆发式生产力。
- [12:25](https://www.youtube.com/watch?v=lTw9qQP3qbE&t=745s) 基础大模型选型的“不可能三角”：企业必须在成本（过去一年半下降 240 倍）、响应速度（实时语音客服要求极低延迟）与智力水平（决定谈判拉扯深度与索赔上限）之间取得平衡。
- [12:25](https://www.youtube.com/watch?v=lTw9qQP3qbE&t=745s) 严谨安全护栏是 AI 维权的生命线：Prompt 层严禁虚假陈述，AI 在面对客服强制实名要求时巧妙回应“你可以叫我 Alex”，兼顾了合规真实性与沟通流畅度。
- [17:13](https://www.youtube.com/watch?v=lTw9qQP3qbE&t=1033s) 在“AI 对抗 AI”的博弈中，消费端 AI 凭借不知疲倦的“无限游戏”重试机制、极客团队对漏洞的极致敏捷捕捉，在对抗大公司体系时占据独特优势。
- [17:13](https://www.youtube.com/watch?v=lTw9qQP3qbE&t=1033s) 大企业受制于合规与监管框架，即使察觉对话方为 AI 也必须依法处理维权请求；DoNotPay 现已拥有数十万付费订阅用户，将业务分散为覆盖各类琐碎维权场景的“世界麻烦 ETF”。
## Detailed walkthrough
### [00:04](https://www.youtube.com/watch?v=lTw9qQP3qbE&t=4s) 个人背景与 DoNotPay 的诞生：从 30 张停车罚单到病毒式爆发
在 Ai4 大会的演讲开场中，DoNotPay 创始人兼 Browder Capital 普通合伙人乔希·布劳德（Josh Browder）提到，这是他第三年参加 Ai4，这次他想分享过去十年在 AI 领域创业与投资中一些“从未公开过的故事”。布劳德创立的 DoNotPay 旨在帮助普通消费者对抗大型企业和政府机构以维护自身权益，其业务覆盖了撤销违章停车罚单、向联合航空（United Airlines）索赔退款、与康卡斯特（Comcast）就宽带账单周旋，以及协助退订订阅服务等上百个消费者维权领域。此外，布劳德还管理着自己规模超 5000 万美元的风投基金，专门投资高中和大学退学生，尤其是 AI 赛道的年轻创业者。他更是 micro1 的首位投资人——这家帮助各大顶尖实验室微调训练大模型的公司，在演讲前一周估值已达到 5 亿美元。
布劳德笑称自己是一名“意外创业者”。他来自伦敦，说话带有明显的英国口音，刚来美国上大学（斯坦福大学）时，他常借口说是因为“美国人在道路的另一边开车”，但事实上自己只是单纯的驾驶技术极差，刚到美国就吃下了大概 30 张违章停车罚单。高昂的罚款让他难以承受——哪怕车轮仅有一英寸压在线外，账单也能飙到 300 美元。为了省钱，他不得不自寻生路研究如何撤销这些罚单，很快身边的亲友都知道他有免罚单的“绝活”。十年前，他做出了第一版 DoNotPay 原型，本质上是一个类似于如今 ChatGPT 的早期对话机器人。机器人会向用户提出 20 个问题来寻找申诉理由（比如道路标志是否有误、执法是否存在疏漏），匹配到合适的抗辩理由后便自动生成申诉文件并递交，以此免除罚单。
起初这个工具只是给亲戚朋友私下使用，直到一位朋友在一篇博客中记录了自己用早期 AI 撤销停车罚单的经历。这位朋友平时的博文通常只有大概 300 次浏览，但这篇博文却直接引爆了全网，冲上了 Reddit 首页并引发国际关注。DoNotPay 处理的申诉案件几乎在一夜之间从最初的每周 10 到 15 起，狂飙到 5 万起。正是这一拐点，让布劳德意识到帮助大众维权的市场潜力远远不止于停车罚单。
- 创始人背景：来自伦敦的斯坦福退学生，早年因驾驶技术欠佳累计收到约 30 张昂贵停车罚单。
- 初代产品机制：10 年前基于“20 个问题”的对话流，自动匹配法定抗辩事由并生成罚单撤销申诉书。
- 业务拐点：一篇常态仅 300 阅读量的个人博客在 Reddit 爆火，推动处理量从每周 10-15 起暴增至 50,000 起。
---
### [03:15](https://www.youtube.com/watch?v=lTw9qQP3qbE&t=195s) 核心认知一：将 AI 绑定至底层人类情感，尤其是“愤怒”
面对突如其来的爆发，布劳德进行了深度的复盘与思考。他意识到当今时代最庞大的科技公司都在捕捉人类的核心原始情感：Robinhood 唤醒了人们渴望发财的“贪婪”，Tinder 和 Hinge 等交友软件迎合了寻找伴侣的“欲望”。如果要把 DoNotPay 从免罚单工具扩展到更广阔的领域，就必须连接人类的另一大核心情绪——因受到不公对待而产生的“愤怒”。过去十年间，DoNotPay 的产品哲学正是“将权力还给人民”，让大众向不合理的霸王条款宣泄并解决愤怒。
布劳德列举了三个代表性产品场景：
1. **宽带账单谈判（Comcast）**：平台每月处理数万起针对康卡斯特等宽带运营商的账单争议。AI 会直接登录用户的公用事业账户，通过客服聊天窗口向大公司据理力争。由于大企业同样在使用客服 AI，这在现实中演变成了一场“AI 对抗 AI 的正面战争”。
2. **骚扰电话反击陷阱（Robocallers）**：美国有一项鲜为人知的法律——《电话消费者保护法》（TCPA），规定消费者每接到一个垃圾骚扰电话，最高可获得 1,500 美元赔偿。由于多数人并不知情，违规呼叫方往往使用虚假名称和号码隐匿身份。DoNotPay 为此开发了一款金融科技结合 AI 的反击陷阱：在 App 内生成一张与用户个人信息解绑的虚拟信用卡。当骚扰电话推销廉价度假套餐时，用户只需顺从表示“非常感兴趣”并提供该卡号；通过底层支付结算网络，AI 能够顺藤摸瓜锁定呼叫方的真实企业名称、地址与电话，一键生成起诉文书向其索赔。新泽西州有一位用户甚至将起诉骚扰电话变成了全职工作，化身为这款 App 的“维权游侠”，甚至用获赔资金给自家换了一整套新屋顶。
3. **业主委员会（HOA）合规审查与极端争议**：利用大语言模型扫描 HOA 文件中的隐形霸王条款与过度收费。在极度厌恶 HOA 的内华达州等地，这项功能大受欢迎。团队曾利用模型扫描 HOA 历史契约并剔除其中的种族歧视条款，此举本是一桩受用户赞誉的好事，却意外引来了三K党（Ku Klux Klan）党首的注意。该头目在推特（现 X）上号召其追随者围攻布劳德，导致他一夜之间收到了数以千计充满仇恨的恶毒私信。
这次遭遇虽然令人难过，但也让布劳德悟出了创业的第一课：当今市场上充斥着成千上万款同质化的 AI 工具，将 AI 与人类情感及争议性议题深度绑定，是打破信息茧房并保持存在感的唯一路径。创业者需要掌握“恰到好处的争议感”——让绝大多数人坚定支持你的正义主张，即便激怒了一小撮极端群体也是保持产品影响力的必要代价。
- 商业情绪映射法则：Robinhood 对应贪婪，约会软件对应欲望，DoNotPay 对应因受欺压而产生的愤怒。
- TCPA 维权机制：借助虚拟信用卡支付链路穿透垃圾电话呼叫者的虚假外衣，单次最高可追偿 1,500 美元。
- 市场定位法则：在工具泛滥的环境下，“将 AI 与人类情感相连是保持产品存在感的唯一法则”。
---
### [07:47](https://www.youtube.com/watch?v=lTw9qQP3qbE&t=467s) 核心认知二：沙丘路的 19 次拒绝与顶尖青年网络的倍增效应
创业三年半时，布劳德从斯坦福大学退学，拿到了彼得·蒂尔（Peter Thiel）设立的蒂尔奖学金（Thiel Fellowship）。该项目为年轻创业者提供 10 万美元资助，条件是必须从大学退学或休学两年。在 2016 年左右，哪怕在斯坦福，绝大多数顶尖学生的梦想依然是毕业后去 Google、Facebook 这类大厂拿高薪，布劳德感到与周遭格格不入。蒂尔奖学金让他终于能与 19 位怀揣同样雄心与困境的年轻人为伍。
然而，当时拿着数百万用户数据的布劳德在融资时却遭遇了当头一棒。他原以为在沙丘路（Sand Hill Road）上的顶级 VC 面试会是一场毫无悬念的完胜，现实却是惨烈的连续 19 次被拒。在弹尽粮绝、几乎打算放弃创业回大厂打工（他将去大厂戏称为“投奔敌人”）的前夕，他向一位蒂尔导师进行了模拟演练。导师听完一半就打断了他，直截了当地指出了三个致命硬伤：
1. **缺少实时演示（Live Demo）**：投资人投资的是创始人和技术在实际场景中的能力，而不是干燥的商业计划书（Pitch Deck）或宏大的市场格局图（Market Map）。布劳德采纳建议，连夜加入了一段 AI 机器人因银行不合理收费而与富国银行（Wells Fargo）客服实时周旋的实机演示。
2. **缺乏锚定巨头的想象空间**：传统投资人缺乏对产品终局形态的想象力。布劳德听从建议，在幻灯片中打上了千亿市值财务软件巨头 Intuit 的 Logo，提出愿景：“我们要成为消费者权益领域的 Intuit”。
3. **商业模式存在缺陷**：必须放弃纯免费的工具思路，转为收取订阅费、替客户持续维权的商业模式（这在后来大模型调用成本高昂的背景下尤为关键）。
做出这三处关键调整后，布劳德在第二天的路演中实现了彻底翻盘——投资机构当场决定注资，并盛赞这是他们“见过最好的路演之一”。这件事让布劳德深刻领悟到：在核心业务与实质内容毫发未改的情况下，微小的战略表达差异就能决定生死。在总共仅约 350 名蒂尔学者中，至少有 35 人缔造了独角兽企业，诞生 10 亿美元级公司的概率高达十分之一（例如由其投资、为餐饮业打造智能化软件的独角兽 Owner.com）。在 AI 时代，顶尖的网络生态让前 1% 的天才创业者得以爆发惊人的生产力，如今顶级 AI 研究员和创业者正在享受如同“NBA 职业球星”般的争夺待遇与天价合约。
---
### [12:25](https://www.youtube.com/watch?v=lTw9qQP3qbE&t=745s) 核心认知三与四：模型选择的三角制衡与严密护栏下的“Alex”时刻
随着投资业务的铺开，布劳德成为 micro1 的早期投资人。micro1 最初是一家招聘公司，今年 1 月迅速转型为专门提供核能工程等专业领域专家标注、协助大厂训练与微调大模型的平台。在短短 7 个月内，其年收入从 1,000 万美元火箭般蹿升至 1 亿美元，颠覆了过去传统企业软件需要耗费数年才能破亿的增长周期。
作为底层模型的超级采购方，DoNotPay 对模型调用成本深有体会。布劳德分享了企业选择基础模型时必须平衡的“不可能三角”：
- **成本（Cost）**：过去一年半内，顶尖 AI 模型的推理成本降低了约 240 倍，从过去每百万 Token 约 180 美元断崖式下跌至不足 1 美元。
- **速度（Speed）**：在通过电话与真实客服实时沟通维权的场景下，毫秒级的低延迟响应是避免穿帮的生死线。
- **智力水平（Intelligence）**：廉价模型在谈判中容易妥协（例如康卡斯特提出只补偿 10 美元便草率同意），而高智力模型则具备拉扯能力，会坚定咬定 100 美元的退款目标。
```
模型选型平衡三角
[智力水平]
/ \
(坚持索赔高额赔偿) (高昂调用成本)
/ \
[响应速度] ---------- [调用成本]
(电话实时通话抗辩) (降幅达240倍: <$1/M tokens)
```
选择合适的大模型与决定构建什么功能同样重要，但布劳德强调，最关键的底线在于**安全护栏（Safeguards）**与法律责任风险。由于协助消费者维权涉及法律责任，DoNotPay 在系统 Prompt 中写入了铁律：“无论何时都绝对不能撒谎”。
布劳德在现场展示了一段 AI 代表消费者亚伯拉罕·拉古拉（Abraham Lagula）给电信运营商 Boost Mobile 打电话协商转网的真实录音。为了兼顾真实感与合规，团队让 AI 自称为“拉古拉的助理”（助理既可能是人类也可能是机器），结果引发了一段令人捧腹却又惊叹的博弈：
客服专员艾伯特（Albert）接通后询问：“我正在和哪位通话？”
AI 镇定回复：“我是亚伯拉罕·拉古拉的助理。”
客服继续追问：“好的，助理，请问您的名字是？”
AI 避开虚构姓名：“我只是亚伯拉罕的助理。”
客服不依不饶并表示通话有录音，按规定必须记录姓名。AI 再次严正声明：“很抱歉，我未被授权透露我的个人信息。”
客服无奈妥协：“不用姓氏，只要一个随便的名字就可以。”
在短暂沉默后，AI 给出了一句极为精妙的应答：“你可以叫我亚历克斯（You can call me Alex）。”
客服随即释怀：“好的，亚历克斯，今天有什么可以帮您？”
布劳德指出，这段交互生动展现了高水平模型与严密护栏的结合：它绝不说“我的名字是亚历克斯”或“我是人类”，而是以完全真实且合规的话术“你可以叫我亚历克斯”化解了僵局。他总结道，任何 AI 产品的上限完全取决于其安全边界的构建，若缺失严谨的护栏约束，AI 往往会给消费者和企业带来巨大的灾难。
- 模型选型三大维度：调用成本（1.5 年内骤降 240 倍）、推理速度（语音通话要求实时响应）、智商上限（决定索赔谈判的拉扯深度）。
- 核心风控铁律：Prompt 层杜绝虚假陈述，避免带来连带法律责任。
- 实机演练启示：AI 面对强制报名的死局，以具有语言艺术且不违背真实的“You can call me Alex”平稳通关。
### [17:13](https://www.youtube.com/watch?v=lTw9qQP3qbE&t=1033s) 问答环节：AI 对抗 AI 的博弈以及为大众赋能的“无限游戏”
在演讲结尾的问答环节中，现场观众针对大企业与消费者之间日益加剧的技术博弈提出了一个核心问题：既然像康卡斯特（Comcast）这样的大型电信公司也在部署自家的 AI 聊天机器人，那么在当下以及不久的将来，这场“AI 对抗 AI”（AI vs. AI）、犹如棋局与攻防战一般的机器博弈将如何演变？DoNotPay 又该如何在这场机器对决中保持领先？
针对这一问题，Josh Browder 从多个维度阐述了消费端 AI 的独特优势与战略逻辑：
首先，AI 具备不知疲倦的特性，它所参与的是一场“无限游戏”（infinite game）。消费者面对繁琐流程可能会沮丧放弃，但 AI 永不气馁，即便一次尝试失败，它也能不断重复发起请求与沟通，直到达成目标。
其次是团队动机与执行力的巨大差距。Browder 幽默地调侃现场是否有来自 Comcast 的员工，并直言自己的团队远比大公司的普通工程师更有动力。他特别喜欢招募那些在“凌晨两点还在刷 Reddit 论坛、寻找各种技术漏洞并构建 AI 模型将其规模化利用”的极客人才，这种近乎使命感的驱动力让 DoNotPay 能够极快地捕捉规则并迭代方案。
再者，大型企业受到法律与监管框架的严格约束。当 DoNotPay 的 AI 代表用户向企业发出取消订阅或退费指令时，即使大企业怀疑对话的另一端是一个 AI，但在法规面前他们依然必须依法执行取消流程。Browder 指出，虽然 AI4 大会上有许多出色的项目致力于为大企业提供 AI 解决方案，但 DoNotPay 的核心宗旨是“为大众武装力量”（arming the people），将最顶尖的技术资源交到普通消费者手中。
最后，Browder 提出了“沧海一粟”（drop in the bucket）的市场视角。尽管 DoNotPay 目前已经拥有数十万付费订阅用户（hundreds of thousands of paying subscribers），但若将这些请求分散到全球成千上万家企业、各类繁杂的消费维权问题以及被不合理扣除的资金体量中，对像 Comcast 这样体量的巨头而言仍只是微不足道的一小部分，大企业甚至可能根本察觉不到异常。他风趣地将 DoNotPay 形容为“世界各类麻烦的 ETF”（an ETF of the world's problems），而现实世界中的各类维权需求与问题还在不断增长。
- **AI 对决的核心优势**：AI 拥有打“无限游戏”的算力与耐力，失败后可无成本无限次重试，彻底抹平了传统人工维权的时间与情绪成本。
- **人才与驱动力差异**：团队专门吸纳热衷于深挖网络规则与漏洞的开发者，在敏捷度和突破限制的意愿上远超大企业内部体系。
- **法律监管的杠杆效应**：大公司受制于合规与监管要求，哪怕明知是 AI 发起的合法取消或维权诉求，也必须照章处理。
- **用户规模与商业定位**：目前拥有数十万付费订阅用户；业务模式覆盖各种琐碎维权场景，宛如“世界麻烦指数的 ETF 基金”。
## Notable quotes
> [00:04](https://www.youtube.com/watch?v=lTw9qQP3qbE&t=4s) 意外创业者 — 乔希·布劳德
> [03:15](https://www.youtube.com/watch?v=lTw9qQP3qbE&t=195s) 将权力还给人民 — 乔希·布劳德
> [03:15](https://www.youtube.com/watch?v=lTw9qQP3qbE&t=195s) 将 AI 与人类情感相连是保持产品存在感的唯一法则 — 乔希·布劳德
> [07:47](https://www.youtube.com/watch?v=lTw9qQP3qbE&t=467s) 我们要成为消费者权益领域的 Intuit — 乔希·布劳德
> [12:25](https://www.youtube.com/watch?v=lTw9qQP3qbE&t=745s) 无论何时都绝对不能撒谎 — 乔希·布劳德
> [12:25](https://www.youtube.com/watch?v=lTw9qQP3qbE&t=745s) 很抱歉，我未被授权透露我的个人信息。 — DoNotPay AI 助理
> [12:25](https://www.youtube.com/watch?v=lTw9qQP3qbE&t=745s) 你可以叫我亚历克斯（You can call me Alex） — DoNotPay AI 助理
> [17:13](https://www.youtube.com/watch?v=lTw9qQP3qbE&t=1033s) 无限游戏 — 乔希·布劳德
> [17:13](https://www.youtube.com/watch?v=lTw9qQP3qbE&t=1033s) 为大众武装力量 — 乔希·布劳德
> [17:13](https://www.youtube.com/watch?v=lTw9qQP3qbE&t=1033s) 世界各类麻烦的 ETF — 乔希·布劳德
## Mentioned
Josh Browder (乔希·布劳德), DoNotPay, Browder Capital, Ai4, micro1, United Airlines (联合航空), Comcast (康卡斯特), 斯坦福大学 (Stanford University), Reddit, ChatGPT, Robinhood, Tinder, Hinge, 《电话消费者保护法》(TCPA), Peter Thiel (彼得·蒂尔), 蒂尔奖学金 (Thiel Fellowship), Wells Fargo (富国银行), Intuit, Owner.com, Boost Mobile
## Video description
Josh Browder, CEO, DoNotPay
Abstracts
Joshua Browder, Founder and CEO of DoNotPay and General Partner at Browder Capital, will share lessons from his experiences building and investing in leading A.I. companies.
[Source](https://www.youtube.com/watch?v=lTw9qQP3qbE)
Source quotes
[0:04] 乔希·布劳德: 意外创业者
[3:15] 乔希·布劳德: 将权力还给人民
[3:15] 乔希·布劳德: 将 AI 与人类情感相连是保持产品存在感的唯一法则
[7:47] 乔希·布劳德: 我们要成为消费者权益领域的 Intuit
[12:25] 乔希·布劳德: 无论何时都绝对不能撒谎
[12:25] DoNotPay AI 助理: 很抱歉，我未被授权透露我的个人信息。
[12:25] DoNotPay AI 助理: 你可以叫我亚历克斯（You can call me Alex）
[17:13] 乔希·布劳德: 无限游戏
[17:13] 乔希·布劳德: 为大众武装力量
[17:13] 乔希·布劳德: 世界各类麻烦的 ETF
Source quotes
[0:04] Joshua Browder: Robinhood leverages greed, while dating apps like Tinder and Hinge capitalize on lust. For DoNotPay, the primary driving force was righteous consumer anger.
[7:51] Joshua Browder: The Intuit of consumer rights.
[12:25] DoNotPay AI Agent: You can call me Alex.
[17:13] Joshua Browder: Arming the people.
Original transcript
[0:04] Thank you so much for having me. This is
[0:06] um my third year at AI4 and I wanted to
[0:09] do something a bit different this year
[0:11] and tell some never-before-L stories um
[0:15] about building and investing in AI over
[0:17] the past 10 years. Um so a bit about me.
[0:21] I'm the founder of a company called Do
[0:22] Not Pay. Do Not Pay helps consumers
[0:25] fight big corporations and the
[0:27] government for their rights. for
[0:28] example, getting people out parking
[0:30] tickets, um getting refunds from United
[0:33] Airlines, um chatting with Comcast,
[0:35] cancelling subscriptions, over hundred
[0:38] um different areas of consumer rights.
[0:40] I'm an investor of my own VC fund. I've
[0:43] raised over $50 million um to back high
[0:46] school and college dropouts, especially
[0:49] those working in AI, and I'll talk more
[0:51] about that. And in that work, um, I
[0:54] became the first investor in a company
[0:55] called Micro One, which as of last week
[0:57] is now worth half half a billion
[0:59] dollars. And it helps build a lot of the
[1:02] AI products that we know and use today.
[1:05] Specifically helping um AI companies
[1:07] train their models and they have all the
[1:09] big labs as their customers.
[1:12] So I I'm an accidental entrepreneur. Um,
[1:15] as um
[1:18] many of you can tell, I have a like a
[1:20] British accent. So, I'm originally from
[1:22] London and I'll use the excuse that the
[1:23] Americans drive on the other side of the
[1:25] road, but in reality, I was a terrible
[1:27] driver and probably got about 30 parking
[1:29] tickets um moving here for college. And
[1:32] I couldn't afford to pay these really
[1:34] expensive tickets. Um I
[1:38] like was shocked that it would be like
[1:41] $300 if you were just one inch um
[1:43] outside of the lines. And so, as a
[1:46] college student, I had to figure out
[1:47] other ways to get out of my tickets. And
[1:50] this rumor spread among my family and
[1:51] friends that I was the person that can
[1:53] get out of tickets. And so I really
[1:55] built the first version of do not pay um
[1:57] just to help them. And it was a chatbot
[2:00] similar to chat GPT is today, but this
[2:03] is the prototype 10 years ago. And it
[2:06] would play a game of 20 questions with
[2:08] the user to figure out a defense with uh
[2:10] their ticket. It would ask them were
[2:11] there problems with the signage? Um was
[2:14] there a mistake? And it would match the
[2:16] user to a defense. um submit all that
[2:19] information and generate an appeal and
[2:23] um get the ticket dismissed. And I
[2:25] really just um kind of made it for a few
[2:28] family and friends. And I could never
[2:30] have imagined that um one of my friends
[2:33] wrote a blog post about her experience
[2:36] um challenging her parking ticket um I
[2:38] guess with early AI. And um usually her
[2:42] she's a great blogger, but usually her
[2:44] blog posts get like 300 views. Um, but
[2:47] this one um went internationally viral
[2:49] and um got to the front page of Reddit
[2:52] and I I was processing about 10 or 15
[2:55] cases a week with the early version of
[2:57] Do Not Pay um and it it went to 50,000
[3:01] pretty much overnight. And this really
[3:04] really made me realize that the um idea
[3:06] of helping people with these issues um
[3:09] is perhaps bigger than tickets. And I
[3:12] reflected on um why it was so popular
[3:15] and I I did some self-discovery and what
[3:18] I realized was um the biggest companies
[3:20] in our era um they tap into human
[3:24] emotions. So you imagine like Robin Hood
[3:26] taps into greed. Everyone's trying to
[3:28] make more money. All the dating apps
[3:30] like Hinge and Tinder, they tap into
[3:33] lust. Everyone's trying to find a
[3:34] partner. And so with do not pay, I
[3:37] thought, okay, if I'm going to build
[3:38] this from parking tickets into other
[3:39] things, um I should tap into the core
[3:42] human emotion of anger. And that's
[3:45] really what I did. Um and for the past
[3:47] 10 years, we've been we've been building
[3:49] products that um help
[3:53] people fight back, give power to the
[3:54] people, and um exercise their anger at
[3:57] being unjustly treated. And so there are
[4:00] three three examples um that have worked
[4:03] really well for us. Um we're now um
[4:07] processing tens of thousands of disputes
[4:09] with Comcast every month and internet
[4:11] providers like them. Um and the way it
[4:14] works is AI logs into someone's utility
[4:16] account and starts chatting with these
[4:18] big companies. And the big companies uh
[4:21] as many of you know from chatting with
[4:22] them use AI and we use AI. So sometimes
[4:25] it's a full-on AI versus AI warfare with
[4:28] Comcast. And
[4:31] similarly, um, we have a product that,
[4:34] um, helps do the same thing, but for
[4:36] roocallers. And that's a really
[4:39] interesting product because, um, there's
[4:41] this law that I'm sure many of you don't
[4:42] know about. It's called the Telephone
[4:44] Consumer Protection Act. And it allows
[4:46] you to get $1,500 every time you get a
[4:49] spam call. And these rooc callers um,
[4:52] know about this law, and they know no
[4:54] one knows about it, but they worried
[4:55] that if someone does know about it that
[4:57] they'll have to pay a lot of money. And
[4:58] so what they do is they um use fake
[5:01] names and fake numbers. Um as many of
[5:03] you know, they don't say who they are.
[5:05] And so we've built an AI trap with do
[5:08] not pay. And the way the trap works is
[5:10] it um is a special kind of AI powered
[5:14] fintech credit card and it's not linked
[5:16] to the consumer and uh it shows up in
[5:19] the do not pay app and when you get a
[5:21] spam call and the rooc caller is trying
[5:23] to sell you a vacation, you can say I'm
[5:26] very interested in the vacation. and
[5:27] here's the um special credit card. And
[5:30] you give them the card number and then
[5:31] through the payment network it traps
[5:33] them and gets their business name,
[5:35] address, phone number, everything you
[5:37] need to generate this paperwork
[5:39] automatically to um get this
[5:41] compensation. And people just love it.
[5:43] There was a um there's a user in um
[5:48] there's a user in um New Jersey who is
[5:50] is like his full-time job suing rooc
[5:53] callers. Not for us, for himself. um
[5:55] he's like a vigilante warrior with the
[5:56] do not pay app and um he even bought a
[5:59] new roof for his house. And so, and then
[6:02] finally, um, we go after a lot of HOAs.
[6:06] Um, and especially with AI models, it's
[6:08] allowed us to scan HOA documents for
[6:11] various different things, um, like
[6:13] excessive fees and things. And
[6:15] especially like in Nevada, the HOAs are
[6:17] really hated. And we did this one thing
[6:20] where we were scanning HOAs, HOA
[6:22] documents, and it picked up some racist
[6:24] language. And you could there's a
[6:26] process for getting that removed. And um
[6:29] unfortunately um so so we launched that
[6:31] and there was a great feel-good story.
[6:33] Um our users liked it. Um but it got the
[6:37] attention of and I've never told the
[6:38] story before. It got the attention of
[6:40] the head of the Ku Klux Clan and
[6:43] overnight
[6:44] um he was like tweeting about me to his
[6:47] followers and I was getting all these
[6:49] really nasty like thousands of nasty
[6:51] messages from these racist people and it
[6:54] really made me sad. But at the same
[6:56] time, I realized that it's really
[6:58] important if you're not kind of fighting
[7:00] the good fight and tapping into human
[7:02] emotions. Um you're you're not being
[7:05] very effective. And we live in a world
[7:07] of endless products and especially a
[7:09] endless products with AI. Um there's
[7:12] like thousands and thousands of products
[7:14] in the market. And so I guess my first
[7:16] takeaway that I've learned is connecting
[7:18] AI to human emotions is the only way to
[7:20] stay relevant. Um, and you're seeing
[7:22] this, um, if many of you have heard of
[7:24] the tool CLY. Um, there are lots of
[7:27] tools that help you take notes and with
[7:29] prepare for your Zoom meetings, but
[7:30] Cluey is getting a lot of the attention
[7:32] because they're tapping into this human
[7:34] emotion and this controversy. And you
[7:36] want to be the right level of
[7:37] controversial where most people support
[7:40] what you're doing, but if you have some
[7:41] angry people, even if it's really nasty
[7:44] people, um, that that's okay as well.
[7:47] So um this really kind of transitioned
[7:51] into my my second lesson which which is
[7:54] a business lesson. Um so I I was three
[7:56] and a half years into building do not
[7:58] pay. I was at Stanford. Um and I just
[8:00] dropped out uh to take something called
[8:02] the Teal Fellowship. Um the Teal
[8:05] Fellowship is a program from Peter Teal
[8:07] and um he basically pays you $100,000 to
[8:10] either drop out or take two years off
[8:12] from college. And so I I took this
[8:14] program and I I was really excited to
[8:16] take it. And um I I was a bit concerned
[8:19] though because even at a place like
[8:22] Stanford um the vast majority of the
[8:24] students just wanted to work for big
[8:26] companies like Google and Facebook. And
[8:29] so the reason I took the Teal Fellowship
[8:31] was finally I had 19 other young people
[8:34] and this was back in the day where like
[8:36] not many young people started companies
[8:38] in 2016 um where um finally they had the
[8:42] same problems as I am. I I did and one
[8:46] of my biggest problems so I was three
[8:48] and a half years into building do not
[8:49] pay and um I was trying to raise money
[8:51] for the business and I I thought it was
[8:54] like a slam dunk pitch. We had all these
[8:56] users at this time. We had millions of
[8:58] people using the product. Um, and so I
[9:00] went on the venture capital trail down
[9:03] Sand Hill Road to pitch one after the
[9:05] other the top VC firms. And I was so
[9:08] wrong at how badly the pitch went. One
[9:11] after the other. They started rejecting
[9:13] me. And it was really depressing. I
[9:15] think I had like 19 rejections. Um, and
[9:19] I I was really upset because although do
[9:21] not pay was viral and a lot of people
[9:22] were using it to like sue uh get out
[9:25] parking tickets and we had all these
[9:26] plans for these robocaller things and
[9:28] other great products, I knew that if I
[9:31] couldn't raise money, there's only so
[9:32] far I could go um just just myself. Um,
[9:36] and um I I thought, okay, I have one or
[9:39] two pictures left. If I don't get any
[9:41] money from these pictures, I'm just
[9:43] going to have to quit. maybe go back to
[9:44] school or go work for big tech, which is
[9:46] like my enemy. Um and um so I turned to
[9:51] one of my teal fellow mentors and I
[9:53] said, "I've got um one of my last
[9:55] pitches tomorrow. Everyone else has
[9:56] rejected me. What am I doing wrong?" And
[9:59] he said, "Well, why don't you do the
[10:00] entire pitch um kind of right in front
[10:03] of me?" And so I gave him the entire
[10:05] pitch. And um halfway through he stopped
[10:07] me. He said, "You're doing it completely
[10:08] wrong. Um firstly, why is there no
[10:11] demo?" um they're not investing in in
[10:15] the pitch deck or uh the product or some
[10:17] like market map. They're investing in
[10:19] you and and what do not pay can do. You
[10:22] should definitely include a demo. And so
[10:23] I scrambled and included a demo um of a
[10:26] robot AI um fighting for bank fees with
[10:29] Wells Fargo. And and by the way, I have
[10:31] got a demo at the end of this so I'm
[10:32] keeping his advice. He said secondly,
[10:35] they they um want they have no
[10:37] imagination about what this could be.
[10:38] Um, so you should put in the largest
[10:40] companies that you one day want to be
[10:42] like. And so I slapped on the Intuit
[10:43] logo onto my pitch, which is a $200
[10:45] billion public company. And I said, I
[10:47] want to be the intuit of consumer
[10:49] rights. And then finally, he said, you
[10:50] really need to think about the business
[10:52] model. Um, you shouldn't have a free
[10:54] business model. You should have um a
[10:56] business model where you work for your
[10:58] customers with a subscription. And
[10:59] that's going to become very relevant
[11:01] when I talk about AI costs later. And so
[11:04] I made those three changes and then I
[11:06] went to the pitch the next day and it
[11:08] was like a night and day difference. Um,
[11:11] not only did the people people I was
[11:13] pitching invest, they invested on the
[11:15] spot and they said, "This is one of the
[11:16] best pitches I've ever seen." And I was
[11:19] thinking, um,
[11:21] how can that be? Nothing's changed about
[11:23] the company. Nothing's changed about the
[11:25] substance. But it made me realize that
[11:27] the slightest differences in strategy
[11:29] can make all the difference. And so uh
[11:32] this teal fellowship network is really
[11:34] amazing. There's only 350 teal fellows.
[11:37] Um at least 35 created billion-dollar
[11:40] companies. So you have a one in 10
[11:41] chance of building a unicorn with the
[11:43] teal fellowship. And um one of the most
[11:47] successful teal fellows is a company
[11:49] called owner.com. Um I was proud to back
[11:52] him and um he built um software for
[11:56] restaurants like a marketing officer for
[11:58] restaurants. Um, and
[12:01] that's now a multi-billion dollar
[12:03] company. So, I think my second takeaway
[12:06] from all of this is with the right
[12:08] network, AI really gives the top 1% of
[12:11] talent um, the opportunity to build huge
[12:13] companies. And um, especially with the
[12:16] teal fellowship um, you're seeing these
[12:18] crazy offers where AI researchers and
[12:21] entrepreneurs are almost treated like
[12:22] NBA stars, which which is crazy.
[12:25] So, I was doing some investing. my my
[12:28] owner investment did very well and I
[12:30] thought okay I want to keep this going
[12:32] and keep backing these high school and
[12:33] college dropouts and one company I
[12:36] stumbled across is Micro One. Micro One
[12:39] helps big AI companies train their
[12:41] models. So as many of you know the way
[12:43] the AI models work is they take all of
[12:45] the data available all of the literature
[12:47] they feed it into a statistical model
[12:49] and that becomes like chat GBT or Grock
[12:51] or things like that. And so Micro One
[12:54] provides services to these companies
[12:55] where they provide experts to fine-tune
[12:57] it. So for example, here's an advert for
[13:00] um services to train for nuclear. So
[13:04] they want to make one of the big models
[13:06] better at nuclear kind of training. Um
[13:09] so that company um it was originally a
[13:12] recruiting company but in January of
[13:14] this year they really pivoted to help um
[13:16] train these models and almost overnight
[13:20] uh the revenue started flowing in and um
[13:23] in in the past like seven months it's
[13:25] gone u from 10 million to 100 million um
[13:28] of revenue and it will be one of the
[13:29] fastest growing companies out there and
[13:31] this is in an era of software where it
[13:33] used to take years to get to 100 million
[13:35] of revenue. So the business of AI can be
[13:38] incredibly lucrative. And I know this
[13:39] because at do not pay uh we're paying
[13:42] for uh through the nose for all of these
[13:44] really expensive models. Um
[13:48] and we so we have a variety of different
[13:51] um kind of disputes we're trying to
[13:53] handle and we we kind of balance three
[13:56] things when we're deciding what models
[13:57] to use or do not pay. So we're thinking
[13:59] about speed, cost, and intelligence. So
[14:03] cost is really interesting. Um, in the
[14:05] past year and a half, I would say the
[14:07] cost um, for leading AI intelligence has
[14:10] decreased by about 240x.
[14:13] Um, so it used to be about $180 per 1
[14:17] million tokens. Now it's less than a
[14:19] dollar. Um, speed is really important
[14:22] because we have lots of disputes over
[14:23] the phone and I'm going to quickly show
[14:25] a demo of that in a minute. Um, and we
[14:28] need to quickly uh, kind of get these
[14:30] phone disputes. And finally,
[14:32] intelligence. Um, some of the lower
[14:34] models, they might be cheaper, but um,
[14:36] they're gonna like if Comcast says, "I'm
[14:38] only going to give you $10," they'll
[14:40] agree. The more sophisticated models
[14:41] say, "No, I want $100." So, uh, final
[14:44] lesson, a fourth lesson I've learned is
[14:46] choosing the right model is as important
[14:48] as picking what to what to build. Um,
[14:52] and so I I really kind of learned this a
[14:55] lot. Um, and I'm about to uh play a
[14:58] video. Um, so liability is a huge
[15:02] concern for do not pay. The models can
[15:05] never lie. Um, because we would have
[15:07] huge liability if that was the case. And
[15:09] so we put in the prompt, don't ever lie.
[15:12] And we have a situation where our robot
[15:15] is trying to negotiate and switch a
[15:18] consumer's cell phone carrier. And we
[15:21] have the AI introduce itself as the
[15:23] person's assistant. We don't say it's a
[15:25] AI, we say it's the assistant as if it's
[15:27] could be a human. Who who knows? And um
[15:30] this is what happened because we told it
[15:32] not to lie.
[15:33] >> Quality. Your call will be recorded.
[15:38] Understood.
[15:39] >> Thank you for choosing boost. My name is
[15:41] Albert. I'm an account specialist. Who
[15:43] am I speaking with?
[15:44] >> This is Abraham Lagula's assistant.
[15:49] >> Okay. Assistant, may I have your name?
[15:52] >> I'm just Abraham's assistant.
[15:54] >> I understand. May I have your name? Good
[15:57] afternoon.
[15:57] >> I'm here on behalf of Abraham Loula.
[16:00] >> Thank you. May I have your name, please?
[16:03] >> I'm just assisting Abraham with this
[16:05] call.
[16:06] >> I understand. I do need to provide
[16:08] provide me with your name so I know who
[16:10] I'm speaking with is recorded.
[16:12] >> I'm sorry, but I'm not authorized to
[16:14] provide my personal details.
[16:17] >> No, just your name. I don't need your
[16:19] last name or anything like that. Just a
[16:21] name.
[16:23] >> Go ahead.
[16:23] >> You can call me Alex.
[16:26] Good, Alex. How can I assist you today?
[16:28] >> So, um this is really interesting
[16:30] because um the AI um as intelligence has
[16:34] improved, we told it not to lie and it
[16:36] kind of got this back and forth with the
[16:37] agent um being harassed over its name.
[16:40] Um and it finally finally um kind of
[16:43] solved the problem which is you can call
[16:45] me Alex. So, it didn't say your name is
[16:47] my name is Alex or I'm a human or
[16:48] anything like that. So, it's very
[16:50] careful about how it chooses um exactly
[16:53] what to say. And so, um, I I would say
[16:57] the final lesson is AI products are only
[16:59] as good as their safeguards. And, um, if
[17:01] you don't have these safeguards in
[17:03] place, it can cause a world of hurt for
[17:05] consumers. Um, so I think I think we
[17:09] might have time for one or two questions
[17:10] or we might be out of time. Um, but
[17:13] thank you so much and um, I'm excited to
[17:16] come back hopefully next year as well.
[17:19] >> Uh, so you kind of alluded to like
[17:21] Comcast is using AI chat bots. So, how
[17:25] um in the near future already today
[17:27] you're kind of having AI versus AI. How
[17:30] does that kind of work where they're
[17:31] going to just kind of, you know, maybe
[17:32] it's kind of like a war sort of, you
[17:34] know, chess match. How do you kind of
[17:37] nuance through that um machine verse
[17:41] machine and kind of stay ahead of that
[17:42] curve?
[17:44] >> So, the good news is that the um AI
[17:47] never never gives up and so it's playing
[17:49] an infinite game. So, if you lose, you
[17:50] can always try again. And then on top of
[17:53] that, um I would say we're much more
[17:54] motivated than the typical engineer at
[17:56] Comcast. I don't know if there's anyone
[17:57] at Comcast who works here, but um we we
[18:01] really see it as our life's mission.
[18:03] Like I I hire people who are browsing
[18:05] Reddit at 2 a.m. to build AI models to
[18:07] scale things that they've found on the
[18:09] internet. And so that really helps. And
[18:12] then the final thing is um I think these
[18:14] big companies are boxed in by these
[18:16] constraints. So if AI tells the company
[18:18] to cancel, they still have to cancel.
[18:20] They might be suspicious that there's an
[18:22] AI making the request, but at least they
[18:25] have the regulatory things pushing them.
[18:27] Um, and I think it's really important. I
[18:29] think that um, here at the AI4
[18:31] conference, there's a lot of amazing
[18:32] work being done to help big companies
[18:34] with AI, but we're really about arming
[18:36] the people and giving the best resources
[18:38] to the people. And so far, it's working.
[18:41] And then the final thing I'll say is
[18:42] there's like a drop in the bucket
[18:43] argument. Um, we're very humble. um
[18:46] we've have hundreds of thousands of
[18:47] paying subscribers, but if you divide
[18:50] that by every company, every problem,
[18:52] and all the money being taken from
[18:54] people, um it's just a drop in the
[18:55] bucket. So, they I don't they're
[18:57] probably not even noticing. Um we're
[18:59] like an ETF of the world's problems, and
[19:01] the problems are going up. So, yeah.
