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

Changes, page 21 of 37

Every substantive change to the register, newest first. Speaker events record what the people covered said or changed. Site events record what this site published, corrected, flagged, or withdrew. The two never share a label.

Substantive changes (7383)

Page 21 of 37: changes 4001 to 4200 of 7383, newest first. Dates are when the change was recorded.

  1. Assessment publishedBy this siteGary Marcus

    If AI companies are not held liable for harms, we are in for a very rough ride.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-openais-disconcerting-hack-of-huggingface-b9d697-a1.

  2. Assessment publishedBy this siteGary Marcus

    OpenAI's production classifiers are likely to be permeable, just like all guardrails anybody has built to date.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-openais-disconcerting-hack-of-huggingface-6707a5-a1.

  3. HeldBy this siteGary Marcus

    Held record marcus-openais-disconcerting-hack-of-huggingface-53668a

    The two reviews differ: anthropic historical_observation/not_applicable/mixed, openai historical_observation/not_applicable/contradicted.

  4. Assessment publishedBy this siteGary Marcus

    Spending trillions on data centers, which risk blowing up the economy, is only adding fuel to the fire.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-openais-disconcerting-hack-of-huggingface-4ff47b-a1.

  5. Assessment publishedBy this siteGary Marcus

    The incident probably has implications for open-weight models, but they are complex and not yet clear.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-openais-disconcerting-hack-of-huggingface-1c558a-a1.

  6. HeldBy this siteGary Marcus

    Held record marcus-openais-disconcerting-hack-of-huggingface-10465f

    The openai review found the criteria unfair to the speaker's words.

  7. Assessment publishedBy this siteGary Marcus

    OpenAI's zero-day hack of HuggingFace should be treated as a wake-up call.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-openais-disconcerting-hack-of-huggingface-0d48b2-a1.

  8. Assessment publishedBy this siteGary Marcus

    It is still disconcerting that OpenAI's systems were able to hack into HuggingFace.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-openais-disconcerting-hack-of-huggingface-01be38-a1.

  11. HeldBy this siteGary Marcus

    Held record marcus-openais-dirty-december-o3-demo-doesnt-ae5158

    The two reviews differ: anthropic historical_observation/not_applicable/supported, openai historical_observation/not_applicable/unresolved.

  12. HeldBy this siteGary Marcus

    Held record marcus-openais-dirty-december-o3-demo-doesnt-aa698d

    The openai review found the criteria unfair to the speaker's words.

  14. HeldBy this siteGary Marcus

    Held record marcus-openais-dirty-december-o3-demo-doesnt-5de705

    The openai review found the criteria unfair to the speaker's words.

  15. HeldBy this siteGary Marcus

    Held record marcus-openais-dirty-december-o3-demo-doesnt-4d827e

    The anthropic screen reads the claim differently.

  16. HeldBy this siteGary Marcus

    Held record marcus-openais-dirty-december-o3-demo-doesnt-4a8019

    The final-text check failed: citations_correct; overclaims: The report documents timeouts and only 37 returned answers across 100 ARC-AGI-1 tasks for o3-high..

  17. HeldBy this siteGary Marcus

    Held record marcus-openais-dirty-december-o3-demo-doesnt-3bd254

    The openai screen reads the claim differently.

  18. Assessment publishedBy this siteGary Marcus

    Marcus says he is getting Theranos vibes from OpenAI's handling of the o3 results.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-openais-dirty-december-o3-demo-doesnt-233158-a1.

  21. Assessment publishedBy this siteGary Marcus

    OpenAI's code red may stem in part from a recent exodus of users.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-openais-code-red-feccc6-a1.

  22. HeldBy this siteGary Marcus

    Held record marcus-openais-code-red-e0029e

    hedged language is not scored, even on a plain reading

  23. Assessment publishedBy this siteGary Marcus

    If most or all investors opt out, OpenAI's valuation could drop, perhaps precipitously, like WeWork's.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-openais-code-red-c53012-a1.

  24. Assessment publishedBy this siteGary Marcus

    If OpenAI drops, Nvidia might well drop too.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-openais-code-red-c48fd2-a1.

  25. HeldBy this siteGary Marcus

    Held record marcus-openais-code-red-c3bc65

    The openai review found the criteria unfair to the speaker's words.

  26. Assessment publishedBy this siteGary Marcus

    OpenAI's unraveling has begun.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-openais-code-red-a687a9-a1.

  27. Assessment publishedBy this siteGary Marcus

    If OpenAI drops, other companies with many future OpenAI orders on their books might also drop.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-openais-code-red-a52e2a-a1.

  28. HeldBy this siteGary Marcus

    Held record marcus-openais-code-red-97a216

    e1 measures something else (metric net income (loss), implied from Microsoft's equity-method share is not net income (loss)); it can only be context unless it is derived with stated inputs; e2 measures something else (metric net income (loss), projected is not net income (loss)); it can only be context unless it is derived with stated inputs; e3 is decisive for a financial claim but states no metric scope; e4 measures something else (metric net loss excluding one-time charges is not net income (loss)); it can only be context unless it is derived with stated inputs

  30. HeldBy this siteGary Marcus

    Held record marcus-openais-code-red-6ad305

    e1 is decisive for a financial claim but states no metric scope

  31. Assessment publishedBy this siteGary Marcus

    In the worst case, if OpenAI falls, the government might seriously consider a taxpayer-funded bailout.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-openais-code-red-3613cb-a1.

  32. HeldBy this siteGary Marcus

    Held record marcus-openais-code-red-339082

    e1 is decisive for a financial claim but states no metric scope

  33. HeldBy this siteGary Marcus

    Held record marcus-openais-code-red-24da7e

    The final-text check failed: citations_correct; overclaims: OpenAI did not fall..

  34. Assessment publishedBy this siteGary Marcus

    OpenAI is burning billions of dollars per month.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-openais-code-red-111356-a2.

  35. Assessment publishedBy this siteGary Marcus

    Astra's math success is possible because math allows external tools for verification and synthetic data generation.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-openais-amazing-but-vastly-oversold-fc16b6-a2.

  36. Assessment publishedBy this siteGary Marcus

    OpenAI's Astra announcement was marketing rather than science and disclosed no information about how the results were achieved.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-openais-amazing-but-vastly-oversold-efd1af-a1.

  37. Assessment publishedBy this siteGary Marcus

    Astra is unlikely to lead directly to a cancer cure or enormous advances in materials, energy, or drug discovery.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-openais-amazing-but-vastly-oversold-e93f85-a1.

  38. Assessment publishedBy this siteGary Marcus

    Astra is unlikely to score even 5 out of 10 on the 2024 Marcus–Brundage bet.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-openais-amazing-but-vastly-oversold-d10cdf-a1.

  39. Assessment publishedBy this siteGary Marcus

    Astra is very impressive, but Marcus sees no reason whatsoever to think Astra is AGI, let alone ASI.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-openais-amazing-but-vastly-oversold-cfd4e8-a1.

  40. Assessment publishedBy this siteGary Marcus

    The 'AGI-is-near' community keeps committing the same logical fallacy, the fallacy of composition, over and over.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-openais-amazing-but-vastly-oversold-9ae3b7-a1.

  41. Assessment publishedBy this siteGary Marcus

    Astra is very unlikely to solve all the reliability problems of current models.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-openais-amazing-but-vastly-oversold-94f6c4-a1.

  42. HeldBy this siteGary Marcus

    Held record marcus-openais-amazing-but-vastly-oversold-90ceaa

    The openai screen reads the claim differently.

  43. HeldBy this siteGary Marcus

    Held record marcus-openais-amazing-but-vastly-oversold-8bc3e0

    The anthropic screen reads the claim differently.

  44. Assessment publishedBy this siteGary Marcus

    Astra does not represent the Singularity.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-openais-amazing-but-vastly-oversold-7dd7e1-a1.

  45. Assessment publishedBy this siteGary Marcus

    Astra's apparent strength in some math does not imply it will avoid hallucinations or solve GenAI reliability problems.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-openais-amazing-but-vastly-oversold-78885a-a1.

  46. Assessment publishedBy this siteGary Marcus

    Models shine at math and coding because those domains allow verification and cheap correct synthetic data, which is not true of domains in general.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-openais-amazing-but-vastly-oversold-763b2e-a1.

  47. Assessment publishedBy this siteGary Marcus

    Astra is good at some math, but it is very doubtful that math has been solved.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-openais-amazing-but-vastly-oversold-6df0ae-a1.

  48. Assessment publishedBy this siteGary Marcus

    Astra, the new model OpenAI is testing internally, is amazing.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-openais-amazing-but-vastly-oversold-6a6c5c-a1.

  49. Assessment publishedBy this siteGary Marcus

    Astra's results contain no new theory or techniques and are nowhere near the significance of history's major mathematical developments.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-openais-amazing-but-vastly-oversold-6905ff-a1.

  51. Assessment publishedBy this siteGary Marcus

    Astra will probably not be able to write usable scripts for Sabine Hossenfelder's YouTube series.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-openais-amazing-but-vastly-oversold-54cac3-a1.

  52. Assessment publishedBy this siteGary Marcus

    Astra's results do not mean we have entered the era of automated scientific discovery.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-openais-amazing-but-vastly-oversold-4bbb80-a1.

  53. Assessment publishedBy this siteGary Marcus

    There is an important, principled reason to think that AI success on math is a special case that will not generalize as much as people might hope.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-openais-amazing-but-vastly-oversold-3e407d-a1.

  54. Assessment publishedBy this siteGary Marcus

    Astra will probably not be able to reliably extract numbers from arbitrary PDFs.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-openais-amazing-but-vastly-oversold-3cc78d-a1.

  55. Assessment publishedBy this siteGary Marcus

    People hoping Astra will be magic are likely to be disappointed.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-openais-amazing-but-vastly-oversold-026f12-a1.

  56. Assessment publishedBy this siteGary Marcus

    OpenAI recently invested in a $60M fundraise for a webcam company and is planning a hardware joint venture with it.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-openai-on-a-path-to-becoming-the-fdfbc0-a1.

  57. Assessment publishedBy this siteGary Marcus

    Sam Altman wants to know everything about people and make money from it.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-openai-on-a-path-to-becoming-the-f59e29-a1.

  58. Assessment publishedBy this siteGary Marcus

    Sam Altman might succeed in knowing and monetizing everything about people.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-openai-on-a-path-to-becoming-the-f0b070-a1.

  59. HeldBy this siteGary Marcus

    Held record marcus-openai-on-a-path-to-becoming-the-cdfc48

    The final-text check failed: citations_correct.

  60. Assessment publishedBy this siteGary Marcus

    Sam Altman has acknowledged that he wants to train AI on everyone's personal documents, such as Word files and email.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-openai-on-a-path-to-becoming-the-c9ddff-a1.

  61. Assessment publishedBy this siteGary Marcus

    Something terrifying has just snapped into place, as shown by the developments listed after this line.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-openai-on-a-path-to-becoming-the-359df1-a1.

  62. Assessment publishedBy this siteGary Marcus

    OpenAI recently put Paul Nakasone, formerly of the NSA, on its board.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-openai-on-a-path-to-becoming-the-097a75-a1.

  63. Assessment publishedBy this siteGary Marcus

    Microsoft and OpenAI's only way to make generative AI profitable is around-the-clock surveillance of users.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-openai-is-becoming-a-surveillance-ee342e-a1.

  64. Assessment publishedBy this siteGary Marcus

    ChatGPT's new Apple Messages plugin certainly raises privacy concerns.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-openai-is-becoming-a-surveillance-c0fdfa-a1.

  65. Assessment publishedBy this siteGary Marcus

    It is plausible that OpenAI will turn into a surveillance company.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-openai-is-becoming-a-surveillance-9d9a76-a1.

  66. HeldBy this siteGary Marcus

    Held record marcus-openai-is-becoming-a-surveillance-94a209

    The openai review found the criteria unfair to the speaker's words.

  67. HeldBy this siteGary Marcus

    Held record marcus-openai-is-becoming-a-surveillance-4b588f

    The final-text check failed: excerpt_matches, conditions_shown, rationale_supported_by_evidence.

  68. HeldBy this siteGary Marcus

    Held record marcus-openai-is-becoming-a-surveillance-43029f

    The two reviews differ: anthropic historical_observation/not_applicable/mixed, openai historical_observation/not_applicable/unresolved.

  69. Assessment publishedBy this siteGary Marcus

    Marcus expects the two new members of OpenAI's reconstituted board to be neutral.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-openai-is-back-in-business-a494ed-a1.

  70. Assessment publishedBy this siteGary Marcus

    Marcus is disappointed that OpenAI's new board has no women and hopes the board will expand in due course.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-openai-is-back-in-business-5e2aaa-a1.

  71. Assessment publishedBy this siteGary Marcus

    The public may never fully learn what happened in the November 2023 OpenAI board crisis.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-openai-is-back-in-business-1d3bd5-a1.

  72. Assessment publishedBy this siteGary Marcus

    The most important thing for humanity is that OpenAI sticks to its original mission.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-openai-is-back-in-business-1a26a2-a1.

  74. Assessment publishedBy this siteGary Marcus

    Altman was obviously fibbing when he said 'We are now confident we know how to build AGI as we have traditionally understood it'.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-openai-in-deep-trouble-f86d83-a1.

  75. Assessment publishedBy this siteGary Marcus

    GPT-4.5 appears to be the system intended as GPT-5, renamed to lower expectations.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-openai-in-deep-trouble-cf0147-a1.

  76. HeldBy this siteGary Marcus

    Held record marcus-openai-in-deep-trouble-a9e42a

    The final-text check failed: citations_correct; overclaims: e2 explanation: "The reviewers identified practical writing strengths, cautioning against treating coding results as the whole product.".

  77. Assessment publishedBy this siteGary Marcus

    Potential investors should question Altman's technical management and OpenAI's ability to deliver.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-openai-in-deep-trouble-9d0172-a1.

  78. Assessment publishedBy this siteGary Marcus

    GPT-4.5 was OpenAI's best attempt after DeepSeek to show it still leads, and it failed.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-openai-in-deep-trouble-8d14d0-a1.

  79. Assessment publishedBy this siteGary Marcus

    OpenAI is probably still losing money on every prompt.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-openai-in-deep-trouble-8973e6-a1.

  80. Assessment publishedBy this siteGary Marcus

    GPT-4.5 is the best Altman has got, and Altman was bluffing the whole time.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-openai-in-deep-trouble-80805d-a1.

  81. HeldBy this siteGary Marcus

    Held record marcus-openai-in-deep-trouble-735095

    The anthropic screen reads the claim differently.

  82. Assessment publishedBy this siteGary Marcus

    OpenAI still has not released GPT-5 after two years of hype.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-openai-in-deep-trouble-68a900-a1.

  83. HeldBy this siteGary Marcus

    Held record marcus-openai-in-deep-trouble-663d82

    The openai review does not accept e4 as contradicts.

  84. Assessment publishedBy this siteGary Marcus

    Altman may lack the intellectual vision to take OpenAI to the next level.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-openai-in-deep-trouble-10576f-a1.

  85. Assessment publishedBy this siteGary Marcus

    Sam Altman's credibility has declined.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-openai-in-deep-trouble-040164-a1.

  86. Assessment publishedBy this siteGary Marcus

    OpenAI may well have used content from YouTube, the New York Times, and many artists and writers without permission, much as DeepSeek allegedly did to OpenAI.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-openai-cries-foul-ccfa02-a1.

  87. HeldBy this siteGary Marcus

    Held record marcus-openai-cries-foul-82b9ec

    The anthropic screen reads the claim differently.

  88. Assessment publishedBy this siteGary Marcus

    OpenAI being copied by DeepSeek is karmic payback for its own use of others' IP.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-openai-cries-foul-5dcbf6-a1.

  89. Assessment publishedBy this siteGary Marcus

    It is plausible that DeepSeek broke OpenAI's Terms of Service by distilling OpenAI's models without permission.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-openai-cries-foul-1fdf3c-a1.

  90. Assessment publishedBy this siteGary Marcus

    The media and everyone else should not confuse open-weight models with open-source models.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-open-source-is-not-the-same-as-open-f838f9-a1.

  92. Assessment publishedBy this siteGary Marcus

    How much of what a model does is regurgitation versus how much extends beyond its training data is arguably the central question in AI.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-open-source-is-not-the-same-as-open-869077-a1.

  93. Assessment publishedBy this siteGary Marcus

    Like pigeons being 'just rats with better PR', open-weight is open-source with all the good press but far fewer advantages.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-open-source-is-not-the-same-as-open-27bb2b-a1.

  95. HeldBy this siteGary Marcus

    Held record marcus-open-source-is-not-the-same-as-open-2451f8

    The anthropic screen finds the claim checkable, so it must be graded.

  96. Assessment publishedBy this siteGary Marcus

    EU leaders should not gut (drastically weaken) the EU AI Act, because it is needed now more than ever.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-open-letter-to-all-eu-leaders-e4c6eb-a1.

  97. Assessment publishedBy this siteGary Marcus

    The EU AI Act is needed now more than ever, given the turmoil at OpenAI.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-open-letter-to-all-eu-leaders-e4c6eb-2-a1.

  98. Assessment publishedBy this siteGary Marcus

    The November 2023 events at OpenAI are likely to lead to considerable, unpredictable instability.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-open-letter-to-all-eu-leaders-bb3b9b-a1.

  100. Assessment publishedBy this siteGary Marcus

    AI models need more innate structure than pure multilayer perceptrons, even before learning.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-open-letter-responding-to-yann-lecun-c7efc3-a1.

  101. HeldBy this siteGary Marcus

    Held record marcus-open-letter-responding-to-yann-lecun-b3f296

    The final-text check failed: citations_correct; overclaims: e2 excerpt "outperforms all baselines" (truncated from "outperforms all baselines on data efficiency").

  102. HeldBy this siteGary Marcus

    Held record marcus-open-letter-responding-to-yann-lecun-81296b

    e1 is dated 2024, which is not inside the period described (2024-01-01 to 2024-06-04); it can only be context

  106. Assessment publishedBy this siteGary Marcus

    Marcus and LeCun are among the minority of people who have come to fully understand just how limited LLMs are and what needs to be done next.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-open-letter-responding-to-yann-lecun-09dedf-a1.

  107. HeldBy this siteGary Marcus

    Held record marcus-oops-how-google-bombed-while-doing-e303fd

    The final-text check failed: citations_correct.

  109. Assessment publishedBy this siteGary Marcus

    Observers should be asking what competitive moat exists in chatbot search when Google and Microsoft are about to offer basically the same thing.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-oops-how-google-bombed-while-doing-d6992b-a1.

  110. Assessment publishedBy this siteGary Marcus

    The hallucination problem in neural networks has persisted since 2001, so assuming it will soon go away ignores 20 years of history.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-oops-how-google-bombed-while-doing-d17e5e-a2.

  111. HeldBy this siteGary Marcus

    Held record marcus-oops-how-google-bombed-while-doing-bb9161

    e1 is decisive for a financial claim but states no metric scope

  112. Assessment publishedBy this siteGary Marcus

    Hallucinating web search could be dangerous if left to run amok in domains like medicine, and promises to fix it are scientifically problematic.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-oops-how-google-bombed-while-doing-772c19-a1.

  113. HeldBy this siteGary Marcus

    Held record marcus-oops-how-google-bombed-while-doing-6726ff

    The anthropic review does not accept e4 as contradicts.

  114. Assessment publishedBy this siteGary Marcus

    If hallucinations are not fixed soon, which they may not be, people may quickly tire of chat-based search and return to traditional search.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-oops-how-google-bombed-while-doing-35c27e-a1.

  116. Assessment publishedBy this siteGary Marcus

    Most likely neither Google's nor Microsoft's chatbot search product is particularly reliable.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-oops-how-google-bombed-while-doing-0dfcae-a1.

  118. HeldBy this siteGary Marcus

    Held record marcus-one-genai-image-ten-errors-c5349e

    The anthropic review reads the claim differently.

  120. Assessment publishedBy this siteGary Marcus

    The physical errors (and one conceptual error) in the Midjourney image discussed in the post are of a piece with hallucinations and near-plagiarism.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-one-genai-image-ten-errors-37e75c-a1.

  122. Assessment publishedBy this siteGary Marcus

    Readers should never trust AI hype, such as claims that ChatGPT writes poetry indistinguishable from Shakespeare.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-on-hype-and-the-unbearable-banality-3ae6b2-a1.

  124. Assessment publishedBy this siteGary Marcus

    Employees should not panic yet about AI replacing their jobs.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-on-employment-dont-panic-yet-2603c4-a1.

  125. Assessment publishedBy this siteGary Marcus

    For now, employers should not focus on replacing humans with AI and should instead use AI to help their existing workers.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-on-employment-dont-panic-yet-219c0b-a1.

  127. HeldBy this siteGary Marcus

    Held record marcus-o3-agi-the-art-of-the-demo-and-what-cdebf1

    The anthropic review does not accept e9 as contradicts.

  128. Assessment publishedBy this siteGary Marcus

    The o3 demo does not change the view that AGI requires true logical reasoning rather than statistical tricks.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-o3-agi-the-art-of-the-demo-and-what-b9eae1-a1.

  130. Assessment publishedBy this siteGary Marcus

    No one outside OpenAI has evaluated o3's robustness across different problem types.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-o3-agi-the-art-of-the-demo-and-what-a8b6ba-a1.

  131. Assessment publishedBy this siteGary Marcus

    o3 may work on problems allowing massive data augmentation but not on problems that do not.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-o3-agi-the-art-of-the-demo-and-what-a36471-a1.

  132. Assessment publishedBy this siteGary Marcus

    During 2025, many AI influencers and possibly some companies will claim AGI has been reached.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-o3-agi-the-art-of-the-demo-and-what-9f5bee-a1.

  133. Assessment publishedBy this siteGary Marcus

    OpenAI's o3 is not the long-awaited GPT-5.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-o3-agi-the-art-of-the-demo-and-what-9b659d-a1.

  134. Assessment publishedBy this siteGary Marcus

    o3's best demo results showed truly impressive major advances on ARC and FrontierMath.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-o3-agi-the-art-of-the-demo-and-what-9a94fb-a2.

  135. Assessment publishedBy this siteGary Marcus

    As of the demo, o3 had not been released to the public nor vetted by the scientific community.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-o3-agi-the-art-of-the-demo-and-what-8bf959-a1.

  136. HeldBy this siteGary Marcus

    Held record marcus-o3-agi-the-art-of-the-demo-and-what-6ddd11

    e2 is dated 2024-12, which is not inside the period described (… to 2024-12-20); it can only be context

  137. HeldBy this siteGary Marcus

    Held record marcus-o3-agi-the-art-of-the-demo-and-what-612921

    "likely" is hedged or probabilistic; one outcome does not score it; hedged language is not scored, even on a plain reading

  138. Assessment publishedBy this siteGary Marcus

    The environmental impact of such high-compute AI systems could be immense.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-o3-agi-the-art-of-the-demo-and-what-55f5eb-a1.

  139. Assessment publishedBy this siteGary Marcus

    AI demos should not be taken seriously until the products are released and subjected to outside scrutiny.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-o3-agi-the-art-of-the-demo-and-what-4ff1bc-a1.

  140. Assessment publishedBy this siteGary Marcus

    Those claiming AGI are likely to craft definitions post hoc or give none at all.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-o3-agi-the-art-of-the-demo-and-what-4abc95-a1.

  141. Assessment publishedBy this siteGary Marcus

    Many basic problems on the path to AGI remain unsolved.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-o3-agi-the-art-of-the-demo-and-what-256352-a1.

  142. Assessment publishedBy this siteGary Marcus

    People should not assume o3 is reliable until many people have tested it on many kinds of tasks.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-o3-agi-the-art-of-the-demo-and-what-21773e-a1.

  143. Assessment publishedBy this siteGary Marcus

    Spending a quarter trillion dollars to get systems 1000x more expensive than humans is maybe not the right approach.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-o3-agi-the-art-of-the-demo-and-what-17e769-a1.

  144. HeldBy this siteGary Marcus

    Held record marcus-o3-agi-the-art-of-the-demo-and-what-11221b

    The anthropic screen reads the claim differently.

  145. Assessment publishedBy this siteGary Marcus

    The economics of generative AI could get wild.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-nvidia-could-soon-take-a-serious-fec58e-a1.

  146. HeldBy this siteGary Marcus

    Held record marcus-nvidia-could-soon-take-a-serious-90b3b7

    The openai screen reads the claim differently.

  149. Assessment publishedBy this siteGary Marcus

    Marcus doubts that Altman would have given up his OpenAI board seat if there had really been nothing against him that even looked like a legitimate grievance.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-not-consistently-candid-821d92-a1.

  150. Assessment publishedBy this siteGary Marcus

    Altman's indirect stake in OpenAI through Y Combinator may be worth tens of millions of dollars.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-not-consistently-candid-6f602e-a1.

  152. Assessment publishedBy this siteGary Marcus

    If Altman was already working on the Rain chip deal when he testified to the Senate in May 2023, he should have disclosed it.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-not-consistently-candid-53ab62-a1.

  153. Assessment publishedBy this siteGary Marcus

    If the OpenAI board was trying to return OpenAI to its original mission of AI for human interest rather than profit, it should be praised, not condemned.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-not-consistently-candid-17f348-a1.

  154. Assessment publishedBy this siteGary Marcus

    LaMDA is not sentient.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-nonsense-on-stilts-d6fb09-a1.

  155. HeldBy this siteGary Marcus

    Held record marcus-nonsense-on-stilts-d3b80d

    The anthropic screen reads the claim differently.

  156. Assessment publishedBy this siteGary Marcus

    Language models like LaMDA string words together without any coherent understanding of the world.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-nonsense-on-stilts-cd38ec-a1.

  157. Assessment publishedBy this siteGary Marcus

    Marcus doubts that LaMDA and similar language models will play any important role in the future of AI.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-nonsense-on-stilts-a29178-a1.

  158. Assessment publishedBy this siteGary Marcus

    No AI system that anyone knew how to build in 2022 is sentient.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-nonsense-on-stilts-9d49da-a1.

  159. Assessment publishedBy this siteGary Marcus

    LaMDA-style language models only match statistical patterns of human language, and their output has no real meaning.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-nonsense-on-stilts-96e570-a1.

  160. Assessment publishedBy this siteGary Marcus

    Neither LaMDA nor similar large language models such as GPT-3 are remotely intelligent.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-nonsense-on-stilts-75d580-a1.

  161. Assessment publishedBy this siteGary Marcus

    Fooling people into thinking a program is intelligent is not the same as building an intelligent program.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-nonsense-on-stilts-48f130-a1.

  162. Assessment publishedBy this siteGary Marcus

    Everything LaMDA says is bullshit with no real meaning behind it.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-nonsense-on-stilts-28eaa9-a1.

  163. Assessment publishedBy this siteGary Marcus

    Current AI does not firmly have the sound logic, reliable causal reasoning, and originality that Nobel-calibre science requires.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-nobel-prizes-and-the-ai-hype-hall-cd169c-a1.

  164. HeldBy this siteGary Marcus

    Held record marcus-nobel-prizes-and-the-ai-hype-hall-c80802

    the horizon on 2026-10-28 is not_due, not open_horizon

  165. HeldBy this siteGary Marcus

    Held record marcus-nobel-prizes-and-the-ai-hype-hall-c3ca51

    hedged language is not scored, even on a plain reading

  166. Assessment publishedBy this siteGary Marcus

    Even in the best AI-for-science work, such as Google's AI co-scientist, human scientists still choose the research goals.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-nobel-prizes-and-the-ai-hype-hall-bfc85e-a1.

  167. Assessment publishedBy this siteGary Marcus

    In AI it is easy to build demos covering 80% of a task such as driving or a scientific problem, but getting the last 20% is often very hard.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-nobel-prizes-and-the-ai-hype-hall-b2dc5a-a1.

  168. HeldBy this siteGary Marcus

    Held record marcus-nobel-prizes-and-the-ai-hype-hall-9dfc90

    The anthropic screen reads the claim differently.

  169. HeldBy this siteGary Marcus

    Held record marcus-nobel-prizes-and-the-ai-hype-hall-8558de

    The anthropic screen reads the claim differently.

  170. Assessment publishedBy this siteGary Marcus

    Nobel-calibre science is much harder than driving, and AI still has not reliably solved driving.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-nobel-prizes-and-the-ai-hype-hall-81f40e-a1.

  171. Assessment publishedBy this siteGary Marcus

    Current AI cannot reliably solve river-crossing puzzles.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-nobel-prizes-and-the-ai-hype-hall-612ee6-2-a1.

  172. Assessment publishedBy this siteGary Marcus

    Elon Musk's 2019 promise of over a million Tesla robotaxis has largely failed to come true.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-nobel-prizes-and-the-ai-hype-hall-33890b-a1.

  173. HeldBy this siteGary Marcus

    Held record marcus-nobel-prizes-and-the-ai-hype-hall-137618

    hedged language is not scored, even on a plain reading

  174. HeldBy this siteGary Marcus

    Held record marcus-nobel-prizes-and-the-ai-hype-hall-0ac7a1

    the conclusive event's date differs from its evidence

  175. HeldBy this siteGary Marcus

    Held record marcus-nobel-prizes-and-the-ai-hype-hall-08b51c

    The openai review found the criteria unfair to the speaker's words.

  176. HeldBy this siteGary Marcus

    Held record marcus-noam-chomsky-and-gpt-3-faccda

    The openai review found the criteria unfair to the speaker's words.

  177. Assessment publishedBy this siteGary Marcus

    GPT-3 cannot avoid producing misinformation because it models how words relate to each other, not how language relates to the world.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-noam-chomsky-and-gpt-3-f648bc-a1.

  178. Assessment publishedBy this siteGary Marcus

    LLMs' failures argue strongly for exploring Kantian AI systems that start with built-in representations of space, time, causality and some universal grammar.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-noam-chomsky-and-gpt-3-d8d192-a1.

  179. Assessment publishedBy this siteGary Marcus

    For AI to understand language, it should process language through built-in commonsense understanding instead of starting as a blank slate and learning statistics.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-noam-chomsky-and-gpt-3-c49391-a1.

  180. Assessment publishedBy this siteGary Marcus

    Because systems like GPT-3 don't tell us why human languages have the special character that they do, they have little explanatory value.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-noam-chomsky-and-gpt-3-c476d0-a1.

  182. HeldBy this siteGary Marcus

    Held record marcus-noam-chomsky-and-gpt-3-75df3d

    The final-text check failed: rationale_supported_by_evidence; overclaims: e2 studies GPT-2 and smaller models, not GPT-3..

  183. Assessment publishedBy this siteGary Marcus

    Systems like GPT-3 do not gravitate towards human language, and they never develop a cohesive understanding of the world, no matter how much experience they have.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-noam-chomsky-and-gpt-3-61b596-a1.

  185. Assessment publishedBy this siteGary Marcus

    New AI systems with a firmer innate, Kantian basis might eventually tell us more about the human mind.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-noam-chomsky-and-gpt-3-5ad6a3-a1.

  186. Assessment publishedBy this siteGary Marcus

    Doing well on a benchmark does not amount to being intelligent.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-no-virginia-agi-is-not-imminent-e8ff52-a1.

  187. Assessment publishedBy this siteGary Marcus

    The AI field should not lower its AGI goal from Star Trek computer-level to merely beating an average person.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-no-virginia-agi-is-not-imminent-d9df8f-a1.

  189. Assessment publishedBy this siteGary Marcus

    A genuine AGI must reliably do multidigit arithmetic.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-no-virginia-agi-is-not-imminent-cda2a4-a1.

  190. Assessment publishedBy this siteGary Marcus

    Around early 2022, before LLM euphoria, the field assumed AGI meant more than beating ordinary humans.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-no-virginia-agi-is-not-imminent-cc7110-a2.

  191. Assessment publishedBy this siteGary Marcus

    AI still has a long way to go on reasoning, compositionality, planning, and factuality.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-no-virginia-agi-is-not-imminent-c974be-a1.

  192. Assessment publishedBy this siteGary Marcus

    An AGI should do what it is asked, decline what it cannot do, and know the difference.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-no-virginia-agi-is-not-imminent-c8e452-a1.

  193. Assessment publishedBy this siteGary Marcus

    LLMs cannot reliably write summaries without hallucinating.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-no-virginia-agi-is-not-imminent-b7e604-a1.

  194. HeldBy this siteGary Marcus

    Held record marcus-no-virginia-agi-is-not-imminent-aa764b

    The final-text check failed: citations_correct.

  195. HeldBy this siteGary Marcus

    Held record marcus-no-virginia-agi-is-not-imminent-6f7105

    The anthropic review reads the claim differently.

  197. Assessment publishedBy this siteGary Marcus

    Marcus doubts LLMs or any extant AI can summarize and reliably answer questions about original films and TV programs.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-no-virginia-agi-is-not-imminent-544b2c-a1.

  198. Assessment publishedBy this siteGary Marcus

    Beating AI benchmarks is often a statistical artifact and says less than people assume.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-no-virginia-agi-is-not-imminent-5387d8-a1.

  199. Assessment publishedBy this siteGary Marcus

    Without knowing the training corpus, one can hardly conclude what a machine understands or how general it is.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-no-virginia-agi-is-not-imminent-447af8-a1.

  200. Assessment publishedBy this siteGary Marcus

    AI benchmarks are often quite gameable.

    First assessment published after both reviews and the final-text check passed.

    Assessment marcus-no-virginia-agi-is-not-imminent-39aa05-a2.