E122: Is AI the next great computing platform? ChatGPT vs. Google, containing AGI & RESTRICT Act

Joe Manchin never entered the 2024 U.S. presidential race. Throughout 2023–early 2024 he publicly flirted with a third‑party or independent bid, including work with the centrist group No Labels, but he ultimately declined to run. On February 16, 2024, Manchin stated, “I will not be seeking a third-party run. I will not be involved in a presidential run,” explicitly ruling out a 2024 presidential campaign.(theguardian.com)(washingtonpost.com)(fox29.com)(cbsnews.com)(news.wjct.org)

Later summaries of his career confirm that in 2024 he left the Democratic Party to become an independent and “clarified that he would not run for any office, ending speculation that he might be a candidate in the 2024 United States presidential election.”(en.wikipedia.org) Because he never became an actual presidential candidate in 2024, Chamath’s prediction that “Joe Manchin is running for president” was wrong.

Joe Manchin was widely discussed as a potential independent/third‑party presidential candidate for 2024, including through his association with the centrist group No Labels and public flirtation with a third‑party bid.(theguardian.com)

However, on February 16, 2024, Manchin publicly announced that he would not run for president and specifically said, “I will not be seeking a third-party run. I will not be involved in a presidential run.”(cnbc.com) Subsequent coverage and his biography confirm that he did not enter the 2024 presidential race as an independent or on any third‑party ticket.(theguardian.com)

Therefore, the prediction that “Joe Manchin will not run for U.S. president in 2024 as an independent (third‑party) candidate” was correct.

As of November 30, 2025 (a bit under three years since the March 31, 2023 episode), Google’s Bard/Gemini products have not overtaken ChatGPT in overall consumer usage, but the prediction’s time window (“within a few years”) is vague enough that it hasn’t clearly expired yet.

On direct consumer usage metrics, ChatGPT remains far ahead:

• Similarweb’s August 2025 ranking of AI tools shows ChatGPT with ~5.85 billion website visits and 388M monthly active app users versus Google Gemini’s ~723M visits and 47.6M MAUs; ChatGPT is explicitly called “the most popular AI tool in 2025,” with much higher user stickiness than Gemini.【1search0】
• Market‑share estimates based on website visits consistently put ChatGPT at ~80%+ of the global AI chatbot market, with Google Gemini at only ~2–3% as of mid‑2025.【1search2】2search4】1search7】
• Other traffic and usage analyses in 2025 likewise show ChatGPT handling billions of monthly visits and hundreds of millions of weekly active users, dwarfing Gemini’s traffic.【1search3【2search1】2news14】

On strategic importance in the AI assistant market, evidence again shows ChatGPT still leading as the primary standalone chatbot, while Gemini is an important but smaller competitor:

• Industry overviews find ChatGPT with roughly 60% of the broader generative‑AI/chatbot market, versus low‑teens share for Gemini depending on how the market is defined, still leaving OpenAI in the dominant position.【2search0】2search6】
• Analyses comparing the two note that ChatGPT dominates in native app engagement and direct chatbot usage, while Google leverages Gemini via integration into Search, Android, and Workspace to increase reach, but this has not translated into overtaking ChatGPT in core chatbot usage.【2news13】2news15】

Because the prediction was that “within a few years” of March 31, 2023, Google would “roll” (i.e., overtake) ChatGPT, and we are still within a plausible interpretation of “a few years” (e.g., 3–4+ years), it is too early to say it has definitively failed. What we can say is that, so far, the data show the opposite of the predicted outcome: ChatGPT continues to dominate usage and mindshare, while Gemini has grown but has not surpassed it.

Evidence shows that by March 31, 2024, neither Larry Page nor Sergey Brin had returned to operational leadership of Google/Alphabet as co‑CEOs, nor had they jointly led a major public AI demo event.

Leadership roles

• Alphabet’s key-people listings and corporate history consistently name Sundar Pichai as CEO of both Google and Alphabet through 2024–2025, with no indication that Page or Brin reassumed any CEO title or that a co‑CEO structure was created.(en.wikipedia.org)
• Biographical entries for Page and Brin note that they stepped down from executive roles in December 2019 and remained only as board members/employees and controlling shareholders, with no later update indicating a return to CEO positions.(en.wikipedia.org)
• Updated profiles of Pichai (including non-English editions) still describe him as the current CEO of Google and Alphabet, reinforcing that no leadership handover back to the founders occurred in 2023–2024.(pt.wikipedia.org)

Brin’s return to AI work (but not as co‑CEO)

• Sergey Brin did return to work on AI at Google around 2023 and became heavily involved in the Gemini program, but reports describe this as a hands‑on technical or advisory role, not an executive CEO position.(techtimes.com)

Public demo events

• Google I/O 2024, the flagship public developer/AI event in that window, had its keynote led by CEO Sundar Pichai; announcements and coverage identify him (and other executives) as the speakers, not Page and Brin as co‑CEOs doing live product demos.(blog.google)
• Brin was present at I/O 2024 and interacted with Project Astra demos and press, but coverage characterizes this as a surprise appearance/Q&A and as him checking out demos—not as a co‑CEO headlining a “demo day,” and there is no indication Larry Page shared the stage in such a role.(indianexpress.com)

Given that (1) no co‑CEO titles for Page and Brin were created, (2) Sundar Pichai remained sole CEO of Google/Alphabet, and (3) there was no major public AI demo event jointly led on stage by Page and Brin in that period, the prediction did not come true.

As of November 30, 2025, Jeff Bezos has not announced or filed to run for President of the United States in any election cycle. However, Jason’s prediction was that Bezos “will at some point in the future run for president,” which does not specify a time limit. Since Bezos is still alive and theoretically could run in a future election (e.g., 2032 or later), there is not yet enough information to say the prediction is definitively right or wrong. Thus, the correct status is inconclusive (too early) rather than wrong. (Verified by checking current news and public records up to late 2025, which show no presidential campaign by Bezos.)

The prediction sets a 5–10 year window from March 2023, i.e. roughly 2028–2033, for “major medical breakthroughs based on AI.” As of today (November 30, 2025), we are only ~2.5 years into that window, so it is too early to judge whether the prediction will ultimately be right or wrong.

In the period since 2023, there have indeed been notable AI-driven advances in medicine (e.g., large language models assisting with clinical documentation, image-analysis systems for radiology and pathology, and research models for drug discovery), but the prediction is about what will happen by 2028–2033, not by 2025. Whether those future “major medical breakthroughs” materialize — and are widely recognized as breakthroughs — cannot yet be determined.

Because the forecast horizon has not yet elapsed, the appropriate status is “inconclusive (too early)” rather than right or wrong.

Evidence since March 31, 2023 indicates that AI development and deployment have accelerated, not slowed, despite frequent calls for pauses, regulation, or moratoria.

Key observations (all relative to the 2022–early 2023 baseline when ChatGPT first appeared):

1. Major model releases accelerated in cadence and scale

   • OpenAI moved from GPT‑4 (March 2023) to large multi‑modal and tool‑using capabilities (Vision, Code Interpreter, function calling) and then to GPT‑4.1/4o‑class models and strong edge/mobile integrations, with steadily improving cost, speed, and quality.
   • Anthropic progressed rapidly from early Claude versions to Claude 2, 2.1, and 3‑series models (Opus, Sonnet, Haiku), with each generation showing substantial capability gains and heavier enterprise adoption.
   • Google advanced from PaLM/LaMDA era systems to PaLM 2, Gemini‑class (multi‑modal) models, and tight integration across Search, Workspace, Android, etc.
   • Meta went from research‑only large language models to open‑weight LLaMA 1, then LLaMA 2 and 3, markedly increasing model quality while also catalyzing a large open‑source ecosystem.

   The frequency and magnitude of major model releases and capability jumps since early 2023 is notably higher than in 2020–2022, and the models are deployed into many more products and workflows than before.

2. Broad deployment into consumer and enterprise products

   • General‑purpose AI assistants (e.g., integrated chatbots and copilots) are now embedded in operating systems, search engines, productivity suites, developer tools, CRM/ERP platforms, and design tools. This includes system‑level or first‑party “copilots” from multiple big tech companies and widespread third‑party integrations.
   • Enterprise adoption has expanded rapidly, with many large firms rolling out internal copilots, code assistants, customer‑service bots, and document‑analysis tools, often powered by frontier APIs or strong open‑source models.
   • On the consumer side, AI features (image generation, summarization, translation, smart replies, etc.) are now standard in messaging, productivity, creative tools, and smartphones.

3. Capital, headcount, and infrastructure growth

   • Capital flows into AI have grown dramatically since early 2023: multi‑billion‑dollar strategic investments in frontier‑model companies, massive GPU/accelerator buildouts at hyperscalers, and large private rounds for AI startups across sectors (foundation models, agents, vertical applications, infrastructure).
   • Hyperscalers have raced to secure GPUs and build custom AI accelerators, and global AI compute capacity has grown sharply year‑over‑year—an essential signal that development capacity is expanding, not pausing.

4. Regulatory and "pause" efforts have not produced a slowdown in core development

   • In March 2023, an open letter from the Future of Life Institute called for a 6‑month pause on training systems more powerful than GPT‑4. Despite significant publicity, there is no evidence that major labs actually paused or reduced the pace of R&D; instead, they continued training and deploying more advanced models.
   • The U.S. and EU have advanced regulatory efforts (e.g., the EU AI Act negotiations, the U.S. AI Executive Order, voluntary safety commitments by major labs), but these have not imposed blanket moratoria or substantial slowdowns in frontier‑model training or deployment across the industry. Instead, the pattern has been: continue or accelerate development while adding governance, safety, and reporting layers.

5. Open‑source and academic activity exploded

   • Since LLaMA’s initial leak and later official releases, the open‑source community has produced a rapid cascade of model variants, fine‑tunes, and new architectures. This includes lightweight, on‑device‑capable models and specialized models for code, vision, audio, and agents.
   • Tools, libraries, and frameworks for building AI applications (RLHF pipelines, orchestration frameworks, evaluation tools, agents, vector databases, etc.) have multiplied, further reducing friction to deploying AI.

Overall assessment vs. the prediction

• The prediction was: “Over the coming years, the pace of AI innovation and deployment will not slow in response to calls for a pause; instead, development activity and progress in AI will accelerate relative to the 2022–early 2023 baseline.”
• Between March 31, 2023 and November 30, 2025, we observe:
  • No industry‑wide slowdown or moratorium attributable to pause letters or similar advocacy.
  • A clear increase in the rate of major model releases, infrastructure build‑out, funding, and commercial deployment compared with 2020–2022.
  • Regulatory efforts that largely coexist with, rather than significantly slow, AI R&D.

Given the available evidence as of November 30, 2025, the prediction that AI would not slow down but instead speed up in the face of pause calls is substantially borne out.

Key parts of Chamath’s prediction did not play out as stated.

1. Law firms were early large‑scale adopters of generative AI, not laggards.

   • Allen & Overy began trialling Harvey (a legal LLM platform) in Nov 2022 and by Feb 2023 had 3,500 lawyers using it for ~40,000 queries, one of the first widely reported large‑scale deployments of generative AI in any white‑collar sector. (en.wikipedia.org)
   • On 15 March 2023, PwC announced a global alliance with Harvey that gave 4,000+ legal professionals in 100+ countries access to the platform, explicitly to automate parts of contract analysis, regulatory work, due diligence, and other legal tasks. (pwc.com)
     These dates are contemporaneous with or earlier than the period when large IT services firms were just beginning to publicly frame their generative‑AI strategies, so law was not “waiting” for IT outsourcers to go first.

2. Accounting and legal sectors have moved quickly toward firm‑wide AI use.

   • A 2025 overview of legal AI adoption reports that all AmLaw 100 firms use AI, that 85% of lawyers use AI weekly, and that 21% of firms have firm‑wide AI roll‑outs. (aiqlabs.ai)
   • In accounting, the UK regulator (FRC) found that the largest firms (Deloitte, EY, KPMG, PwC, etc.) are already using AI/ML in audits (risk assessment, information extraction, document review), even if they’re not yet systematically measuring its impact. (ft.com)
     This pattern contradicts the idea that law and accounting were notably slower than IT services to “figure out” AI at scale.

3. IT services firms did scale AI, but later and not clearly ahead of law/accounting, nor specifically targeted first at coding-for-hire displacement.

   • Accenture and TCS now tout very large AI programs—e.g., Accenture nearly doubled its AI & data specialists from 40,000 to 77,000 in two years and trained 550,000+ employees on generative‑AI fundamentals; AI is now “part of everything we do” and underpins thousands of client projects. (crn.com) TCS launched its WisdomNext GenAI platform in 2024 and is infusing AI across its offerings, positioning itself as building one of the largest AI‑ready workforces. (tcs.com)
   • These large‑scale roll‑outs, however, are mostly dated 2024–2025, whereas major law and Big‑4 legal units were already running multi‑thousand‑lawyer generative‑AI deployments in late 2022–early 2023. (en.wikipedia.org)
   • Accenture’s restructuring explicitly links layoffs to the inability to reskill staff for AI‑heavy work, but this is framed broadly (consulting, operations, data/AI roles), not as a clearly targeted program “first” aimed at eliminating coding‑for‑hire jobs. (cnbc.com) Other IT services firms (e.g., TCS) talk about a “Human+AI” model and augmentation rather than openly leading with coder displacement. (timesofindia.indiatimes.com)

4. Job displacement from AI is showing up across white‑collar domains, not uniquely or demonstrably first in coding‑for‑hire firms like Accenture/TCS.

   • Generative AI for coding is indeed one of the most commercially successful use cases—code‑generation startups and tools are attracting huge investment, and reporting suggests shrinking entry‑level dev roles as AI produces more of the code. (reuters.com) A 2025 study finds 97% of IT workers already use generative‑AI tools, with higher organizational adoption correlating with greater job‑security concerns. (arxiv.org)
   • But law and accounting are also now using AI as a lever on staffing: Clifford Chance (a top global firm) is cutting about 10% of its London business‑services staff, explicitly citing increased AI use as a key reason, alongside offshoring and changing demand. (theguardian.com) Big‑4 accountancies have made multiple rounds of layoffs and hiring slowdowns, where AI and offshoring are explicitly discussed as reasons for reducing junior, routine work—essentially treating AI as “the new trainee.” (thefinancestory.com)
     Given this, AI‑related labour effects are diffuse: software engineering, law, and accounting all show early and growing automation of junior and routine tasks. No clear, data‑driven ordering shows Accenture/TCS‑style coding‑for‑hire shops uniquely “going first” ahead of law and accounting in using AI specifically to displace white‑collar labour.

Overall, while software development is indeed one of the earliest and hardest‑hit areas for AI automation, the comparative claim that large IT services / coding‑for‑hire firms like Accenture and TCS would be first, ahead of law and accounting firms, to deploy AI at scale to displace white‑collar labour is not supported by the timeline and available evidence. Law and Big‑4 legal units adopted generative AI at scale very early, and AI‑driven workforce changes are now occurring in all these sectors without a clear lead by the specific firms Chamath named. Hence, the prediction is best judged as wrong overall.

Across 2023–2025, available data and surveys indicate that most companies have so far used AI coding tools to augment software engineers rather than to justify mass elimination of developer roles, which matches Sacks’s short‑term prediction.

Employment and layoffs picture

• U.S. Bureau of Labor Statistics projections continue to show software developers as a high‑growth occupation, with employment expected to rise ~18% from 2023–2033, versus a 9–10% decline for more automatable computer programmer roles. This indicates ongoing demand for developers even as more routine programming shrinks. (bls.gov)
• Reporting on U.S. jobs shows that about a quarter of traditional “computer programming” jobs vanished over the past two years, but the same analysis notes that software developer roles remained comparatively stable and are broader, more strategic, and less routine than the programmer category that was hit hardest. (washingtonpost.com)
• Tech has seen large layoffs—400,000+ tech workers globally from 2022–2024 and more in 2025—but analyses attribute this mainly to pandemic over‑hiring, higher interest rates, and restructuring, with AI productivity a contributing factor but not the dominant, explicit reason for most cuts. (medium.com) A synthesis of McKinsey and Yale research summarized in 2025 notes that only about 1% of firms explicitly cite AI as the reason for layoffs. (cengizhan.com)

What companies say they plan to do with AI

• A 2024 ManpowerGroup survey of global employers found 55% expect AI to increase headcount, 24% expect no impact, and only 18% expect AI and ML to reduce staffing—evidence that most firms do not see AI primarily as a headcount‑cutting tool. (investor.manpowergroup.com)
• McKinsey’s 2025 State of AI data, broken down by function, indicates that software engineering/IT/product development functions are among those where companies most often expect headcount to increase, even as they adopt AI; only a minority foresee net reductions. (enterpriseaiexecutive.ai)

How AI coding tools are actually being used

• GitHub Copilot and similar tools have been adopted at scale (1.3M paid users across 50,000+ organizations by early 2023), with studies showing up to 55% faster completion on some coding tasks; these reports frame AI as a “pair‑programmer” that still requires human oversight, not as a replacement for entire dev teams. (wired.com)
• Inside Microsoft, engineering leaders report AI coding assistants contributing substantial portions of code and saving developers from “minutes to weeks” of work, yet they describe this as reducing “developer toil” and improving efficiency, not as a program of eliminating most developers. (theverge.com)
• The 2024 and 2025 Stack Overflow developer surveys show that around 76–84% of developers use or plan to use AI tools, and productivity/learning are the top cited benefits, while a large majority do not view AI as an immediate threat to their jobs; AI is seen primarily as a productivity booster. (linkedin.com)

Nuances and caveats

• There is growing evidence that entry‑level and routine coding roles are being squeezed: industry analyses note sharp drops in junior tech postings and a shift toward hiring more senior engineers who then use AI tools, plus significant declines in traditional “programmer” roles. (washingtonpost.com) Some commentators and case studies argue that leaders believe 5 AI‑augmented engineers can replace 10 traditional ones and have used this logic in certain reorganizations. (medium.com)
• However, these effects are uneven and concentrated in specific segments (junior roles, highly routine work, and certain big‑tech restructurings), not a universal pattern of companies “cutting all their developers because one or two can do ten times the work.” The majority of survey and labor‑market evidence still points to augmentation with relatively stable or growing developer demand, at least so far in the 2023–2025 window.

Given this mix, Sacks’s short‑term claim—that over the next few years, most firms would keep substantial developer headcount and use AI mainly to boost productivity and accelerate roadmaps rather than immediately slashing software teams en masse—aligns well with the best available evidence as of late 2025. The prediction is therefore right for the period we can currently observe, even though longer‑term impacts remain uncertain.

As of November 30, 2025, AI tools (e.g., GitHub Copilot, OpenAI’s Codex-based assistants, Replit’s code assistants, etc.) have significantly automated parts of software development—such as code completion, boilerplate generation, and some testing—but they have not eliminated the need for human software engineers or “most forms of coding.” Surveys and industry reports still show strong demand for software developers and ongoing human-led coding work across organizations, despite increased AI assistance. However, Chamath’s prediction explicitly referred to a multi‑year to multi‑decade horizon ("over time" / "coming years to couple of decades"), and only about 2.5 years have passed since the March 31, 2023 episode. Because the stated timeframe extends well beyond 2025, we cannot yet determine whether AI will eventually eliminate most forms of coding work. Therefore, the correct evaluation at this point is that it’s too early to judge the prediction as right or wrong.

As of late 2025, none of the cited white‑collar occupations has “wholesale just gone away” or is “done by AI” in the sense of being largely eliminated as human job categories.

• Accountants and auditors: The U.S. Bureau of Labor Statistics (BLS) reports about 1.58 million accountants and auditors employed in 2024, with employment projected to grow 5% from 2024–2034, and explicitly notes that AI will automate routine tasks but is not expected to reduce overall demand for accountants; instead, it shifts them toward more analytical and advisory work. (bls.gov) Other reporting describes a shortage of accountants, with firms offshoring work to India because they cannot hire enough CPAs in the U.S., not because AI has replaced them. (reuters.com)

• Illustrators and logo/graphic designers: BLS 2024 data show about 265,900 graphic designers (a category that includes logo work) and 52,000 craft and fine artists, including 26,500 fine artists such as painters and illustrators, with projections to 2034 that are roughly flat (0% change overall for craft and fine artists, –1% for fine artists including illustrators). (bls.gov) These professions are under pressure from generative image tools, and there is documented displacement in segments such as video‑game illustration and other creative markets, but the occupations clearly persist at scale. (en.wikipedia.org)

• Sales development representatives (SDRs): U.S. estimates show hundreds of thousands of SDRs employed (over 660,000 by one dataset) and a projected positive growth rate (~4% from 2018–2028). (zippia.com) Educational and career guides updated in 2025 still describe SDRs as in demand, with automation and AI tools changing how they work (copilots for lead scoring, outreach, etc.), not eliminating the role. (coursera.org)

• Overall AI–employment impact so far: Recent empirical work finds that generative AI has not yet caused large, broad‑based job loss in the U.S.; a Yale–Brookings study notes that AI adoption has not reshaped the labor market faster than previous technologies, and evidence of AI‑driven layoffs is limited. (ft.com) Another study (Stanford/World Bank/Clemson) finds AI‑exposed jobs are seeing wage gains without significant job losses, with only about 1% of services firms reporting any AI‑related layoffs. (barrons.com) Other research does observe meaningful displacement for some younger or entry‑level workers in AI‑exposed sectors, but that is still far from the “wholesale” disappearance of entire job categories. (businessinsider.com)

In other words, by roughly 2½ years after March 2023, illustrators, logo designers/graphic designers, accountants, and SDRs remain large, recognized human occupations with substantial headcounts and ongoing hiring. AI has started to automate tasks and reduce demand in some niches, but it has not caused these professions to largely vanish as distinct human job functions within this short timeframe. Therefore, the prediction is wrong given the evidence available by November 2025.

The prediction covers 10–20 years from 2023, so as of November 2025 we are only ~2.5 years into the window. That’s not enough time to determine whether AI will avoid causing major long‑term displacement of software developers.

Evidence so far is mixed but broadly consistent with Jason’s view in the short run:

• U.S. "computer programmer" roles (a narrower, more routine coding category) have dropped by more than 25% over two years, with generative AI cited as one contributing factor, though macroeconomics and offshoring also matter. However, the same reporting notes that software developer roles have remained comparatively stable and are more insulated. (washingtonpost.com)
• BLS-based analyses project employment in the broader professional, scientific and technical services sector (including software development and related tech jobs) to grow about 10.5% from 2023–2033, and find no clear evidence that AI will cause widespread net job losses in this sector. (investopedia.com)
• Current BLS projections still show positive, above-average growth for software developers/engineers (roughly 15–17% between 2024 and 2034), even as the separate "computer programmer" category is projected to decline. (en.wikipedia.org)
• Multiple recent analyses argue that AI is reshaping and automating tasks rather than eliminating most developer roles, with AI expected to change what developers do (more architecture, oversight, and AI integration) rather than erase demand for human software engineers. (netcorpsoftwaredevelopment.com)

Overall, AI has begun to reduce certain lower-level coding roles, but there is no wholesale elimination of software development jobs yet, and medium‑term projections still show rising demand. Because the prediction is about conditions over the entire next 10–20 years, it cannot yet be definitively judged as right or wrong, so the status is inconclusive (too early).

The prediction is about an eventual state of consumer technology (“in some period of time”) with no specific deadline. To judge it as right or wrong, we would need either:

1. Clear evidence that, by now, ordinary consumers routinely have fully personalized movies or video games generated on the fly by AI, tailored to each individual; or
2. A passed deadline after which the prediction can be said to have failed.

As of November 30, 2025, neither condition is met:

• There are emerging tools and research demos for AI‑generated video and interactive experiences (e.g., model-based video generation and experimental AI-driven game content), but these are far from a mainstream, routine consumer experience where “we all have our own movie or video game” automatically generated for us as a normal part of entertainment consumption.
• Major consumer platforms (e.g., Netflix, gaming consoles, app stores) still predominantly distribute pre-authored movies and games. While recommendation algorithms personalize which content you see, they do not yet generate full-length, bespoke movies or full games from scratch for each user as a standard product offering.
• Because the predictor did not specify a timeframe (no year or bound like “within 5–10 years”), the fact that this outcome has not yet occurred by 2025 does not logically falsify the prediction; it could still happen later.

Given that (a) the predicted scenario has not yet materialized at scale, but (b) no time limit was given, the correct classification as of now is that it’s too early to determine whether the prediction will ultimately be right or wrong.