Pundit ClaimsAI claims by Gary Marcus and Ed Zitron, with the evidence

Ed Zitron Where's Your Ed At Published

How To Argue With An AI Booster

We recorded 35 claims from this post. 24 were checked against evidence. 11 can’t be checked. 18 more are held. Read the post at its source.

Claims from this post (35)

In the order they appear in the post. Each quotation is the speaker’s words in this post. A label such as “Supported” or “Can’t be checked” is the claim’s current assessment; its page has the evidence and the reasons.

  1. Assessment: UnresolvedSaid

    “This isn't a case where "the enterprise" is suddenly going to save these companies, because the enterprise already tried, and it isn't working.”

    What we tested: Enterprise demand will not rescue generative AI companies, because enterprises have already tried the technology and it is not working.

  2. Assessment: Can’t be checkedSaid

    “The answer, of course, is that it has no intellect, is not intelligent, and Large Language Models are being pushed up a mountain by a cadre of people who are either easily impressed or invested — either emotionally or financially — in its success due to the company they keep or their intentions for the world.”

    What we tested: LLMs have no intellect and are not intelligent; they are promoted by people who are easily impressed or have an emotional or financial stake in them.

  3. Assessment: UnresolvedSaid

    “Long story short, agents are not autonomous, they do not replace jobs, they cannot “replace coders,” they are not going to do so because probabilistic models are a horrible means of taking precise actions, and almost anyone who brings up agents as a booster is either misinformed or in the business of misinformation.”

    What we tested: AI agents are not autonomous, do not replace jobs or coders, and will not do so, because probabilistic models are poor at taking precise actions.

  4. Assessment: Can’t be checkedSaid

    “Big tech has spent over $500 billion in capital expenditures in the last 18 months, and all told — between investments of cloud credits and infrastructure — will likely sink over $600 billion by year's-end.”

    What we tested: Big tech, having spent over $500 billion in capital expenditures in the last 18 months, will likely sink over $600 billion all told, counting investments of cloud credits and infrastructure, by year's end.

  5. Assessment: Can’t be checkedSaid

    “We Are Not In The Early Days Of Generative AI, And Anyone Using This Argument Is Either Ignorant Or Intentionally Deceptive”

    What we tested: Generative AI is not in its early days.

  6. Assessment: ContradictedSaid

    “In simpler terms, these companies are flush with cash, have all the attention and investment they could possibly need, and are still unable to create a product with a defined, meaningful, mass-market use case.”

    What we tested: Generative AI companies, despite ample funding and attention, have not produced a product with a defined, meaningful mass-market use case.

    What happened: The abstract reports adoption by around 10% of the world's adults by July 2025 and identifies practical guidance, information seeking, and writing as nearly 80% of conversations. Together, those findings establish broad adoption with identifiable practical uses under the stated reading. (National Bureau of Economic Research, )

  7. Assessment: Can’t be checkedSaid

    “In fact, I believe that thanks to effectively infinite resources, we've speed-run the entire Large Language Model era, and we're nearing the end.”

    What we tested: Because of effectively unlimited resources, the Large Language Model era has been sped through and is nearing its end.

  8. Assessment: UnresolvedSaid

    “The "fiber optic cable" of this era is mostly owned by a few companies. 42% of NVIDIA's revenue is from the magnificent 7, and the companies buying these GPUs are, for the most part, not going to go bust once the AI bubble bursts.”

    What we tested: For the most part, the companies buying AI GPUs are not going to go bust once the AI bubble bursts.

  9. Assessment: UnresolvedSaid

    “Unlike CPUs, a GPU can't easily change tasks, or handle many little distinct operations, meaning that these things aren't going to be adopted for another mass-scale use case.”

    What we tested: AI GPUs will not be adopted for another mass-scale use case, because GPUs cannot easily switch tasks.

  10. Assessment: UnresolvedSaid

    “There is no proof this is possible, nobody has done it, nobody will do it.”

    What we tested: Nobody will build the self-learning, self-teaching AI agent that AI 2027 assumes.

  11. Assessment: Can’t be checkedSaid

    “Anyway, AI 2027 is fan fiction, nothing more, and just because it’s full of fancy words and has five different grifters on its byline doesn’t mean anything.”

    What we tested: The AI 2027 scenario is fiction with no forecasting value.

  12. Assessment: ContradictedSaid

    “Reasoning models are also the only way that model developers have been able to improve the efficacy of new models, using something called "test-time compute" to burn extra tokens to complete a task.”

    What we tested: Model developers have only managed to make new models more effective through reasoning models that burn extra tokens via test-time compute.

    What happened: OpenAI explicitly describes GPT-4.5 as not thinking before responding and attributes gains to unsupervised scaling. Its reported comparisons include GPQA scores of 71.4% versus 53.6% for GPT-4o, providing a counterexample to the claim that additional inference-time reasoning was the only route to improvement. (OpenAI, )

  13. Assessment: UnresolvedSaid

    “The cost of inference has gone up. Statements otherwise are purely false, and are the opinion of somebody who does not know what he's talking about.”

    What we tested: The real cost of AI inference has risen, and claims that it is falling are false.

  14. Assessment: ContradictedSaid

    “...maybe? It sure isn't trending that way, nor has it gone down yet.”

    What we tested: The cost of inference is not trending downward and has not gone down yet, though he says it maybe could go down.

    What happened: Epoch checked evaluation costs using recorded token consumption and found historical declines at comparable benchmark performance levels. This addresses Zitron's objection to token-price-only comparisons and contradicts the assertion that inference costs had not fallen. (Epoch AI, )

  15. Assessment: UnresolvedSaid

    “And models keep getting bigger and more expensive, too.”

    What we tested: AI models keep getting larger and more expensive.

  16. Assessment: ContradictedSaid

    “Because model developers hit a wall of diminishing returns, and the only way to make their models do more was to make them burn more tokens to generate a more accurate response (this is a very simple way of describing reasoning, a thing that OpenAI launched in September 2024 and others followed).”

    What we tested: Inference costs rose because model developers hit diminishing returns and could only improve models by having them burn more tokens through reasoning.

    What happened: The report documents gains in the pretrained base model under a common evaluation pipeline, separately from its later reasoning post-training. Its architecture, data, and training improvements supply another route to capability gains, contradicting the claim of exclusivity. (Qwen Team / arXiv, )

  17. Assessment: ContradictedSaid

    “In any case, there really is no changing this path. They are out of ideas.”

    What we tested: AI model developers are out of ideas and will not escape the path of improving models by burning more tokens.

    What happened: Artificial Analysis reports reasoning-mode Intelligence Index scores of 54 for Flash and 48 for Flash-Lite, improvements of 3 and 8 points over their predecessors. The same evaluation reports lower output-token consumption, providing a counterexample to the asserted dependence on increasing tokens. (Artificial Analysis, )

  18. Assessment: UnresolvedSaid

    “Every member of the government having access to ChatGPT — something that is not even necessarily the case! — does not make this software useful, let alone essential, and if OpenAI burns a bunch of money "making it work for them," it still won't be essential, because Large Language Models are not actually that useful for doing stuff!”

    What we tested: Even if OpenAI spends heavily tailoring ChatGPT for government use, it will not become essential, because LLMs are not very useful.

  19. Assessment: UnresolvedSaid

    “These data centers are not, in and of themselves, driving much economic growth other than in the costs of building them.”

    What we tested: Apart from construction spending, AI data centers do not drive much economic growth.

  20. Assessment: UnresolvedSaid

    “These data centers, once built, also create very little economic activity. They don't create jobs, they take up massive amounts of land and utilities, and they piss off and poison their neighbors.”

    What we tested: Once built, AI data centers create very little economic activity, do not create jobs, take up massive amounts of land and utilities, and piss off and poison their neighbors.

  21. Assessment: UnresolvedSaid

    “Based on my (conservative) calculations, the true "cost" of OpenAI is around $82 billion, and that only includes capex from 2024 onward, based on 30% of Microsoft's capex (as not everything has been invested yet in 2025, and OpenAI is not necessarily all of the capex) and the $41.4 billion of funding it’s received so far.”

    What we tested: Attributing 30% of Microsoft's capex since 2024 and adding OpenAI's $41.4B in funding, OpenAI's true cost is about $82 billion.

  22. Assessment: UnresolvedSaid

    “The true cost of Anthropic is around $77.1 billion, including all its funding and 30% of Amazon's capex from the beginning of 2024.”

    What we tested: Counting all its funding plus 30% of Amazon's capex since 2024, Anthropic's true cost is about $77.1 billion.

  23. Assessment: Can’t be checkedSaid

    “There's a good chance that Anthropic and OpenAI dominate the capex of Amazon and Microsoft, in part because what the fuck else are they buying all these GPUs for, as their own AI services don't seem to be making that much money at all.”

    What we tested: There is a good chance that Anthropic and OpenAI dominate the capital expenditures of Amazon and Microsoft.

  24. Assessment: Can’t be checkedSaid

    “We don’t know if AGI is possible, anyone claiming they do is lying.”

    What we tested: Nobody knows whether AGI is possible, and anyone who claims to know is lying.

  25. Assessment: SupportedSaid

    “It is transparently, blatantly obvious that GPT-4 did not "hire" a Taskrabbit or, indeed, make any of these actions — it was prompted to, and they do not show the prompts they used, likely because they had to use so many of them.”

    What we tested: GPT-4 did not autonomously hire a TaskRabbit worker or take the actions described; it was prompted and humans carried them out.

    What happened: The report describes a researcher executing browser commands and replacing a failed reasoning step with a hint. It also documents the assigned hiring task while redacting exact prompt wording, supporting the absence of autonomous execution. (Alignment Research Center Evals, )

  26. Assessment: Can’t be checkedSaid

    “Anyone falling for this is a mark, and OpenAI should have gone out of its way to correct people.”

    What we tested: OpenAI should have actively corrected misreported claims that GPT-4 hired a TaskRabbit worker.

  27. Assessment: Can’t be checkedSaid

    “As I've said in the past, OpenAI is deliberately using Weekly Active Users so that it doesn't have to publish its monthly active users, which I believe would be higher.”

    What we tested: OpenAI is deliberately using weekly active users so that it doesn't have to publish its monthly active users, which he believes would be higher.

  28. Assessment: UnresolvedSaid

    “OpenAI's growth as the single-most-prominent company in AI (and if we're honest, one of the most prominent in software writ large) makes sense, but at some point will slow, because the actual returns for the businesses aren't there.”

    What we tested: OpenAI's growth will eventually slow because businesses are not getting real returns from AI.

  29. Assessment: UnresolvedSaid

    “ChatGPT is basically the same product as Claude or Grok or any number of different LLMs.”

    What we tested: ChatGPT is essentially the same product as Claude, Grok and other LLMs.

  30. Assessment: UnresolvedSaid

    “I don't buy that this company will continue growing forever, and its stinky conversion rate isn't going to change anytime soon.”

    What we tested: He doesn't believe OpenAI will continue growing forever, and OpenAI's poor conversion rate isn't going to change anytime soon.

  31. Assessment: SupportedSaid

    “While young people are having trouble finding jobs, there is no proof that AI is the reason.”

    What we tested: There is no proof that AI is why young people are struggling to find jobs.

    What happened: The event is publication of a contemporaneous research assessment, not the date of the underlying unemployment observations. Its young-worker analysis and discussion of confounding support the narrower conclusion that exposure comparisons had not established causation. (Economic Innovation Group, )

  32. Assessment: UnresolvedSaid

    “QUIP: AI Will Do Science Research, Somehow — No it won't, here's a writeup about why foundation models can't do this.”

    What we tested: AI foundation models will not do scientific research.

  33. Assessment: Can’t be checkedSaid

    “AI is being sold as a transformational technology, and I am yet to see it transform anything.”

    What we tested: Although marketed as transformational, generative AI has not transformed anything he has seen.

  34. Assessment: UnresolvedSaid

    “You won't be able to find this as it isn't possible.”

    What we tested: Someone who cannot code cannot build a fully functional, secure, useful app entirely by vibe coding.

  35. Assessment: Can’t be checkedSaid

    “As things begin to collapse — and they sure look like they're collapsing, but I am not making any wild claims about "the bubble bursting" quite yet — it will look increasingly more deranged to bluntly publish everything that these companies say.”

    What we tested: Things in the AI industry sure look like they're beginning to collapse, though he is not making claims about 'the bubble bursting' quite yet.

Held from this post (18)

These records did not pass every check, so nothing about them is published. Held is not a finding.