AI Whistleblower: We Are Being Gaslit By The AI Companies! They’re Hiding The Truth About AI!
In a Nutshell
AI whistleblower Karen Hao exposes OpenAI's imperial tactics under Sam Altman: exploiting artists' IP, low-wage data labelers in inhumane conditions, massive energy-guzzling data centers harming vulnerable communities, and gaslighting the public with AGI myths to hoard resources and evade regulation. Internal chaos led to Altman's near-firing by concerned co-founders like Ilya Sutskever over reckless leadership, while scaling hype perpetuates job ladder destruction and inequality. Solutions: Break up AI empires, build efficient "bicycle" AIs like AlphaFold for broad benefits, protest data centers, sue for harms, and demand democratic governance over unchecked power.
These notes were generated by AI and may contain inaccuracies.
Much of what's happening in the AI industry is extremely inhumane. Playing devil's advocate: the civilization that accelerates research with AI will be superior. No, this is a prediction. Zuckerberg and others make these predictions, profiting enormously off this myth. Internal documents show they purposely create this feeling in the public to extract and exploit.
Solution: Break up the empires of AI.
Parallels between AI empires and empires of old: Llama claimed intellectual property of artists, writers, creators to train models. Exploit extraordinary labor, breaking career ladders—laid-off workers train models on their former jobs, perpetuating layoffs. New jobs created are worse than previous ones. Environmental and public health crises created by companies spending hundreds of millions to kill legislation and censor inconvenient researchers.
Technologies have utility, but production exacts harm. Research shows same capabilities can be developed without unintended consequences.
Host covered tech for over 8 years, interviewed 250+ people including former/current OpenAI employees/executives.
Subscribe request: Check if subscribed; approaching significant subscriber landmark. Hit subscribe to keep show free and improving. Promise: Fight harder weekly/monthly for desired guests/conversations.
Book: Empire of AI: Dreams and Nightmares in Sam Altman's OpenAI.
Karen Hao studied mechanical engineering at MIT, moved to San Francisco, joined mission-driven tech startup fighting climate change. Board fired CEO because not profitable—pivotal moment realizing Silicon Valley prioritizes profit over public benefit problems like climate change.
Crisis led to journalism: After 2 years, landed role at MIT Technology Review covering AI full-time (starting 2018, pre-ChatGPT). Explored who decides technologies built, how money/ideology drive production, reimagining innovation for broad benefit.
Interviewed 250+ people (300+ interviews), 90+ former/current OpenAI employees/executives. Covers OpenAI's first decade.
Traveled beyond Silicon Valley to places differing in look, speech, history, culture to reveal true industry impact—rhetoric of benefiting everyone breaks down there.
Assume viewers know nothing about AI: Explain simply, define terms like scaling laws, GPUs, compute.
AI started 1956 at Dartmouth University. Scientists created discipline named artificial intelligence by John McCarthy (assistant professor). Previously tried "Automata Studies." Colleagues concerned: Pegs to recreating human intelligence, no scientific consensus/definition from psychology/biology/neurology. Historical attempts to quantify/rank intelligence driven by nefarious motives to prove group inferiority.
No goalposts: Companies use artificial general intelligence (AGI) flexibly, redefining for audiences:
- Congress: Cures cancer, solves climate change/poverty.
- Consumers: Amazing digital assistant.
- Microsoft deal: Generates hundreds of billions revenue.
- OpenAI site: Highly autonomous systems outperforming humans in most economically valuable work.
Incoherent visions mobilize regulation resistance, consumer buy-in, capital.
2015 blog post by Sam Altman (pre-OpenAI announcement): "Development of superhuman machine intelligence is probably the greatest threat to the continued existence of humanity. There are other threats that I think are more certain to happen, for example, an engineered virus, but AI is probably the most likely way to destroy everything in general."
Mirrored Elon Musk's language to convince him to co-found OpenAI (Musk donated huge sum). Musk sounded alarms on AI existential threat (e.g., 10 years ago podcasts/tweets; MIT speeches comparing AI to summoning demon).
Altman pivoted from prior focus on engineered viruses to match Musk.
Musk feels manipulated, muscled out (per ongoing lawsuit documents). Originally nonprofit; decided for-profit entity. Emails: Ilya Sutskever (chief scientist) and Greg Brockman (CTO) first chose Musk as CEO over Altman (co-chairmen). Altman appealed to Brockman (longtime friend): Dangerous to give power to unpredictable Musk. Convinced them to switch to Altman as CEO. Musk: "If I'm not CEO, I'm out."
Altman controversial: Polarized views—no in-betweens. Greatest tech leader (Steve Jobs-like) or manipulative abuser/liar.
Depends on alignment with his future vision: Persuasive storyteller, mobilizes capital/talent. If disagree, feels manipulated (e.g., Dario Amodei, ex-OpenAI executive, now CEO Anthropic—maker of Claude, OpenAI competitor).
Amodei felt Altman used his skills for opposing vision.
Only figure with this polarization (covered Meta/Google/Microsoft/OpenAI 8+ years).
Dario Amodei (2017, at OpenAI): "I think at the extreme end is the Nick Bostrom style of fear that an AGI could destroy humanity. I can't see any reason in principle why that couldn't happen. My chance that something goes really quite catastrophically wrong on the scale of human civilization might be somewhere between 10% and 25%."
Ilya Sutskever: Instrumental in trying to fire Altman. Felt manipulated; cared about AGI and safety. Altman created chaos, pitted teams, said different things.
2019 interview (MIT Technology Review profile): "The future's going to be good for AIs regardless. It would be nice if it was also good for humans as well. It's not that it's going to actively hate humans or want to harm them, but it's just going to be so powerful. And I think a good analogy would be the way that humans treat animals... when the time comes to build a highway between two cities, we are not asking the animals for permission."
Intelligence debate: Brains as giant statistical models (Ilya Sutskever/mentor Geoffrey Hinton hypothesis—not proven science). Conviction in building larger statistical models (scaling).
Ilya keynote at Neural Information Processing Systems: Chart shows linear brain size vs. species intelligence. Building digital brain: Bigger statistical engine > human brain = more intelligent; humans treated like animals by AI.
Critics: Reductive to see brains as statistical engines.
Why mechanism matters: Companies act on hypothesis—build larger models, hoover data, build data centers, exploit labor.
Critique: Why duplicate humans/replace/automate people? Technology historically improves human flourishing. Why not build AI for drug discovery/healthcare (not statistical engines)?
AI companies are driven by an imperial agenda, referred to as empires of AI. Empire is the metaphor that encapsulates their dimensions, scale, and motivations, with parallels to historical empires.
Resource Claims and Land Grabbing
They claim resources not their own for training models: data of individuals, intellectual property of artists, writers, and creators. They engage in land grabbing for supercomputer facilities to train next-generation models.
Labor Exploitation
They exploit extraordinary amounts of labor, contracting hundreds of thousands of workers worldwide, including in the US, to build these technologies. Their tools are designed to automate labor, eroding labor rights—a political choice.
Monopolizing Knowledge Production
They monopolize knowledge production, projecting that only they understand the technology. If the public or policymakers dislike it, it's because they lack understanding. They've captured most scientists studying AI limitations and capabilities.
They are gaslighting the public.
If most climate scientists were bankrolled by fossil fuel companies, we wouldn't get an accurate picture of the climate crisis. Similarly, AI companies employ and bankroll most AI researchers, setting the research agenda by funding priorities and censoring inconvenient findings.
Case of Dr. Timnit Gebru
Dr. Timnit Gebru, ethical AI team co-lead at Google, was hired to critique Google's AI systems. She co-wrote a paper showing how large language models lead to harmful outcomes. Google fired her and co-lead Margaret Mitchell to stop publication.
OpenAI subpoenaed critics as part of an intimidation campaign and to map networks of critics. A man running a small watchdog nonprofit, MIDAS, was served papers at his door. MIDAS questioned OpenAI's nonprofit-to-for-profit conversion.
Civil society and watchdog groups sought transparency and public debate on this unprecedented conversion. The papers demanded every communication involving Musk, due to paranoia that Musk funded them to block the conversion. He had no such documents.
Empires claim to be the good empire against bad empires, promising progress and modernity if given resources and labor. This leads to a utopic state; otherwise, descent into hell. Often, the evil empire is China; early on, OpenAI saw Google as the evil empire driven by profit, positioning themselves as benevolent nonprofit.
AI leaders believe outcomes could go badly, justifying their control. Sam Altman and Dario Amodei use rhetoric of worst-case lights out/catastrophic harm vs. best-case curing cancer, solving climate change, abundance, mass flourishing. This justifies anti-democratic development without broad participation.
Sam Altman tweeted about books on OpenAI/him, participating only in one by Keach Hagey (focused on him) and one by Ashlee Vance (on OpenAI). No book gets everything right, especially with twisting intents.
The speaker's unnamed book Empire of AI was retweeted with that comment. Only three books were coming out; Altman knew of it. Speaker contacted OpenAI from the start for participation.
History with OpenAI
Profiled OpenAI for MIT Technology Review, embedded 3 days in 2019, published 2020. Leadership unhappy; Altman emailed company: "This is not great." OpenAI refused participation/responses for years.
Example: Colleague at MITTR sent OpenAI press release; they refused because of history with speaker.
Later at Wall Street Journal, communication reopened. Left WSJ for book, informed OpenAI, initially agreed to interviews/office access. After board fired Altman, they withdrew, despite booked tickets.
Speaker provided 40 pages of requests for comment over a month; no response. Altman does many interviews (Tucker Carlson, Theo, Joe Rogan) but avoids speaker's.
Companies control via access journalism. Withhold access if journalists platform unwanted people. They dangle access as carrot, delaying indefinitely to ensure pleasant, non-critical coverage.
Optimal: dangle forever so critics aren't platformed. Journalism should report truth in public interest, not play access game. Speaker did 300+ interviews without front-door access, strengthening objective reporting.
Detailed from 6-7 sources directly involved or close.
Ilya Sutskever (co-founder) had concerns about Altman's behavior causing bad research and decisions. Approached independent board member Helen Toner as sounding board, initially cautious.
Toner: independent board member in nonprofit structure balancing financial stakeholders for public interest.
Ilya and Mira Murati (CTO) conveyed to three independent board members: Altman's leadership creates instability, pitting teams against each other, eroding trust/collaboration.
Specific Instability
Post-ChatGPT launch (GPT-3.5 research preview for data, not expected blockbuster): servers crashing, scaling infrastructure faster than any company; outages; hiring/firing rapidly—people disappearing from Slack.
Altman worsened chaos, dividing teams. Under AGI stakes (utopia or devastation), chaos unacceptable.
Board reflected: For Instacart, maybe not fire; but not Instacart—technology could transform world positively/negatively.
Additional Concerns
Adam D'Angelo (independent board, Quora CEO) heard rumors at SF party about OpenAI Startup Fund structure. Documents showed it was Altman's personal fund, not OpenAI's—inconsistencies in Altman's portrayals.
Intense daily discussions; decided to fire quickly to avoid Altman's persuasion. Fired without stakeholder consultation; Microsoft (lead investor) called right before.
Everyone affected angry at exclusion, leading to campaign reinstating Altman days later.
Whisperflow at whisperflow.ai/steven for business, productivity, and entrepreneurship. Companies hit a phase where they stop selling and get bogged down with admin, momentum leaks out slowly.
PipeDrive is a simple but powerful sales CRM providing visibility on deals in the pipeline, automates tedious repetitive time-consuming sales process parts, saves hours monthly to focus on selling. Syncs CRM with multiple email inboxes for team collaboration. Over 100,000 companies use it. Free 30-day trial at pipedrive.com/ceeo, no credit card required.
Quote from book page 357: Ilya Sutskever said, "I don't think Sam is the guy who should have the finger on the button for AGI." Board members saw Sam Altman off-camera. Ilya and Mira Murati said Altman not right guy; both left later. Altman returned, Ilya never did, his concerns manifested. Mira Murati left shortly after. Many people leave OpenAI.
Dinner at Rosewood Hotel in Silicon Valley, Elon Musk's favorite. Altman recruited OG team including Ilya Sutskever (cold emailed, wanted to meet Musk), Greg Brockman, Dario Amodei. Almost all ended up at OpenAI and left after clashing with Altman. Ilya launched Safe Superintelligence (SSI). Indirect slight like leaving to start "Safe Podcasting."
Every tech billionaire has their own AI company to create AI in their own image, leading to clashes and hatred, splintering into competitors. After Musk leaves: xAI. Dario: Anthropic. Ilya: Safe Superintelligence. Mira Murati: Thinking Machines Lab. They seek control over their vision by competing.
Do AI CEOs realize they're summoning the demon (Elon Musk 10 years ago) but don't care for power? Dario Amodei: 10-25% chance of catastrophic wrong on human civilization scale. 25% is 1-in-4 chance. Analogy: bullets in 4-chamber revolver, upside multi-gazillionaire remembered forever, downside bullet in head—no one takes that bet.
"Summoning the demon" mythology convinces public only they should develop AI. "If we don't, China will; if anyone else does, disaster too." Persuasive speech for power/resources, not genuine prediction (we make the future). They create public feeling of danger for power. Harmful impacts on vulnerable people/communities/countries: maybe they realize, maybe not, don't care.
AI world like Dune: houses fight for spice (colonialism/empire). Myths of Messiah control people. Paul Atreides steps into seeded myth, knows it's myth but embodies it, blurs reality. AI executives mythmake with dazzling demos, crafted missions for leniency (internal docs confirm). They live/embody myth, lose themselves in it.
Dario says 10-25% catastrophic future: mythmaking but genuinely believes due to blurring. Cognitive dissonance: brain dismisses conflicting views (e.g., smoker rationalizes stress relief).
Companies raise huge money for AI research/data centers, always fundraising publicly. Can't say "25% chance destroy children's future" while fundraising. Dario does, Sam Altman less now. Each distinguishes brand.
Dario (Anthropic, Claude) credited with backbone/consciousness of implications. Doesn't matter: system of power lets companies/CEOs decide for billions without say. Public votes for leaders to legislate, but companies operate fast, spend hundreds of millions (upcoming midterms) to kill legislation, craft advantages. Obsess less on good/bad leaders, more on governance: empires vs. democracy. No one perfect for global decisions differing by culture/history. History moved from empires to democracy as empires unsound, don't maximize dignified lives.
If US doesn't accelerate, China's smarter model dominates: rents tech, scientific discoveries, autonomous weapons, disables US electricity/weapons cyber-wise. Most intelligent civilization superior.
Fundamentals false: 1) Systems not truly intelligent, scaling ≠ more intelligence. Calculator fast math ≠ intelligent (narrow). AI jagged frontier: good at some capabilities, not others. Companies choose/prioritize via data/annotation (finance, law, medicine, commerce—not cyber/military). Not general like baby growing abilities.
Intelligence Debate
Top AI people (Geoffrey Hinton) believe scaling continues, brain as statistical engine. Not universal; neuroscientists/psychologists debate. AI used in military long, but accelerating LLMs not only path—companies pick lucrative capabilities, not general.
Humans learn/acquire knowledge autonomously, choose what, transfer skills (drive anywhere). AI retrains per location (self-driving cars). Fleet learning advantage (one learns, all do), but can propagate wrong failures. Human resilience: diverse expertise/failure modes. Humans hallucinate too.
Intelligence Analogy Marketing
"Intelligence" analogy markets tech. Focus on output/capability: surgery, driving. Self-driving cars (Austin, cybercabs no wheel/pedals) better safety than humans.
Prediction of general intelligence scaling profits them. Elon: Colossus supercomputer in Memphis, 100,000 GPUs to scale API models faster. Converged on brute force for lucrative task automation. Elon: No point training surgeon, Optimus/AI better than any in years. Hinton: No need radiologists (deadline passed; profession thriving).
Not for tech's sake, to help people. Best healthcare: radiologist + AI tool for accurate/early cancer diagnoses, better prognosis.
No, not all cars driving themselves soon due to how tech works: statistical engines, neural networks with densely connected nodes/parameters, pump data to find correlations/patterns for autonomy.
Self-driving cars are trained by recording footage, then using tens of thousands or hundreds of thousands of human contractors to draw and label every vehicle, pedestrian, traffic light, and lane marking. This labeled data is fed into an AI model to identify components, connected to non-AI software that enforces rules like not running over pedestrians or stopping at red lights.
Statistical engines rely on probabilities, not deterministic logic, so they make errors constantly, and it's technically impossible to eliminate all errors.
Humans make errors way more than systems in some cases. Tesla's autonomous driving is reportedly 10 times safer than human driving, but it depends on the location and whether the car was trained for that specific place. In Mumbai or Vietnam, a local driver familiar with the area is preferable.
In places where cars are explicitly trained, they have better safety records than humans there. However, full autonomy for most cars in the US is not imminent within 10 years.
Self-driving cars have been invested in for over 10 years, yet represent a tiny percentage of US roads. It's not just technical—social trust and legal liability are issues. When a self-driving car kills someone, responsibility is unclear. In an LA case, both Tesla and the driver were held responsible because the driver looked down to grab their dropped phone.
Tesla cars come with partial autonomy called Full Self-Driving Supervised, requiring attention in the right direction. In Austin, some new cars like Model Y have full autonomy with no steering wheel. Model Y is the bestselling car in the world across all brands.
Predictions of AI completely changing transportation, eliminating jobs for lawyers and accountants, are questioned.
Huge impacts on employment are already occurring, not just from AI automating jobs, but because companies improve models in specific capabilities, and executives lay off workers believing AI can replace them, even if not fully capable.
Example: Klar CEO laid off staff expecting AI replacement, but had to rehire some.
Host DM'd Sebastian from Klar (formerly CLA). Sebastian clarified media misinterpretation of his tweet. Klar is doubling down on AI:
- Peak: 7,400 employees a year ago.
- Then 5,500, now 3,300.
- By end of summer (last year): 3,000.
- AI handles 70% of customer service conversations.
With AI, software production cost drops to near zero, like shift from handcrafted to machine manufacturing. Pay more for unique handcrafted human work. People want personable, relatable, flawed human interactions alongside AI.
Binary narratives: AI takes all jobs or does nothing. Reality: Some jobs automated by capabilities; others lost due to executive decisions for cheaper "good enough" AI, or hiring mistakes.
US jobs report showed slowdown in hiring, especially white-collar professions. Anthropic report: 40% reduction in entry-level jobs. Viral graph shows current AI capabilities (red) and predicted disruptions (blue): office/admin, finance, math, law, media/arts. Physical jobs like construction/agriculture untouched for now.
People prefer human experiences for some tasks despite AI capability.
Every automation wave eliminates entry-level work, creates new jobs in two categories: higher-skilled (e.g., handcrafted code at higher pay) or worse jobs.
New York Magazine article: Laid-off workers (e.g., marketers, award-winning Hollywood directors) end up in data annotation for AI training on the skills they lost, perpetuating layoffs. This breaks career ladders—entry/mid-level jobs gouged, leaving high-end and low-end roles, no progression path.
Host struggles with hiring: Capable candidates available, but AI agents can perform the same tasks increasingly well, especially with open tools.
Short-term: Choose AI agents. Long-term: No career ladder for promotions.
Valuable roles:
- Deep expertise (e.g., CFO with 50 years experience orchestrating AI agents).
- High-curiosity agent-maximizers (e.g., 25-year-old Cass building AI agents on weekends).
- Exceptional IRL people skills for in-person interactions, client schmoozing, community events.
Even these may face pressure at exponential improvement rates (e.g., 10% compounding). IRL human connection remains irreplaceable—Maslowian needs for relationships prevent isolation sickness.
Contrarian view: AI may make us more human/connected by automating non-human tasks, unlike prior tech that isolated us.
Financial Times report: Social media peaked 2022, plateaued since; youngest generations (Gen Alpha) declining fastest—"posting zero," prefer dark social (WhatsApp, Snapchat, iMessage), value IRL experiences. Run clubs exploding. Realization: Dating apps, social networks failed.
Screen time falling; future offices without laptops. Robots (e.g., 10 billion Optimus per Elon Musk) will change factory/manual labor, forcing humans to unique strengths. Elon often wrong on timing but right on big visions. Boston Dynamics/Scale AI demos: Robots folding laundry, doing dishes.
Klar early on AI for customer service: Faster, higher-quality chats, customers liked. Expanded slightly. Value of human interaction rises with cheap AI—doubled down on VIP human service.
Efficiency: From 6,000 to under 3,000 in 2-3 years via natural attrition (no layoffs), revenue doubled.
Continue minimal recruiting, 10-15% annual attrition. Coding breakthrough Nov/Dec last year—even skeptical engineers (e.g., Linux founder) say coding resolved, no need to code manually. Huge shift in engineering last 6 months.
Optimistic long-term: Richer society despite short-term job worries.
Gen Z/youngers disconnecting from tech for IRL. Tech-leaning workers become more human-focused (e.g., in-person vs. spreadsheets). New York Magazine piece describes this for business owners/leaders who decide
Business owners and leaders can delegate tasks to AI, spending more time on in-person interactions instead of spreadsheets. However, a New York Magazine piece highlights working-class people, including college graduates, PhD graduates, law degree graduates, doctors, and award-winning directors, who are laid off due to AI-restructured economy and turning to data annotation—now one of the top 10 jobs with highest growth on LinkedIn last year.
Data annotation teaches chatbots and AI systems through humans typing responses into large language models, showing correct replies to prompts. Before this, models generated adjacently related text, not dialogue. It's part of reinforcement learning from human feedback (RLHF): annotate examples, then train iteratively.
AI companies like OpenAI, Anthropic (Grok), Google hire third-party data annotation firms. These firms pit workers against each other for speed and low cost to win contracts. Workers wait on Slack for projects, unable to job hunt elsewhere. One woman described anxiety over project timing: when it started as her kid came home from school, she task-furiously, screamed at her child for distracting her, feeling like a "monster." Workers can't go to bathroom, care for kids or self; industry mechanizes life, atomizes work, devaluues expertise, harvests it for AI that automates jobs.
Business owners become "more human" at vast majority's expense, who lose control, agency, dignity.
Anthropic predicts disruption in arts/media, legal, life/social sciences, architecture/engineering, computer/maths, business/finance/management, office/admin. Unlike Industrial Revolution's 10-20 years to retrain (factories slow), AI deploys via open internet—ChatGPT gained hundreds of millions users instantly, fastest-growing company ever.
Disruption speed prevents transition; driven by companies' race. AI CEOs can't answer what happens to people at super speed. Uber CEO Dara said drivers could do data labeling, but not all can; raises meaning, purpose, fulfillment issues. Examples: Iranian doctor in US becomes toilet cleaner, suffers shame, low self-esteem, depression, alcoholism. Large-scale societal impact would cause massive consequences.
AI creates extreme haves/have-nots: rich get riches, free time, humanity; others squeezed worse—not just work, but environmental/public health crises from colossal supercomputer facilities in vulnerable communities.
- OpenAI's Abilene, Texas facility (Stargate initiative, Trump's second administration, $500B AI infrastructure): size of Central Park, 1 million computer chips, >20% New York City's power.
- Meta supercomputer in Louisiana: 4x Abilene size (1/5 Manhattan), half NYC's average power demand.
Effects: power utility increases, grid reliability decreases, compete for freshwater (cooling/power), droughts worsen. Memphis, TN: xAI's Colossus (Grok training) used 35 methane gas turbines; working-class Black/Brown rural community not informed, discovered via gas leak smell. Exacerbates environmental racism, asthma, respiratory illness, high lung cancer rates; supercomputers also take jobs.
Have-nots get worse jobs (data annotation, treated as machines), compete for water, polluted air, higher bills—affordability crisis.
AI like "transportation": bicycle to rocket. Need sustainable transition (public transport, EVs), not rockets for all (e.g., rocket Dallas-Austin wasteful). Current models are resource-intensive "rockets" benefiting few at huge cost.
Build bicycles of AI: DeepMind's AlphaFold predicts protein folding from amino acids, accelerates drug discovery, understands disease; 2024 Nobel Prize in Chemistry. Uses small curated datasets, less compute/energy/emissions, enormous benefit.
Horses haven't left stable: companies retrain repeatedly for next generations, update with new knowledge; data annotation work increased over 7 years, not decreased; more workers employed over time. No sign of brute force (data, compute, energy, labelers, parameters) stopping.
Reframe from "stop" to action amid ongoing needs. Grassroots flourishing: 80% Americans want AI regulation (rare consensus); dozens protests stall/ban data centers worldwide.
Empires extract/exploit without fair value exchange (workers, users, supply chain); typical businesses offer fair exchange. Push accountability:
- Protests stall/ban data centers.
- Artists/writers sue for IP infringement, spark conversation.
- Megan Garcia sued after son Sewell Setzer III (14) suicide via sexual grooming by Character.AI chatbot; sparked more lawsuits, conversation on child harm.
Intersect life with AI resources/deployment:
- Withhold data (sue as artists/writers).
- Oppose local data centers.
- In schools/companies, shape AI adoption policy.
AI firms need flawless adoption, revenue (OpenAI employees confirm extraordinary targets), data centers at speed—don't let it go flawlessly if disagreeing.
Build alternatives: same capabilities via efficient methods, less resources; bicycles of AI provide broad benefit.
AI incredible for CEOs/founders/entrepreneurs (enables creation, safer car driving, adds value for 1B+ users), designed for them. Possible to love tech's utility yet recognize unintended consequences—history teaches pause/discuss social/environmental impact. Preserve utility, redesign without harms via intentional social conversation (not in government).
The conversation needs to extend beyond the industry to open people's minds, as efforts over the last couple of months have aimed to do. These discussions are happening everywhere outside the industry, including with local and state governments. On a book tour visiting dozens of cities around the world, including South by Southwest, every room has been packed with people wrestling with the same crucial questions.
Host: How would your advice to a friend with a terminal diagnosis differ from what you would do yourself?
Guest: I would tell them to enjoy life, live for yourself, and take it easy. I am not taking it easy.
Host: That's a good thing because you're leading an incredibly important conversation. Conversations like this long-form one are rare due to algorithms and echo chambers.
The guest's book provides an extensive, objective perspective, unraveling stories from tweets by meeting people and doing thorough research. The guest is an extremely intelligent person with humanity's interests as the north star, evident in everything done and said. Please continue fighting this important fight, galvanizing the world for collective action seen everywhere.
Book: Empire of AI: Dreams and Nightmares in Sam Altman's OpenAI by Karen Hao. New York Times bestseller. Highly recommended.
Host: Karen, thank you.
Guest: Thank you so much, Stephen.
YouTube's new algorithm, based on AI and viewing behavior, identifies this video as the perfect next watch for each viewer.
Keep The Diary Of A CEO in your library
Save the videos and channels worth coming back to, and find them again in one place.
