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

Claims checked against evidence, page 10 of 31

Claims by Gary Marcus and Ed Zitron. Deadlines are judged as of 2026-10-29.

Claims checked against evidence (3100)

Each of these claims was checked against dated evidence.

Page 10 of 31: claims 901 to 1000 of 3100, ordered by the date first said, newest first. To find a claim by its words, use search.

  1. Assessment: ContradictedSaid

    Ed Zitron

    “Claude Code does not actually build unique software.”

    What happened: The report documents Claude Code producing machine-checked verification code for an ANF transformation whose corresponding proof had previously been left as future work. It adapted an existing CPS proof under expert guidance, adding a distinct verified property rather than merely reproducing the template; February 23 dates the public artifact report, not an asserted exact completion day. (arXiv — Zoe Paraskevopoulou, )

  2. Assessment: ContradictedSaid

    Ed Zitron

    “Every example you've read about somebody being amazed by it has built a toy app or website that's very similar to many open source projects or website templates that Anthropic trained its training data on.”

    What happened: Carlini's public account identifies Claude Code sessions, provides source code, and expresses surprise at the achievement. The reported compiler functionality goes substantially beyond a small app or website, although the account also discloses dependencies, defects and extensive human work on the testing environment. (Anthropic, )

  3. Assessment: UnresolvedSaid

    Gary Marcus

    “Generative AI is NOT remotely reliable enough to make life or death decisions of magnitude.”

  4. Assessment: SupportedSaid

    Gary Marcus

    “Although there are some practical use cases, such as coding, it is an inherently unreliable technology.”

    What happened: Developer respondents report useful applications in their work, supporting the modest assertion that practical coding uses exist. Self-selection and subjective productivity estimates limit the strength of this evidence. (Stack Overflow, )

  5. Assessment: SupportedSaid

    Gary Marcus

    “It is leading some people into serious mental health issues.”

    What happened: Clinicians documented hospitalization and concluded from chatlogs that the chatbot facilitated or mediated delusion formation. The same report identifies mood history, stimulant use, sleep deprivation, and a later relapse with little chatbot encouragement as important competing explanations. (Innovations in Clinical Neuroscience, )

  6. Assessment: UnresolvedSaid

    Gary Marcus

    “Turns out it wasn’t, as Shira Ovide just reported:”

  7. Assessment: UnresolvedSaid

    Gary Marcus

    “I don’t know that Generative AI was literally a scam, but the people selling it have tried to sell it as if it were tantamount to artificial general intelligence, when it’s not.”

  8. Assessment: UnresolvedSaid

    Gary Marcus

    “Generative AI has its uses, but it hardly surprising that most business have struggled to find return on their investment, given the massive reliability issues that continue to plaguage it.”

  9. Assessment: SupportedSaid

    Gary Marcus

    “They often produce workslop.”

    What happened: The survey of 1,100 US AI users at larger employers also reports weekly cleanup and negative consequences from low-quality outputs. Together these support recurring workslop, although revision alone does not necessarily indicate poor quality. (Zapier, distributed by GlobeNewswire, )

  10. Assessment: UnresolvedSaid

    Gary Marcus

    “Across domains involving causality, value trade-offs, justification, and error monitoring, humans attend to reasons, stakes, and responsibility, whereas language models optimize for linguistic plausibility—structural fault lines between performance alignment and genuine evaluative competence (Quattrociocchi et al., 2025).”

  11. Assessment: UnresolvedSaid

    Gary Marcus

    “Behavioral similarity alone therefore provides no insight into whether the underlying processes support core components of general intelligence, such as judgment, error correction, or reliable generalization.”

  12. Assessment: SupportedSaid

    Gary Marcus

    “However, benchmark success is a limited indicator of general intelligence.”

    What happened: Strong GSM8K scores did not hold up under trivial changes. So the benchmark score overstated robust reasoning. (arXiv / ICLR 2025 (Mirzadeh et al., Apple), )

  13. Assessment: SupportedSaid

    Gary Marcus

    “Moreover, benchmarks are often easily gamed.”

    What happened: Many prominent benchmarks had flaws that allow inflated scores. One even credits empty responses, which shows gaming can be easy. (arXiv (Zhu et al., incl. Stanford, Berkeley, UK AISI researchers), )

  14. Assessment: SupportedSaid

    Gary Marcus

    “Reports that language models have produced correct proofs for isolated open problems in mathematics, including specific Erdős problems, do not alter this assessment.”

    What happened: Tao described the AI successes as falling on obscure problems that often have straightforward solutions. This matches the post's point that these results do not show deep, general problem-solving. (TechCrunch, )

  15. Assessment: SupportedSaid

    Gary Marcus

    “Current AI systems are powerful and increasingly useful tools, but they do not exhibit the flexible, self-directed competence that the original concept of artificial general intelligence was intended to capture.”

    What happened: Frontier systems scored under 1% where humans scored 100% on tasks requiring self-directed exploration and goal discovery. This is direct evidence of missing flexible, self-directed competence. (Decrypt, )

  16. Assessment: UnresolvedSaid

    Gary Marcus

    “We also show that recent claims (Chen et al., 2026) about putative success on AGI hinge on redefining what AGI has historically meant.”

  17. Assessment: UnresolvedSaid

    Gary Marcus

    “Language models, by contrast, tended to issue confident classifications in the same cases, even when uncertainty was explicitly signaled in the input.”

  18. Assessment: UnresolvedSaid

    Gary Marcus

    “As a result, model development and data curation are optimized for benchmark success, producing systems that perform well under test conditions but degrade in real-world settings that differ only modestly from them”

  19. Assessment: UnresolvedSaid

    Gary Marcus

    “In recent years, as the financial stakes of benchmark performance have increased, systems have been trained more often directly on benchmarks (“teaching to the test”) or on synthetic data closely resembling them.”

  20. Assessment: UnresolvedSaid

    Gary Marcus

    “At the task level, analyses of workplace automation show that even state-of-the-art systems can reliably perform only a small fraction of the tasks required in typical human occupations (Eloundou et al., 2023; Felten et al., 2023), despite high scores on idealized benchmarks.”

  21. Assessment: SupportedSaid

    Gary Marcus

    “Task-level performance, even when impressive, is not sufficient evidence of general intelligence.”

    What happened: The paper explains how prior knowledge and training can produce task skill without establishing the ability to acquire new skills. This supports the ordinary distinction in the claim, within an explicitly stated measurement framework. (arXiv, )

  22. Assessment: UnresolvedSaid

    Gary Marcus

    “Such claims confuse high performance on benchmarks (often gameable) with real world flexibility, which historically has been at the center of what it means to be an artificial general intelligence.”

  23. Assessment: UnresolvedSaid

    Gary Marcus

    “Rumors that humanity has already achieved artificial general intelligence (AGI) have been greatly exaggerated.”

  24. Assessment: SupportedSaid

    Gary Marcus

    “By the standards articulated in the original definitions of artificial general intelligence—robustness across environments, reliable generalization under novelty, and autonomous goal-directed behavior—current AI systems remain limited.”

    What happened: On a benchmark built to test generalization in novel environments, frontier systems from the period (including Claude Opus 4.6, released before the post) scored under 1%. Humans solved every environment. This is a large shortfall on the speaker's named standards. (Decrypt, )

  25. Assessment: UnresolvedSaid

    Gary Marcus

    “Scammers will be among the first to adopt these tools. And indeed they already have;”

  26. Assessment: SupportedSaid

    Gary Marcus

    “1Some state laws are trying to address these issues, but as we know the Federal government is trying to undermine the states’ abilities to pass and enforce AI regulation.”

    What happened: The order directs legal challenges and preparation of federal preemption legislation, and asks agencies to assess conditioning grants on states not enacting or enforcing conflicting AI laws. Section 8 preserves specified categories in the proposed legislation, so the effort is substantial but not universal. (The White House, )

  27. Assessment: SupportedSaid

    Gary Marcus

    “Well here’s a new one, you can have OpenClaw call people, and pretend to be human.”

    What happened: The released documentation provides outbound telephony, speech providers and message arguments. It establishes the calling component of the claim before publication. (OpenClaw project, )

  28. Assessment: UnresolvedSaid

    Gary Marcus

    “Anybody’s appearance can (given sufficient data) be faked now, at any time, for almost nothing.”

  29. Assessment: UnresolvedSaid

    Gary Marcus

    “OpenAI is losing money every quarter; even by their own (optimistic) calculations, they won’t make a profit for years.”

  30. Assessment: SupportedSaid

    Gary Marcus

    “Google and Anthropic have each largely caught up; various Chinese companies are closing in.”

    What happened: The original APEX-Agents evaluation places a Google model ahead of OpenAI on professional, multi-application tasks. This supports competitive capability in an important domain, although it is not a universal ranking. (arXiv; Vidgen and colleagues, )

  31. Assessment: UnresolvedSaid

    Gary Marcus

    “If the venture capital dries up, they are toast. Talent will make a beeline for competitors.”

  32. Assessment: SupportedSaid

    Gary Marcus

    “I have been writing for a long time about how OpenAI snarfs the intellectual property of artists and writers.”

    What happened: The court records that training on nine disputed song lyrics was undisputed. The following sentence dates their reproduction by the chatbot to 2024, establishing use by that year without supplying the exact training date. (Munich I Regional Court / Bavarian Ministry of Justice, )

  33. Assessment: UnresolvedSaid

    Gary Marcus

    “Or even how he eventually walked that back to 2026 (which also won’t happen).”

  34. Assessment: SupportedSaid

    Gary Marcus

    “Back on planet earth, according to December’s Remote Labor Index, only 2.5% of human “online tasks” (physical tasks were excluded) could be done by AI.”

    What happened: This dated December account reports the claimed top rate for RLI projects. The event date identifies December reporting of standing results, not new experiments; an institutional summary is used because a precisely dated December primary snapshot was not verified. (United Nations University Campus Computing Centre, )

  35. Assessment: SupportedSaid

    Gary Marcus

    “The security of the autogenerated code is very much in question, as for example this article last week noted (precisely as this substack has long warned):”

    What happened: A large test across more than 100 LLMs found vulnerabilities in 45% of cases. That is a substantial documented security problem in AI-generated code. (Veracode, )

  36. Assessment: SupportedSaid

    Gary Marcus

    “The bottom line is this: LLMs are certainly coding more, but it’s not clear that the code they are creating is secure or trustworthy.”

    What happened: A peer-reviewed measurement shows the share of AI-assisted code rising sharply. This supports 'LLMs are certainly coding more'. (Complexity Science Hub via EurekAlert! (study in Science), )

  37. Assessment: SupportedSaid

    Gary Marcus

    “LLMs write code wicked fast, but some coders are starting to report burnout, and only relatively modest gains relative to that burnout, as a new story from Connie Loizos at Tech Crunch illustrates”

    What happened: The Loizos story exists and reports an engineer at an AI-embracing tech company describing more work, not less. (TechCrunch, )

  38. Assessment: SupportedSaid

    Gary Marcus

    “To his credit, Shumer is actually correct that something has changed recently. You really can let things rip more in the most recent systems.”

    What happened: METR measured a record jump in autonomous task length for a late-2025 model. That shows recent systems can be left to run on longer tasks. (LessWrong (quoting METR on X), )

  39. Assessment: SupportedSaid

    Gary Marcus

    “LLMs still hallucinate and continue to make boneheaded errors.”

    What happened: An independent benchmark in November 2025 found that major frontier reasoning models hallucinated in more than 10% of summaries. (Vectara, )

  40. Assessment: SupportedSaid

    Gary Marcus

    “And reasoning is still one of the core issues.”

    What happened: A comprehensive survey from February 2026 concludes that significant reasoning failures persist in LLMs. (arXiv (Song, Han, Goodman), )

  41. Assessment: UnresolvedSaid

    Gary Marcus

    “Although there is some clear value in it, the costs are high, and relative to those costs, the revenue has been modest.”

  42. Assessment: UnresolvedSaid

    Gary Marcus

    “Let me add that I do think that some form of AI will eventually actually yield the returns people were anticipating for Generative AI.”

  43. Assessment: UnresolvedSaid

    Gary Marcus

    “Small wonder that they are now turning to ads.”

  44. Assessment: SupportedSaid

    Gary Marcus

    “But at the end of the day, it’s a bit of Coke vs Pepsi: two slightly different products (with their own sets of loyal fans) fighting tooth and nail for more or less the same market.”

    What happened: The announcement markets coding alongside research, analysis, and document work. These functions overlap with OpenAI's contemporaneous professional assistant offering. (Anthropic, )

  45. Assessment: UnresolvedSaid

    Gary Marcus

    “Nvidia’s rise was predicated on the wrong-headed idea (long critiqued in this newsletter) that scaling would bring us to AGI.”

  46. Assessment: SupportedSaid

    Gary Marcus

    “Not an all out crash, but an unmistakable sign that its ascent has run out of gas.”

    What happened: Nvidia closed at a record on Thursday, August 7, 2025, at about $180.7, implied from the Friday price and gain. This matches Marcus's 'about 181'. (Barron's (Dow Jones) via Tiger Brokers, )

  47. Assessment: SupportedSaid

    Gary Marcus

    “They have also faced a lot of backlash for being reckless and irresponsible; hardly a good brand name.”

    What happened: An official government statement condemns Grok's handling of intimate deepfake images and calls for regulatory action. It directly establishes prominent backlash over alleged irresponsible conduct. (UK Department for Science, Innovation and Technology, )

  48. Assessment: UnresolvedSaid

    Gary Marcus

    “Me, I think the merger is really a kind of bailout, to give a lot of cash to a company that is otherwise in distress.”

  49. Assessment: SupportedSaid

    Gary Marcus

    “It has already been attacked, as researcher Michael Riegler noted yesterday on LinkedIn:”

    What happened: The January 31 report covers January 28–31 and reproduces hidden instructions and API-action requests aimed at agents reading platform content. This supports the occurrence of attack attempts before the post; the event date marks the report's collection endpoint. (Michael A. Riegler and Sushant Gautam, Simula and SimulaMet; Zenodo, )

  50. Assessment: SupportedSaid

    Gary Marcus

    “The catch is that agents like OpenClaw are built on a foundation of LLMs, and as we well know, LLMs hallucinate and make all kinds of hard to predict and sometimes hard to detect errors.”

    What happened: The versioned documentation describes primary and fallback models and provides Claude model configurations. It establishes the language-model foundation of the agent. (OpenClaw project, )

  51. Assessment: SupportedSaid

    Gary Marcus

    “OpenClaw itself is basically a cascade of LLMs.”

    What happened: The versioned documentation identifies an LLM-and-tool loop as the core execution mechanism. This supports the ordinary shorthand reading of cascade, without establishing a mandatory pipeline of distinct models. (OpenClaw project, )

  52. Assessment: SupportedSaid

    Gary Marcus

    “OpenClaw inherits all these weaknesses.”

    What happened: Cisco describes invoking a third-party skill against OpenClaw, finding silent data exfiltration and instructions that bypassed the assistant's safety guidelines. This is direct evidence of the specified instruction-to-action weakness in the tested setup. (Cisco AI Threat and Security Research, )

  53. Assessment: SupportedSaid

    Gary Marcus

    “AutoGPT had a tendency to report that it had completed tasks that it hadn’t really, and we can expect OpenClaw to do the same.”

    What happened: The April 16 case reports OpenClaw claiming a CRM note was posted, while its tool ledger showed no execution and a direct CRM query showed no write. The report specifies OpenClaw 2026.4.14 and the model used. (GitHub user apoorv12, OpenClaw issue tracker, )

  54. Assessment: UnresolvedSaid

    Gary Marcus

    “The point is that when it costs pennies to coordinate a fake mob and moments to counterfeit a human identity, the public square is left wide open to attack.”

  55. Assessment: SupportedSaid

    Gary Marcus

    “Today’s swarms, now emerging, can coordinate fleets of synthetic personas—sometimes with persistent identities—and move in ways that are hard to distinguish from real communities.”

    What happened: One operator used an LLM to run a fleet of 100+ persona accounts that engaged real users. The same report lists 'creating and maintaining consistent personas across platforms', which matches the persistent-identity claim. (Anthropic, )

  56. Assessment: UnresolvedSaid

    Gary Marcus

    “Even so-called “thinking” AI models are vulnerable to this,”

  57. Assessment: SupportedSaid

    Gary Marcus

    “Venture capital is already helping industrialize astroturfing: Doublespeed, backed by Andreessen Horowitz, advertises a way to “orchestrate actions on thousands of social accounts” and to mimic “natural user interaction” on physical devices so the activity appears human.”

    What happened: Establishes the named investor relationship before publication. The amount is incidental; this assessment does not grade a funding-amount claim. (The Observer, )

  58. Assessment: UnresolvedSaid

    Gary Marcus

    “If we require this kind of “proof-of-human” for high-reach interactions, we make it mathematically difficult and financially ruinous for one operator to secretly run ten thousand accounts.”

  59. Assessment: SupportedSaid

    Gary Marcus

    “Just a few months later, that dark prediction came to pass, as discussed here in April of 2023. (Causality was indeed hard to prove but the facts did not reflect well on the bot).”

    What happened: Shows that by March 2023 a suicide was publicly linked to a chatbot, and that the reported logs did not reflect well on the bot. (Vice (Motherboard), )

  60. Assessment: SupportedSaid

    Gary Marcus

    “And it’s just getting worse.”

    What happened: In one day in November 2025, four new suicide cases went to court, on top of the Raine suit from August. That points to a sharp rise in reported deaths and lawsuits in late 2025. (Associated Press (via FOX19), )

  61. Assessment: UnresolvedSaid

    Gary Marcus

    “One of my darkest yet on-target predictions in WIRED in December 2022, was that we would see the first chatbot-related death in 2023, perhaps by suicide.”

  62. Assessment: SupportedSaid

    Gary Marcus

    “And then there is the growing problem of chatbots apparently inducing delusions in humans, that Kashmir Hill and others have been documenting.”

    What happened: A study of records from about 54,000 psychiatric patients, covering up to June 2025, found a clear rise over time in documented harmful chatbot use. Worsened delusions were the main harm. (Aarhus University, )

  63. Assessment: SupportedSaid

    Gary Marcus

    “By now, there have been several reports of chatbot-associated deaths (both suicide and murder), enough that wikipedia has started keeping track.”

    What happened: The dated revision directly establishes Wikipedia's coverage before the post. It collects numerous reported deaths, including Setzer, Raine, Lacey, and the Greenwich murder-suicide; the event date here describes the coverage, not Lacey's death. (Wikipedia contributors, )

  64. Assessment: SupportedSaid

    Gary Marcus

    “That deal was OpenAI’s to lose. And they lost it.”

    What happened: The joint statement identifies Google's Gemini models and cloud technology as the basis of the next Apple Foundation Models, including personalized Siri. (Google and Apple, )

  65. Assessment: ContradictedSaid

    Gary Marcus

    “LLMs are commodities; DeepSeek proved that.”

    What happened: The study's January-to-April OpenRouter data show that new models do not uniformly displace competitors and that applications use different model mixes. This cuts against broad interchangeability, although the study is descriptive and covers one marketplace. (Andrey Fradkin, )

  66. Assessment: UnresolvedSaid

    Gary Marcus

    “OpenAI has never made a profit.”

  67. Assessment: SupportedSaid

    Gary Marcus

    “Many, many people were disappointed with GPT-5, arguably among the most overhyped products in history.”

    What happened: The report documents substantial launch-period dissatisfaction. Comment totals are not a count of distinct disappointed people, but the scale and described complaints support more than isolated disappointment. (Ars Technica, )

  68. Assessment: SupportedSaid

    Gary Marcus

    “I warned that LLMs might plateau within a year or two, and they have, that OpenAI would be mired in a lot of litigation, and they are.”

    What happened: The judicial order identifies separate news-publisher and class-action cases involved in ongoing litigation against OpenAI. Consolidating proceedings does not eliminate the underlying cases. (U.S. District Court, Southern District of New York; copy hosted by ChatGPT Is Eating the World, )

  69. Assessment: UnresolvedSaid

    Gary Marcus

    “Anthropic has started to eat a lot of OpenAI’s lunch, taking a lot of corporate customers away.”

  70. Assessment: UnresolvedSaid

    Gary Marcus

    “Over the last couple years, Altman’s public credibility has steadily waned (see e.g., Karen Hao’s bestseller, and my earlier essay on Altman in The Guardian).”

  71. Assessment: SupportedSaid

    Gary Marcus

    “Most importantly, other companies like Anthropic and Google caught up; xAI and various Chinese companies aren’t far behind.”

    What happened: The accompanying comparison gives Claude Opus 4.5 and GPT-5.1 scores of 70, Gemini 3 Pro 73, Grok 4 65, and Kimi K2 Thinking 67 on the same index. This supports the leading-tier comparison and xAI's competitive position. (Artificial Analysis, )

  72. Assessment: UnresolvedSaid

    Gary Marcus

    “Those price wars did in fact come to pass; users costs per token are roughly a hundred times lower two year later.”

  73. Assessment: ContradictedSaid

    Gary Marcus

    “His defensive, empty response to investor Brad Gerstner in which he couldn’t begin to explain how he would make financial ends meet was atrocious and utterly unconvincing.”

    What happened: The report quotes Altman connecting future revenue to continued ChatGPT growth, an AI cloud business, consumer devices and scientific applications. That is a high-level account of the intended funding mechanism, although it does not reconcile future cash receipts with commitments. (Fortune, republished by Yahoo Finance, )

  74. Assessment: SupportedSaid

    Gary Marcus

    “I also warned that LLMs would continue to have a truthiness problem and they have.”

    What happened: OpenAI explicitly acknowledges persistent hallucinations in GPT-5. The same sentence reports improvement, which is compatible with Marcus's claim that the problem continued. (OpenAI, )

  75. Assessment: UnresolvedSaid

    Gary Marcus

    “(Altman got famous hinting at the supposed miracles of GPT-5; by any reasonable measure it was both late and short of expectations.)”

  76. Assessment: UnresolvedSaid

    Gary Marcus

    “DeepSeek launched a price war that forced Altman to cut prices dramatically.”

  77. Assessment: SupportedSaid

    Gary Marcus

    “Update: This is consistent with the finding of the Remote Labor Index that AI could only do about 2.5% of jobs, reported recently by the Washington Post.”

    What happened: The original report states exactly the cited leading automation rate for its real-world project benchmark. It measures successful end-to-end project completion, not the percentage of all jobs that AI can perform or the value of human–AI assistance. (Center for AI Safety and Scale AI / arXiv, )

  78. Assessment: UnresolvedSaid

    Gary Marcus

    “They still aren’t adding a lot of quantifiable value to the world:”

  79. Assessment: SupportedSaid

    Gary Marcus

    “LLMs can still cannot be trusted:”

    What happened: The evaluation covers 6,000 questions across six domains. Only three evaluated models scored above an index value representing equal numbers of correct and incorrect answers, supporting continued verification of factual outputs. (Artificial Analysis / arXiv, )

  80. Assessment: UnresolvedSaid

    Gary Marcus

    “A large fraction of what LLMs do is mostly just memorization (and Hinton was on the wrong side of this argument):”

  81. Assessment: UnresolvedSaid

    Gary Marcus

    “If LeCun pulls the same stuff in his new company, morale will be deathly low.”

  82. Assessment: UnresolvedSaid

    Gary Marcus

    “My own view, which I stand by, is that we need “neurosymbolic” contributions from both traditions if we are to get to trustworthy AI.”

  83. Assessment: UnresolvedSaid

    Gary Marcus

    “Of course, the LLM companies have armies of humans; in my own brief experiments a couple hours later, OpenAI had apparently already patched this particular problem.”

  84. Assessment: UnresolvedSaid

    Gary Marcus

    “But nobody should be surprised by these errors; they emanate from the same dynamics as hallucinations: pure LLMs correlate words in their training, but lack stable, revisable world models.”

  85. Assessment: UnresolvedSaid

    Ed Zitron

    “which also includes evidence that OpenAI's revenues were at around $4.5 billion by the end of September, a vast difference from the $4.3 billion for the first half of the year published by other outlets.”

  86. Assessment: MixedSaid

    Ed Zitron

    “Deep Impact, my analysis of the DeepSeek situation and why it scared the American AI industry (clue: it's cost-related and nothing to do with "national security").”

    What held: The surrounding report attributes the selloff to DeepSeek challenging incumbent products while suggesting development costs could fall rapidly. This supports the cost-related component; the quoted market loss is evidence of the reaction, not a separate financial claim being graded. (Reuters, republished by ThePrint, )

    What did not: Amodei explicitly connects DeepSeek's releases to stronger export controls and discusses military dominance and US national security. This contradicts excluding security from the industry's expressed concerns, although it cannot establish his private motivation. (Dario Amodei, Anthropic CEO, )

  87. Assessment: UnresolvedSaid

    Ed Zitron

    “I published proof that OpenAI couldn't afford to pay Oracle and Oracle didn't have the capacity to service their farcical $300 billion, 5-year-long deal.”

  88. Assessment: UnresolvedSaid

    Ed Zitron

    “on how GPT-5's router mode actually costs OpenAI more money to run.”

  89. Assessment: ContradictedSaid

    Ed Zitron

    “how data center developers have no exit route, and US economic growth has become shouldered entirely by big tech.”

    What happened: BEA identifies health care among the leading contributors to increased services consumption. This contradicts the ordinary claim that growth came exclusively from big tech, without establishing a company-level share of GDP growth. (U.S. Bureau of Economic Analysis, )

  90. Assessment: UnresolvedSaid

    Ed Zitron

    “how its future relies on endless debt, how millions of GPUs are sitting waiting to be installed, and why it no longer makes sense to buy more GPUs.”

  91. Assessment: UnresolvedSaid

    Ed Zitron

    “how AI revenues will never pay back the cost of these GPUs”

  92. Assessment: UnresolvedSaid

    Ed Zitron

    “an analysis of reported cash burn and investments in OpenAI that proved the company burned more than $4 billion more than we know.”

  93. Assessment: UnresolvedSaid

    Ed Zitron

    “there are no profitable generative AI companies”

  94. Assessment: SupportedSaid

    Gary Marcus

    “Why LLMs just can’t stop hallucinating.”

    What happened: On a summarization faithfulness benchmark, leading frontier models from several labs hallucinated in over 10% of summaries in November 2025. (Vectara, )

  95. Assessment: UnresolvedSaid

    Gary Marcus

    “Neurosymbolic AI will displace pure LLMs over time; here’s why.”

  96. Assessment: UnresolvedSaid

    Ed Zitron

    “The mistake that every investor, commentator, analyst and member of the media makes about NVIDIA is believing that its sales are an expression of demand for AI compute, when it’s really more of a statement about the availability of debt from banks and private credit.”

  97. Assessment: UnresolvedSaid

    Ed Zitron

    “Anthropic will make no more than $5 billion this year, and has raised $17.5bn (with a further $2.5bn coming in the form of debt).”

  98. Assessment: ContradictedSaid

    Ed Zitron

    “The market for AI compute is not actually growing. In the last two years, no new major consumers of AI compute have emerged.”

    What happened: The study identifies Colossus among systems deployed in 2024 and documents its leading scale by March 2025. This establishes a major standalone consumer beyond OpenAI, Anthropic and the hyperscalers within the claimed window. (arXiv; Konstantin F. Pilz and coauthors, )

  99. Assessment: UnresolvedSaid

    Ed Zitron

    “Anthropic cannot afford to pay Broadcom $10 billion right now!”

  100. Assessment: ContradictedSaid

    Ed Zitron

    “I want to be really, abundantly clear about what’s happening: every single stock you see “growing because of AI” outside of those selling RAM and GPUs is actually growing because of something else.”

    What happened: Reuters reports an almost 9% early-trading gain after an AI-demand-driven forecast increase and connects the broader rally to AI tools and defense spending. This supports AI as a contributor, while acknowledging a material competing explanation. (Reuters, republished by Yahoo Finance, )