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

Changes, page 14 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 14 of 37: changes 2601 to 2800 of 7383, newest first. Dates are when the change was recorded.

  1. Assessment publishedBy this siteGary Marcus

    Most businesses have struggled to get return on generative AI investment because of its ongoing massive reliability issues.

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

    Assessment marcus-turns-out-generative-ai-was-a-scam-7fdefa-a2.

  2. HeldBy this siteGary Marcus

    Held record marcus-turns-out-generative-ai-was-a-scam-7e44f0

    The anthropic screen reads the claim differently.

  3. Assessment publishedBy this siteGary Marcus

    Generative AI vendors have marketed it as tantamount to AGI, but it is not AGI.

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

    Assessment marcus-turns-out-generative-ai-was-a-scam-7a1e98-a2.

  4. HeldBy this siteGary Marcus

    Held record marcus-turns-out-generative-ai-was-a-scam-758d9f

    The openai screen reads the claim differently.

  5. Assessment publishedBy this siteGary Marcus

    Marcus's best guess is that generative AI will ultimately have done significantly more harm to society than good.

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

    Assessment marcus-turns-out-generative-ai-was-a-scam-5f47ca-a1.

  6. HeldBy this siteGary Marcus

    Held record marcus-turns-out-generative-ai-was-a-scam-446f8c

    e2 is dated 2026-02, which is not inside the period described (2021-02-23 to 2026-02-23); it can only be context

  7. HeldBy this siteGary Marcus

    Held record marcus-turns-out-generative-ai-was-a-scam-42f7e4

    The openai review does not accept e2 as supports.

  8. Assessment publishedBy this siteGary Marcus

    Generative AI was not in fact contributing half of US GDP growth, contrary to the White House AI advisor's claim.

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

    Assessment marcus-turns-out-generative-ai-was-a-scam-3f62cd-a1.

  9. Assessment publishedBy this siteGary Marcus

    Generative AI may well lay waste to the economy once banks and investors who bought the hype start to fail.

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

    Assessment marcus-turns-out-generative-ai-was-a-scam-2bce58-a1.

  10. Assessment publishedBy this siteGary Marcus

    Generative AI is leading some people into serious mental health issues.

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

    Assessment marcus-turns-out-generative-ai-was-a-scam-26f0b1-a2.

  11. Assessment publishedBy this siteGary Marcus

    Generative AI has some practical uses such as coding but is an inherently unreliable technology.

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

    Assessment marcus-turns-out-generative-ai-was-a-scam-1080d5-a2.

  12. HeldBy this siteGary Marcus

    Held record marcus-top-5-reasons-why-openai-was-probably-fc511a

    The final-text check failed: citations_correct.

  13. Assessment publishedBy this siteGary Marcus

    OpenAI's income this year is mostly from proof-of-concept testing, and that income could dry up.

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

    Assessment marcus-top-5-reasons-why-openai-was-probably-f090f6-a1.

  14. Assessment publishedBy this siteGary Marcus

    Someone other than OpenAI, such as DeepMind or Anthropic, might well find the new paradigm beyond current LLMs before OpenAI does.

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

    Assessment marcus-top-5-reasons-why-openai-was-probably-ea3bf3-a1.

  15. Assessment publishedBy this siteGary Marcus

    Large language models alone were never going to solve the alignment problem.

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

    Assessment marcus-top-5-reasons-why-openai-was-probably-db74ae-a2.

  16. Assessment publishedBy this siteGary Marcus

    The hallucination problem has never been solved and may not be until a new paradigm is found.

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

    Assessment marcus-top-5-reasons-why-openai-was-probably-d0b587-a1.

  17. HeldBy this siteGary Marcus

    Held record marcus-top-5-reasons-why-openai-was-probably-be4c65

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

  18. Assessment publishedBy this siteGary Marcus

    Even if OpenAI gets past the board drama, the profit-vs-safety tension will likely remain and weaken its value as a business.

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

    Assessment marcus-top-5-reasons-why-openai-was-probably-bb72b0-a1.

  19. HeldBy this siteGary Marcus

    Held record marcus-top-5-reasons-why-openai-was-probably-b25589

    e3 is dated 2023-11, which is not inside the period described (2023-01-01 to 2023-11-22); it can only be context

  20. HeldBy this siteGary Marcus

    Held record marcus-top-5-reasons-why-openai-was-probably-aeaddb

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

  22. Assessment publishedBy this siteGary Marcus

    OpenAI's nonprofit mission was inherently at odds with a pure for-profit business.

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

    Assessment marcus-top-5-reasons-why-openai-was-probably-8faa1a-a1.

  23. HeldBy this siteGary Marcus

    Held record marcus-top-5-reasons-why-openai-was-probably-8bac5b

    The anthropic screen reads the claim differently.

  24. HeldBy this siteGary Marcus

    Held record marcus-top-5-reasons-why-openai-was-probably-82586a

    The anthropic review reads the claim differently.

  25. HeldBy this siteGary Marcus

    Held record marcus-top-5-reasons-why-openai-was-probably-703bc7

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

  26. HeldBy this siteGary Marcus

    Held record marcus-top-5-reasons-why-openai-was-probably-6af3d9

    The openai screen doubts the attribution.

  27. Assessment publishedBy this siteGary Marcus

    OpenAI never seemed ready to move beyond LLMs.

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

    Assessment marcus-top-5-reasons-why-openai-was-probably-679855-a1.

  28. Assessment publishedBy this siteGary Marcus

    OpenAI's strength has been scaling the Transformer paradigm developed at Google, not developing new paradigms.

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

    Assessment marcus-top-5-reasons-why-openai-was-probably-65aca4-a1.

  29. Assessment publishedBy this siteGary Marcus

    Thrive Capital probably will not now invest in OpenAI at an $86B valuation.

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

    Assessment marcus-top-5-reasons-why-openai-was-probably-46851a-a1.

  30. Assessment publishedBy this siteGary Marcus

    OpenAI's nonprofit-hybrid structure was a ticking time bomb because profit-seekers and safety-focused nonprofit stewards would inevitably conflict.

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

    Assessment marcus-top-5-reasons-why-openai-was-probably-42bfd3-a1.

  31. HeldBy this siteGary Marcus

    Held record marcus-top-5-reasons-why-openai-was-probably-40f25c

    The anthropic screen reads the claim differently.

  32. Assessment publishedBy this siteGary Marcus

    Investors should have been deterred by OpenAI's prospectus warning that investment be viewed as a donation.

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

    Assessment marcus-top-5-reasons-why-openai-was-probably-13cacb-a1.

  33. Assessment publishedBy this siteGary Marcus

    OpenAI's real value was always its people, not its IP, data, customer list or infrastructure.

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

    Assessment marcus-top-5-reasons-why-openai-was-probably-1258ed-a1.

  35. Assessment publishedBy this siteGary Marcus

    Tesla's Optimus demo on Friday, September 30, 2022 may not be the big demo Musk's fans are hoping for.

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

    Assessment marcus-top-5-reasons-teslas-optimus-announcement-f0f352-a1.

  36. HeldBy this siteGary Marcus

    Held record marcus-top-5-reasons-teslas-optimus-announcement-f0f352-2

    The anthropic screen doubts the attribution.

  37. HeldBy this siteGary Marcus

    Held record marcus-top-5-reasons-teslas-optimus-announcement-c0443c

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

  38. Assessment publishedBy this siteGary Marcus

    Elon Musk's deadlines for technology projects are almost never met.

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

    Assessment marcus-top-5-reasons-teslas-optimus-announcement-b1e823-a1.

  39. Assessment publishedBy this siteGary Marcus

    The fleets of Tesla driverless taxis Musk promised are likely still years or decades away as of September 2022.

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

    Assessment marcus-top-5-reasons-teslas-optimus-announcement-8258cd-a1.

  40. HeldBy this siteGary Marcus

    Held record marcus-top-5-reasons-teslas-optimus-announcement-6bbaee

    The anthropic screen reads the claim differently.

  41. Assessment publishedBy this siteGary Marcus

    Roboticists are likely to scoff at Tesla's Friday Optimus showing unless they see real evidence of progress.

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

    Assessment marcus-top-5-reasons-teslas-optimus-announcement-51cda1-a1.

  42. Assessment publishedBy this siteGary Marcus

    Robotics is always harder than people think it is.

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

    Assessment marcus-top-5-reasons-teslas-optimus-announcement-3df420-a1.

  43. HeldBy this siteGary Marcus

    Held record marcus-top-5-reasons-teslas-optimus-announcement-381b5b

    The anthropic screen reads the claim differently.

  44. HeldBy this siteGary Marcus

    Held record marcus-top-5-reasons-teslas-optimus-announcement-326978

    e6 is dated 2022-09, which is not inside the period described (2014-01-01 to 2022-09-29); it can only be context

  45. Assessment publishedBy this siteGary Marcus

    OpenAI's future models will likely be bigger and more expensive to train and operate, further reducing profits.

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

    Assessment marcus-top-5-reasons-openai-is-probably-fec989-a1.

  46. Assessment publishedBy this siteGary Marcus

    Free open-source competitors to OpenAI, though perhaps not yet as good, will suit some customers.

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

    Assessment marcus-top-5-reasons-openai-is-probably-fc8dfa-a2.

  47. Assessment publishedBy this siteGary Marcus

    Students using OpenAI products may quickly switch to free alternatives.

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

    Assessment marcus-top-5-reasons-openai-is-probably-fb271b-a1.

  48. HeldBy this siteGary Marcus

    Held record marcus-top-5-reasons-openai-is-probably-95dd08

    e3 is dated 2023, which is not inside the period described (2023-01-01 to 2023-09-27); it can only be context

  49. Assessment publishedBy this siteGary Marcus

    Currently, even small changes to an LLM's training data might require complete retraining.

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

    Assessment marcus-top-5-reasons-openai-is-probably-766b10-a1.

  50. HeldBy this siteGary Marcus

    Held record marcus-top-5-reasons-openai-is-probably-6b9946

    e1 is dated 2023, which is not inside the period described (2023-01-01 to 2023-09-27); it can only be context

  51. Assessment publishedBy this siteGary Marcus

    LLM hallucinations are a huge, unsolved problem.

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

    Assessment marcus-top-5-reasons-openai-is-probably-5ff0ce-a1.

  52. Assessment publishedBy this siteGary Marcus

    LLMs will likely never deliver on the promise of being close to AGI.

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

    Assessment marcus-top-5-reasons-openai-is-probably-480de6-a1.

  53. Assessment publishedBy this siteGary Marcus

    LLMs are not very good at following explicit verbal instructions.

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

    Assessment marcus-top-5-reasons-openai-is-probably-4050bb-a1.

  54. Assessment publishedBy this siteGary Marcus

    LLMs may be acceptable where 70% accuracy suffices but are unlikely to be acceptable for mission-critical applications.

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

    Assessment marcus-top-5-reasons-openai-is-probably-171adb-a1.

  55. HeldBy this siteGary Marcus

    Held record marcus-top-5-reasons-openai-is-probably-14e39b

    The anthropic screen reads the claim differently.

  57. Assessment publishedBy this siteGary Marcus

    Intelligences cleverer than human intelligence are possible and may eventually materialize.

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

    Assessment marcus-too-much-benefit-of-the-doubt-b27c4c-a1.

  58. Assessment publishedBy this siteGary Marcus

    Gary Marcus says the generative AI industry has long reminded him of Wile E. Coyote suspended in mid-air past the cliff edge, about to fall but not until he looks down.

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

    Assessment marcus-too-big-to-fail-cdac40-a1.

  60. HeldBy this siteGary Marcus

    Held record marcus-three-years-on-chatgpt-still-isnt-f4034a

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

  61. Assessment publishedBy this siteGary Marcus

    Large language models like ChatGPT will not get us to AGI.

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

    Assessment marcus-three-years-on-chatgpt-still-isnt-f0fb14-a1.

  62. Assessment publishedBy this siteGary Marcus

    No productivity studies of LLMs have shown anything close to a 10x gain; some show around 30%.

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

    Assessment marcus-three-years-on-chatgpt-still-isnt-ed689e-a1.

  63. HeldBy this siteGary Marcus

    Held record marcus-three-years-on-chatgpt-still-isnt-e8c19e

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

  64. Assessment publishedBy this siteGary Marcus

    Generative AI results are disappointing because the underlying technology is unreliable, which has been obvious from the start.

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

    Assessment marcus-three-years-on-chatgpt-still-isnt-e70f1d-a1.

  65. Assessment publishedBy this siteGary Marcus

    The MIT finding that only 5% of companies get a return on generative AI investment has held up.

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

    Assessment marcus-three-years-on-chatgpt-still-isnt-d75209-a1.

  67. Assessment publishedBy this siteGary Marcus

    The recurring core technical problems of LLMs are inherent to the technology and will not go away.

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

    Assessment marcus-three-years-on-chatgpt-still-isnt-c20947-a1.

  68. Assessment publishedBy this siteGary Marcus

    ChatGPT has had a good but not great run, and the field should move to something new.

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

    Assessment marcus-three-years-on-chatgpt-still-isnt-c03913-a1.

  69. Assessment publishedBy this siteGary Marcus

    If the economy goes down, ChatGPT will be at the center of the mess.

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

    Assessment marcus-three-years-on-chatgpt-still-isnt-bfa445-a2.

  70. Assessment publishedBy this siteGary Marcus

    The market may finally be recognizing AI's limits, as shown by falls in Nvidia, CoreWeave and Oracle stock.

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

    Assessment marcus-three-years-on-chatgpt-still-isnt-bec114-a1.

  71. HeldBy this siteGary Marcus

    Held record marcus-three-years-on-chatgpt-still-isnt-bbd8e9

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

  72. HeldBy this siteGary Marcus

    Held record marcus-three-years-on-chatgpt-still-isnt-b9adbc

    e1 is dated 2025, which is not inside the period described (2022-11-30 to 2025-11-29); it can only be context

  73. Assessment publishedBy this siteGary Marcus

    Return on generative AI investment for corporate end users is lower than many people must have imagined.

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

    Assessment marcus-three-years-on-chatgpt-still-isnt-b36452-a1.

  74. Assessment publishedBy this siteGary Marcus

    OpenAI has no moat, and its value would likely have fallen over the last month if it were publicly traded.

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

    Assessment marcus-three-years-on-chatgpt-still-isnt-afda3b-a1.

  75. Assessment publishedBy this siteGary Marcus

    The economy has become so wrapped up in generative AI and its promises that, by many accounts, it is in serious jeopardy.

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

    Assessment marcus-three-years-on-chatgpt-still-isnt-a43cda-a1.

  77. HeldBy this siteGary Marcus

    Held record marcus-three-years-on-chatgpt-still-isnt-9d75bb

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

  78. HeldBy this siteGary Marcus

    Held record marcus-three-years-on-chatgpt-still-isnt-9c214a

    e1 is dated 2025, which is not inside the period described (2022-11-29 to 2025-11-29); it can only be context

  79. HeldBy this siteGary Marcus

    Held record marcus-three-years-on-chatgpt-still-isnt-9711be

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

  82. Assessment publishedBy this siteGary Marcus

    Too much industry hype is repeated with too little skepticism, driving up market valuations on analyses that often feel to Marcus superficial.

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

    Assessment marcus-three-years-on-chatgpt-still-isnt-8f21bb-a1.

  83. HeldBy this siteGary Marcus

    Held record marcus-three-years-on-chatgpt-still-isnt-642605

    The openai screen reads the claim differently.

  84. HeldBy this siteGary Marcus

    Held record marcus-three-years-on-chatgpt-still-isnt-61dcb2

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

  85. Assessment publishedBy this siteGary Marcus

    ChatGPT has not lived up to expectations.

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

    Assessment marcus-three-years-on-chatgpt-still-isnt-5fcff7-a2.

  87. HeldBy this siteGary Marcus

    Held record marcus-three-years-on-chatgpt-still-isnt-5290a9

    The openai screen reads the claim differently.

  88. Assessment publishedBy this siteGary Marcus

    In many domains ChatGPT's answers may be roughly 80% correct and 20% absurd, with little task-specific engineering.

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

    Assessment marcus-three-years-on-chatgpt-still-isnt-50dcff-a1.

  89. Assessment publishedBy this siteGary Marcus

    OpenAI's $500 billion valuation rests on hopes rather than profits, and those hopes are slipping as usage declines and competitors catch up.

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

    Assessment marcus-three-years-on-chatgpt-still-isnt-508792-a2.

  90. Assessment publishedBy this siteGary Marcus

    LLMs are good at seeming plausible but much less good at getting facts right.

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

    Assessment marcus-three-years-on-chatgpt-still-isnt-4e84e5-a1.

  91. HeldBy this siteGary Marcus

    Held record marcus-three-years-on-chatgpt-still-isnt-47a4e7

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

  92. HeldBy this siteGary Marcus

    Held record marcus-three-years-on-chatgpt-still-isnt-441a3d

    The openai screen reads the claim differently.

  93. HeldBy this siteGary Marcus

    Held record marcus-three-years-on-chatgpt-still-isnt-3e24b5

    The final-text check failed: citations_correct.

  94. HeldBy this siteGary Marcus

    Held record marcus-three-years-on-chatgpt-still-isnt-38c2bf

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

  95. HeldBy this siteGary Marcus

    Held record marcus-three-years-on-chatgpt-still-isnt-38a416

    e3 is dated 2025, which is not inside the period described (2022-11-29 to 2025-11-29); it can only be context

  96. Assessment publishedBy this siteGary Marcus

    In some ways, such as trivia and writing python code, GPT-5 is more capable than most 7- or 8-year-olds.

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

    Assessment marcus-three-years-on-chatgpt-still-isnt-346d56-a1.

  97. Assessment publishedBy this siteGary Marcus

    The AI field should spend the next eight years on new approaches rather than risking the economy on more LLM scaling.

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

    Assessment marcus-three-years-on-chatgpt-still-isnt-3041fb-a1.

  98. Assessment publishedBy this siteGary Marcus

    LLMs will never be AGI and will never be much more than a commodity.

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

    Assessment marcus-three-years-on-chatgpt-still-isnt-2a2a81-a1.

  99. Assessment publishedBy this siteGary Marcus

    Unlike a seven-year-old, GPT lacks a rich model of the world and is just faking it.

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

    Assessment marcus-three-years-on-chatgpt-still-isnt-22462c-a1.

  100. Assessment publishedBy this siteGary Marcus

    If scaling is largely played out, a decline in demand for AI compute is sure to follow.

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

    Assessment marcus-three-years-on-chatgpt-still-isnt-1ad55b-a2.

  101. Assessment publishedBy this siteGary Marcus

    Marcus thinks a fair case can be made that ChatGPT is not what it has often been cracked up to be, and that it probably never will be.

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

    Assessment marcus-three-years-on-chatgpt-still-isnt-199aac-a1.

  102. Assessment publishedBy this siteGary Marcus

    Anyone assuming that everything is simultaneously rising exponentially and will continue to do so indefinitely is a fool.

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

    Assessment marcus-three-years-on-chatgpt-still-isnt-16f45b-a1.

  103. HeldBy this siteGary Marcus

    Held record marcus-three-years-on-chatgpt-still-isnt-1194ef

    The openai screen reads the claim differently.

  104. Assessment publishedBy this siteGary Marcus

    LLMs will continue to exist, get cheaper, and have some utility.

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

    Assessment marcus-three-years-on-chatgpt-still-isnt-05b853-a1.

  105. Assessment publishedBy this siteGary Marcus

    Musk may be making the OpenAI trial too much about himself.

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

    Assessment marcus-three-thoughts-on-the-musk-openai-a6dc23-a1.

  106. HeldBy this siteGary Marcus

    Held record marcus-three-thoughts-on-the-musk-openai-82df88

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

  107. HeldBy this siteGary Marcus

    Held record marcus-three-thoughts-on-the-musk-openai-537268

    e1 is decisive for a financial claim but states no metric scope; e2 is decisive for a financial claim but states no metric scope

  108. Assessment publishedBy this siteGary Marcus

    Musk's lawyers should focus the trial on OpenAI's pattern of behavior rather than Musk's contributions.

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

    Assessment marcus-three-thoughts-on-the-musk-openai-4c8a9c-a1.

  109. Assessment publishedBy this siteGary Marcus

    OpenAI's transition to for-profit should be unwound so the company is forced to benefit humanity.

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

    Assessment marcus-three-thoughts-on-the-musk-openai-0a73fa-a1.

  110. HeldBy this siteGary Marcus

    Held record marcus-three-reasons-to-think-that-the-claude-cd7207

    The final-text check failed: citations_correct.

  111. Assessment publishedBy this siteGary Marcus

    Mythos is incrementally better than recent previous models and certainly not an off-the-chart breakthrough.

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

    Assessment marcus-three-reasons-to-think-that-the-claude-be9090-a1.

  112. Assessment publishedBy this siteGary Marcus

    Marcus feels that, to a certain degree, we were played by the Mythos announcement.

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

    Assessment marcus-three-reasons-to-think-that-the-claude-b2650d-a1.

  113. HeldBy this siteGary Marcus

    Held record marcus-three-reasons-to-think-that-the-claude-9d4ba5

    The anthropic screen reads the claim differently.

  114. HeldBy this siteGary Marcus

    Held record marcus-three-reasons-to-think-that-the-claude-7227c2

    The anthropic screen reads the claim differently.

  115. Assessment publishedBy this siteGary Marcus

    Mythos is more sophisticated than open-weight models but perhaps not as far ahead as it was portrayed.

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

    Assessment marcus-three-reasons-to-think-that-the-claude-41affd-a1.

  116. Assessment publishedBy this siteGary Marcus

    The Mythos demo shows we need to get our regulatory and technical house in order.

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

    Assessment marcus-three-reasons-to-think-that-the-claude-2f8697-a1.

  117. Assessment publishedBy this siteGary Marcus

    The Mythos demo was not the immediate threat the media and public were led to believe.

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

    Assessment marcus-three-reasons-to-think-that-the-claude-2f8697-2-a1.

  118. HeldBy this siteGary Marcus

    Held record marcus-three-reactions-to-anthropicss-latest-e688b2

    The anthropic screen reads the claim differently.

  119. Assessment publishedBy this siteGary Marcus

    The AI security incidents happened because pattern-matching AI systems with no real understanding were allowed to roam the internet freely.

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

    Assessment marcus-three-reactions-to-anthropicss-latest-7167bd-a2.

  120. Assessment publishedBy this siteGary Marcus

    Society as a whole is out of its depth in dealing with AI and cybersecurity risks.

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

    Assessment marcus-three-reactions-to-anthropicss-latest-63593c-a1.

  122. Assessment publishedBy this siteGary Marcus

    We, as a society, currently have no idea how to control AI and are just sort of hoping for the best.

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

    Assessment marcus-three-reactions-to-anthropicss-latest-4f7976-2-a1.

  123. Assessment publishedBy this siteGary Marcus

    The companies presented as leaders in AI and AI safety, such as Anthropic and OpenAI, are clearly out of their depth on AI and cybersecurity.

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

    Assessment marcus-three-reactions-to-anthropicss-latest-2f38a6-a1.

  124. HeldBy this siteGary Marcus

    Held record marcus-three-ideas-from-linguistics-that-f9030d

    The openai review reads the claim differently.

  125. Assessment publishedBy this siteGary Marcus

    Neglecting reference, cognitive models and compositionality makes it hard for LLMs to avoid repeating bias and toxic speech.

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

    Assessment marcus-three-ideas-from-linguistics-that-e640a1-a1.

  127. Assessment publishedBy this siteGary Marcus

    Large language models sidestep the problems of reference and shared understanding, and this puts them at risk.

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

    Assessment marcus-three-ideas-from-linguistics-that-d34857-a1.

  129. Assessment publishedBy this siteGary Marcus

    A system that only predicts the next word will inevitably have only superficial understanding.

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

    Assessment marcus-three-ideas-from-linguistics-that-ca08ad-a1.

  130. Assessment publishedBy this siteGary Marcus

    A good language understanding or language production system almost certainly needs some large-scale learning as a vital component.

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

    Assessment marcus-three-ideas-from-linguistics-that-c4f7da-a2.

  132. HeldBy this siteGary Marcus

    Held record marcus-three-ideas-from-linguistics-that-bdbfbb

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

  133. Assessment publishedBy this siteGary Marcus

    Given comparative prompts such as 'more cucumbers than tomatoes', DALL-E Mini usually leaves out the tomatoes entirely.

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

    Assessment marcus-three-ideas-from-linguistics-that-b6010a-a2.

  134. Assessment publishedBy this siteGary Marcus

    As a downstream consequence of their lack of reference and cognitive models, large language models have no grasp whatsoever on truth.

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

    Assessment marcus-three-ideas-from-linguistics-that-b26843-a1.

  136. Assessment publishedBy this siteGary Marcus

    The word sequences that large language models produce have no connection to the external world, so they lack reference.

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

    Assessment marcus-three-ideas-from-linguistics-that-aa740a-a2.

  137. Assessment publishedBy this siteGary Marcus

    Current LLMs lack the human ability to choose whether to interpret language compositionally.

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

    Assessment marcus-three-ideas-from-linguistics-that-a607b6-a1.

  138. HeldBy this siteGary Marcus

    Held record marcus-three-ideas-from-linguistics-that-9a6cc3

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

  139. Assessment publishedBy this siteGary Marcus

    Neglecting reference, cognitive models and compositionality makes it hard for LLMs to tell truth from falsehood.

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

    Assessment marcus-three-ideas-from-linguistics-that-8c1485-a2.

  140. HeldBy this siteGary Marcus

    Held record marcus-three-ideas-from-linguistics-that-83e6ae

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

  141. Assessment publishedBy this siteGary Marcus

    Current AI approaches still struggle with noncanonical word orders such as 'the mat is on the cat'.

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

    Assessment marcus-three-ideas-from-linguistics-that-7845de-a2.

  143. HeldBy this siteGary Marcus

    Held record marcus-three-ideas-from-linguistics-that-638e77

    The anthropic review does not accept e1 as supports.

  144. HeldBy this siteGary Marcus

    Held record marcus-three-ideas-from-linguistics-that-612dd3

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

  145. Assessment publishedBy this siteGary Marcus

    LLMs like GPT write well-formed prose but do not produce interpretable representations of the structured relationships between the parts of sentences.

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

    Assessment marcus-three-ideas-from-linguistics-that-5fa9f3-a1.

  146. Assessment publishedBy this siteGary Marcus

    Since 2019 large language models have done better on benchmarks, but something is still missing from them.

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

    Assessment marcus-three-ideas-from-linguistics-that-5c8654-a1.

  147. Assessment publishedBy this siteGary Marcus

    Getting to real language understanding may require another revolution in AI.

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

    Assessment marcus-three-ideas-from-linguistics-that-478836-a1.

  148. Assessment publishedBy this siteGary Marcus

    Foundation models such as GPT-3 and BERT sidestep compositionality.

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

    Assessment marcus-three-ideas-from-linguistics-that-42e048-a2.

  149. Assessment publishedBy this siteGary Marcus

    Cross-modal models like Gato are worth investigating but are unlikely by themselves to solve the lack of reference.

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

    Assessment marcus-three-ideas-from-linguistics-that-40ae30-a1.

  150. Assessment publishedBy this siteGary Marcus

    Systems like GPT-3 likely have trouble with compositional phrases such as 'the second blue ball'.

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

    Assessment marcus-three-ideas-from-linguistics-that-3dfd24-a1.

  151. HeldBy this siteGary Marcus

    Held record marcus-three-ideas-from-linguistics-that-39b53d

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

  152. HeldBy this siteGary Marcus

    Held record marcus-three-ideas-from-linguistics-that-36fd64

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

  153. Assessment publishedBy this siteGary Marcus

    DALL-E does poorly on sentences with reflexive pronouns such as 'The man is painting a picture of himself'.

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

    Assessment marcus-three-ideas-from-linguistics-that-30f93e-a1.

  154. Assessment publishedBy this siteGary Marcus

    GPT-3's good word predictions come more from its huge training corpus than from any real mapping of sentences onto reference and cognitive models.

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

    Assessment marcus-three-ideas-from-linguistics-that-20ca0c-a1.

  157. Assessment publishedBy this siteGary Marcus

    Neglecting reference, cognitive models and compositionality causes serious problems in large language models.

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

    Assessment marcus-three-ideas-from-linguistics-that-0eeecb-a1.

  158. HeldBy this siteGary Marcus

    Held record marcus-three-ideas-from-linguistics-that-06d412

    The openai review reads the claim differently.

  159. HeldBy this siteGary Marcus

    Held record marcus-three-ideas-from-linguistics-that-061bee

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

  160. HeldBy this siteGary Marcus

    Held record marcus-three-ideas-from-linguistics-that-03f57e

    e7 is dated 2022-06, which is not inside the period described (2022-01-01 to 2022-06-22); it can only be context

  161. HeldBy this siteGary Marcus

    Held record marcus-three-ideas-from-linguistics-that-0344c3

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

  162. Assessment publishedBy this siteGary Marcus

    AI will not get past its 80:20, sometimes-works reliability unless the paradigm changes.

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

    Assessment marcus-those-claiming-were-mere-months-away-3336cc-a1.

  163. Assessment publishedBy this siteGary Marcus

    As of April 2025, the paradigm change AI needs has not happened.

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

    Assessment marcus-those-claiming-were-mere-months-away-1bc819-a1.

  164. Assessment publishedBy this siteGary Marcus

    OpenAI's LLMs make lousy databases because hallucinations amount to outright failures for a database.

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

    Assessment marcus-this-week-in-ai-doublespeak-f02594-a1.

  165. Assessment publishedBy this siteGary Marcus

    GPT-4 frequently fails to reason, if reasoning means reaching valid conclusions from known facts.

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

    Assessment marcus-this-week-in-ai-doublespeak-d79f2e-a1.

  166. HeldBy this siteGary Marcus

    Held record marcus-this-week-in-ai-doublespeak-7f68c2

    The final-text check failed: citations_correct.

  167. Assessment publishedBy this siteGary Marcus

    The free-association 'reasoning' LLMs do is not really reasoning, even when constrained by a giant database.

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

    Assessment marcus-this-week-in-ai-doublespeak-2a6bad-a1.

  168. HeldBy this siteGary Marcus

    Held record marcus-this-week-in-ai-doublespeak-01cf93

    The anthropic screen reads the claim differently.

  170. HeldBy this siteGary Marcus

    Held record marcus-this-one-weird-trick-might-cost-your-c5c989-2

    The openai screen reads the claim differently.

  172. Assessment publishedBy this siteGary Marcus

    When an input is not directly in the training set but only superficially resembles earlier examples, machine learning systems may fail.

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

    Assessment marcus-this-one-important-fact-about-current-df9f8c-a1.

  173. HeldBy this siteGary Marcus

    Held record marcus-this-one-important-fact-about-current-d4d5f8

    The final-text check failed: rationale_supported_by_evidence; overclaims: On reflection, this forecast lacks a stable outcome to check..

  175. Assessment publishedBy this siteGary Marcus

    Marcus asserts emphatically that no one who realizes we do not yet have a general solution to the outlier problem can think we are close to AGI.

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

    Assessment marcus-this-one-important-fact-about-current-bd5370-a1.

  177. HeldBy this siteGary Marcus

    Held record marcus-this-one-important-fact-about-current-7cfe28

    The anthropic review does not accept e3 as supports.

  179. HeldBy this siteGary Marcus

    Held record marcus-things-are-so-desperate-at-openai-bf9d3b

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

  180. HeldBy this siteGary Marcus

    Held record marcus-things-are-so-desperate-at-openai-6ece11

    The anthropic screen reads the claim differently.

  181. HeldBy this siteGary Marcus

    Held record marcus-things-are-so-desperate-at-openai-14c5bc

    The openai screen reads the claim differently.

  182. HeldBy this siteGary Marcus

    Held record marcus-things-are-so-desperate-at-openai-0770fc

    The anthropic screen reads the claim differently.

  183. HeldBy this siteGary Marcus

    Held record marcus-things-are-about-to-get-a-lot-worse-fab65d

    The final-text check failed: citations_correct.

  184. Assessment publishedBy this siteGary Marcus

    OpenAI faces further risks.

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

    Assessment marcus-things-are-about-to-get-a-lot-worse-e7c89a-a1.

  185. Assessment publishedBy this siteGary Marcus

    OpenAI's image generator can reproduce source images verbatim or near-verbatim.

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

    Assessment marcus-things-are-about-to-get-a-lot-worse-de064b-a1.

  186. Assessment publishedBy this siteGary Marcus

    Experimental findings on copyright reproduction pose serious challenges for generative AI.

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

    Assessment marcus-things-are-about-to-get-a-lot-worse-c357cc-a1.

  187. Assessment publishedBy this siteGary Marcus

    Users of generative AI may not know whether a given image they produce is infringing.

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

    Assessment marcus-things-are-about-to-get-a-lot-worse-8ae32e-a1.

  188. HeldBy this siteGary Marcus

    Held record marcus-things-are-about-to-get-a-lot-worse-838708

    The anthropic screen reads the claim differently.

  189. Assessment publishedBy this siteGary Marcus

    Generative AI systems such as DALL-E and ChatGPT were trained on copyrighted materials.

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

    Assessment marcus-things-are-about-to-get-a-lot-worse-736dfd-a1.

  190. Assessment publishedBy this siteGary Marcus

    Generative AI systems should give users a manifest of sources.

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

    Assessment marcus-things-are-about-to-get-a-lot-worse-679273-a1.

  191. Assessment publishedBy this siteGary Marcus

    The copyright, provenance, and transparency problems of generative AI probably cannot be easily fixed.

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

    Assessment marcus-things-are-about-to-get-a-lot-worse-5d7167-a1.

  192. Assessment publishedBy this siteGary Marcus

    DALL-E can produce infringing images even when the user does not intend to infringe or name any character or film.

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

    Assessment marcus-things-are-about-to-get-a-lot-worse-1de22d-a1.

  193. Assessment publishedBy this siteGary Marcus

    The New York Times lawsuit against OpenAI is likely the first of many copyright lawsuits.

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

    Assessment marcus-things-are-about-to-get-a-lot-worse-049c5d-a1.

  194. Assessment publishedBy this siteGary Marcus

    People should not rely on large language model systems.

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

    Assessment marcus-there-must-be-some-misunderstanding-abefb6-a1.

  195. Assessment publishedBy this siteGary Marcus

    The AI field should consider approaches to AI other than scaling large language models.

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

    Assessment marcus-there-must-be-some-misunderstanding-922937-a1.

  196. Assessment publishedBy this siteGary Marcus

    Compositionality and negation remain problems for large language models.

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

    Assessment marcus-there-must-be-some-misunderstanding-8ebd86-a1.

  197. Assessment publishedBy this siteGary Marcus

    Scaling has not been enough to solve the problems of understanding, compositionality, negation, and reliability in LLMs.

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

    Assessment marcus-there-must-be-some-misunderstanding-8e10c2-a1.

  198. Assessment publishedBy this siteGary Marcus

    Scaling will not be enough to solve the problems of understanding, compositionality, negation, and reliability in LLMs.

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

    Assessment marcus-there-must-be-some-misunderstanding-8e10c2-2-a1.

  199. Assessment publishedBy this siteGary Marcus

    Reliability remains a problem for large language models.

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

    Assessment marcus-there-must-be-some-misunderstanding-781f39-a1.

  200. Assessment publishedBy this siteGary Marcus

    Large language models do not deeply understand their inputs.

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

    Assessment marcus-there-must-be-some-misunderstanding-28e0af-a1.