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

Changes, page 26 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 26 of 37: changes 5001 to 5200 of 7383, newest first. Dates are when the change was recorded.

  1. Assessment publishedBy this siteGary Marcus

    The climate impact of AI models should be considered, especially as they grow.

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

    Assessment marcus-happy-1st-birthday-gpt-4-597563-a1.

  2. Assessment publishedBy this siteGary Marcus

    There will be price wars among LLM providers.

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

    Assessment marcus-happy-1st-birthday-gpt-4-56060e-a1.

  3. Assessment publishedBy this siteGary Marcus

    Hallucinations, instability and occasional nonsensical errors remain unsolved in GPT-4 class LLMs.

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

    Assessment marcus-happy-1st-birthday-gpt-4-2cc74c-a1.

  4. Assessment publishedBy this siteGary Marcus

    The more deeply-trained version of Meta's LLAMA-3, still being trained, may well be in the general vicinity of GPT-4.

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

    Assessment marcus-groupthink-versus-tulips-what-will-de6769-a1.

  5. HeldBy this siteGary Marcus

    Held record marcus-groupthink-versus-tulips-what-will-d6bcee

    The final-text check failed: citations_correct; overclaims: e1 public excerpt "GPT-4 judge favoring GPT-4 outputs." drops the source's hedge "can be due to" and its subject, so a possible bias reads as an established fact..

  6. Assessment publishedBy this siteGary Marcus

    Historians will be puzzled by the AI industry's groupthink.

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

    Assessment marcus-groupthink-versus-tulips-what-will-d559e0-a1.

  7. Assessment publishedBy this siteGary Marcus

    LLMs are becoming a commodity and price wars among providers will be intense.

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

    Assessment marcus-groupthink-versus-tulips-what-will-d50b17-a1.

  8. Assessment publishedBy this siteGary Marcus

    2024 could be remembered as the peak of an AI tulip-mania.

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

    Assessment marcus-groupthink-versus-tulips-what-will-c111e2-a1.

  9. Assessment publishedBy this siteGary Marcus

    The small moat between LLM companies will further diminish.

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

    Assessment marcus-groupthink-versus-tulips-what-will-99ab51-a1.

  10. Assessment publishedBy this siteGary Marcus

    Meta's free near-GPT-4 model threatens the business models of OpenAI, Anthropic, Mistral and similar companies.

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

    Assessment marcus-groupthink-versus-tulips-what-will-5a3395-a1.

  11. Assessment publishedBy this siteGary Marcus

    The AI bubble deflation may have begun as of April 2024.

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

    Assessment marcus-groupthink-versus-tulips-what-will-44e309-a1.

  12. Assessment publishedBy this siteGary Marcus

    An increase in realism about AI might lead to a renaissance of new approaches to AI.

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

    Assessment marcus-groupthink-versus-tulips-what-will-43a618-a1.

  13. HeldBy this siteGary Marcus

    Held record marcus-grok-3-hot-take-eee9ff

    The openai screen reads the claim differently.

  14. Assessment publishedBy this siteGary Marcus

    For now, Grok 3 gives Sam Altman (OpenAI) little reason to worry.

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

    Assessment marcus-grok-3-hot-take-d68627-a1.

  15. HeldBy this siteGary Marcus

    Held record marcus-grok-3-hot-take-a35a63

    The anthropic screen doubts the attribution.

  16. Assessment publishedBy this siteGary Marcus

    Marcus did not notice anything truly original in the Grok 3 demo.

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

    Assessment marcus-grok-3-hot-take-57c74c-a1.

  17. HeldBy this siteGary Marcus

    Held record marcus-grok-3-hot-take-31577b-3

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

  18. Assessment publishedBy this siteGary Marcus

    Grok 3 is probably not smarter than the full, unreleased version of OpenAI's o3.

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

    Assessment marcus-grok-3-hot-take-2b2e50-a1.

  19. Assessment publishedBy this siteGary Marcus

    Grok 3 is neither a game changer nor a major leap forward.

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

    Assessment marcus-grok-3-hot-take-15db12-a1.

  20. HeldBy this siteGary Marcus

    Held record marcus-grok-3-beta-in-shambles-f66fbf

    e1 is dated 2025-02, which is not inside the period described (2024-12-01 to 2025-02-19); it can only be context

  21. HeldBy this siteGary Marcus

    Held record marcus-grok-3-beta-in-shambles-db247f

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

  22. HeldBy this siteGary Marcus

    Held record marcus-grok-3-beta-in-shambles-d638d2

    e1 is dated 2025-02, which is not inside the period described (2022-11 to 2025-02-19); it can only be context; e3 is dated 2025-02, which is not inside the period described (2022-11 to 2025-02-19); it can only be context

  23. Assessment publishedBy this siteGary Marcus

    An AI system like Grok 3 should give today's date consistently, not different answers on different tries.

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

    Assessment marcus-grok-3-beta-in-shambles-ae7667-a1.

  24. HeldBy this siteGary Marcus

    Held record marcus-grok-3-beta-in-shambles-8156e9

    The anthropic screen reads the claim differently.

  25. HeldBy this siteGary Marcus

    Held record marcus-grok-3-beta-in-shambles-6259e6

    e1 measures something else (company xAI is not Leading AI labs (e.g., OpenAI, xAI, Anthropic)); it can only be context unless it is derived with stated inputs; e2 measures something else (company OpenAI is not Leading AI labs (e.g., OpenAI, xAI, Anthropic)); it can only be context unless it is derived with stated inputs

  26. HeldBy this siteGary Marcus

    Held record marcus-grok-3-beta-in-shambles-606bd4

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

  27. HeldBy this siteGary Marcus

    Held record marcus-grok-3-beta-in-shambles-59c4ad

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

  28. HeldBy this siteGary Marcus

    Held record marcus-grok-3-beta-in-shambles-4d1a48

    The openai screen reads the claim differently.

  29. Assessment publishedBy this siteGary Marcus

    LLMs are not close to AGI, and people are increasingly realizing this.

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

    Assessment marcus-gpt-5-overdue-overhyped-and-underwhelming-f13941-a1.

  30. Assessment publishedBy this siteGary Marcus

    Other generative AI systems such as Grok are not faring much better than GPT-5.

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

    Assessment marcus-gpt-5-overdue-overhyped-and-underwhelming-efa4ab-a2.

  31. Assessment publishedBy this siteGary Marcus

    Sam Altman may no longer be the right CEO for OpenAI.

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

    Assessment marcus-gpt-5-overdue-overhyped-and-underwhelming-e1ff35-a1.

  32. HeldBy this siteGary Marcus

    Held record marcus-gpt-5-overdue-overhyped-and-underwhelming-dcdbc6

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

  33. HeldBy this siteGary Marcus

    Held record marcus-gpt-5-overdue-overhyped-and-underwhelming-d8221e

    The anthropic screen doubts the attribution.

  34. Assessment publishedBy this siteGary Marcus

    GPT-5 still has the longstanding problems Marcus has identified, such as chess rule-following and world models.

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

    Assessment marcus-gpt-5-overdue-overhyped-and-underwhelming-ca66af-a1.

  36. Assessment publishedBy this siteGary Marcus

    Over the last several years, we have been fed a steady diet of bullshit about AI.

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

    Assessment marcus-gpt-5-overdue-overhyped-and-underwhelming-bbf4ea-a1.

  37. HeldBy this siteGary Marcus

    Held record marcus-gpt-5-overdue-overhyped-and-underwhelming-baf55b

    The anthropic review does not accept e9 as supports.

  38. HeldBy this siteGary Marcus

    Held record marcus-gpt-5-overdue-overhyped-and-underwhelming-b685c5

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

  39. HeldBy this siteGary Marcus

    Held record marcus-gpt-5-overdue-overhyped-and-underwhelming-b517e6

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

  40. HeldBy this siteGary Marcus

    Held record marcus-gpt-5-overdue-overhyped-and-underwhelming-ace526

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

  41. HeldBy this siteGary Marcus

    Held record marcus-gpt-5-overdue-overhyped-and-underwhelming-ac34f5

    The final-text check failed: proposition_faithful, citations_correct; overclaims: e2 explanation: "Reported AIME and MMMU results also remain below perfect accuracy...".

  42. Assessment publishedBy this siteGary Marcus

    No intellectually honest person should still believe pure scaling will reach AGI.

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

    Assessment marcus-gpt-5-overdue-overhyped-and-underwhelming-9ab158-a1.

  43. Assessment publishedBy this siteGary Marcus

    GPT-5 is not a terrible model.

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

    Assessment marcus-gpt-5-overdue-overhyped-and-underwhelming-941b61-a1.

  44. HeldBy this siteGary Marcus

    Held record marcus-gpt-5-overdue-overhyped-and-underwhelming-9062db

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

  45. Assessment publishedBy this siteGary Marcus

    GPT-5 shows no improvement on the parts-and-wholes problem in generated images.

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

    Assessment marcus-gpt-5-overdue-overhyped-and-underwhelming-876436-a2.

  46. Assessment publishedBy this siteGary Marcus

    The field should give neurosymbolic AI with explicit world models a chance.

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

    Assessment marcus-gpt-5-overdue-overhyped-and-underwhelming-840f57-a1.

  47. Assessment publishedBy this siteGary Marcus

    Even some tech bros are waking up to the reality that 'AGI in 2027' was marketing, not reality.

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

    Assessment marcus-gpt-5-overdue-overhyped-and-underwhelming-8246a9-a1.

  48. HeldBy this siteGary Marcus

    Held record marcus-gpt-5-overdue-overhyped-and-underwhelming-7fefe8

    The anthropic review reads the claim differently.

  49. Assessment publishedBy this siteGary Marcus

    Altman's overhyping caused the GPT-5 backlash; without it people might have accepted an incremental update.

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

    Assessment marcus-gpt-5-overdue-overhyped-and-underwhelming-7ec962-a1.

  50. HeldBy this siteGary Marcus

    Held record marcus-gpt-5-overdue-overhyped-and-underwhelming-79bdf2

    The openai review does not accept e3 as supports.

  51. HeldBy this siteGary Marcus

    Held record marcus-gpt-5-overdue-overhyped-and-underwhelming-7768d1

    The final-text check failed: rationale_supported_by_evidence; overclaims: Both parts hold for launch week in August 2025. | Either way, the gap is small, which fits 'barely better'..

  52. Assessment publishedBy this siteGary Marcus

    GPT-5 is merely an incremental advance, not a radical leap.

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

    Assessment marcus-gpt-5-overdue-overhyped-and-underwhelming-72c33c-a1.

  53. Assessment publishedBy this siteGary Marcus

    After half a trillion dollars spent on scaling, the field should move on from the pure-scaling hypothesis.

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

    Assessment marcus-gpt-5-overdue-overhyped-and-underwhelming-5fbf2d-a1.

  55. Assessment publishedBy this siteGary Marcus

    A few days after launch, hardly anybody believes Altman's claims that GPT-5 is like a PhD-level expert.

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

    Assessment marcus-gpt-5-overdue-overhyped-and-underwhelming-4c54f4-a2.

  56. HeldBy this siteGary Marcus

    Held record marcus-gpt-5-overdue-overhyped-and-underwhelming-440895

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

  57. Assessment publishedBy this siteGary Marcus

    Many of OpenAI's best people have left the company.

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

    Assessment marcus-gpt-5-overdue-overhyped-and-underwhelming-338808-a2.

  58. Assessment publishedBy this siteGary Marcus

    Driverless cars are available in only a couple percent of the world's cities.

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

    Assessment marcus-gpt-5-overdue-overhyped-and-underwhelming-2ca6a1-a1.

  59. HeldBy this siteGary Marcus

    Held record marcus-gpt-5-overdue-overhyped-and-underwhelming-2b9e05

    The final-text check failed: rationale_supported_by_evidence, citations_correct; overclaims: On August 7, 2025, Grok 4 scored 15.9% on ARC-AGI-2 against GPT-5's 9.9% [e1]. | e2 explanation: "...and highlighted leading long-context reasoning.".

  60. Assessment publishedBy this siteGary Marcus

    LLMs' core weakness is their inability to generalize broadly, as confirmed by an ASU study.

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

    Assessment marcus-gpt-5-overdue-overhyped-and-underwhelming-2647a1-a1.

  61. Assessment publishedBy this siteGary Marcus

    Sam Altman's reputation should by now be completely burned given his AGI claims.

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

    Assessment marcus-gpt-5-overdue-overhyped-and-underwhelming-2220ee-a1.

  62. Assessment publishedBy this siteGary Marcus

    In a rational world, OpenAI's valuation would certainly take a hit.

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

    Assessment marcus-gpt-5-overdue-overhyped-and-underwhelming-14b0b0-a1.

  63. Assessment publishedBy this siteGary Marcus

    GPT-5 is unlikely to remain ahead of competing models for more than a couple of months.

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

    Assessment marcus-gpt-5-overdue-overhyped-and-underwhelming-1483a6-a1.

  65. Assessment publishedBy this siteGary Marcus

    OpenAI has probably started training GPT-5 three or four times, with GPT-4 Turbo and GPT-4o perhaps being failed attempts not worthy of the name.

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

    Assessment marcus-gpt-5-now-arriving-gate-8-gate-9-a27b08-a1.

  66. Assessment publishedBy this siteGary Marcus

    Business Insider's sources on an imminent GPT-5 release appear to have been confused or overstated their knowledge.

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

    Assessment marcus-gpt-5-now-arriving-gate-8-gate-9-0dead7-a1.

  67. Assessment publishedBy this siteGary Marcus

    Gary Marcus thinks that perhaps scaling is not in fact all you need.

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

    Assessment marcus-gpt-5-hot-take-eee839-a1.

  68. Assessment publishedBy this siteGary Marcus

    AGI arriving by 2027 looks increasingly unlikely.

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

    Assessment marcus-gpt-5-hot-take-832f3d-a1.

  69. HeldBy this siteGary Marcus

    Held record marcus-gpt-5-hot-take-6b72ff

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

  70. Assessment publishedBy this siteGary Marcus

    Although GPT-5 pricing is good, OpenAI's profits may continue to be elusive.

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

    Assessment marcus-gpt-5-hot-take-25d34f-a1.

  71. Assessment publishedBy this siteGary Marcus

    GPT-5 will not mark the singularity.

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

    Assessment marcus-gpt-5-and-irrational-exuberance-fb1cb5-a2.

  72. HeldBy this siteGary Marcus

    Held record marcus-gpt-5-and-irrational-exuberance-ee7fd5

    hedged language is not scored, even on a plain reading

  74. HeldBy this siteGary Marcus

    Held record marcus-gpt-5-and-irrational-exuberance-e145fa

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

  75. Assessment publishedBy this siteGary Marcus

    Very little of what GPT-4 does is fully replicable; its summarization sometimes works and sometimes fails.

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

    Assessment marcus-gpt-5-and-irrational-exuberance-da17ab-a1.

  77. Assessment publishedBy this siteGary Marcus

    GPT-4 cannot play a good game of chess and is about as good as 1978 chess computers.

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

    Assessment marcus-gpt-5-and-irrational-exuberance-d99615-a2.

  78. Assessment publishedBy this siteGary Marcus

    Marcus says none of his predictions about GPT-4 proved incorrect, and every flaw he predicted would persist in GPT-4 did persist.

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

    Assessment marcus-gpt-5-and-irrational-exuberance-d8c80f-a1.

  79. HeldBy this siteGary Marcus

    Held record marcus-gpt-5-and-irrational-exuberance-c7debd

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

  80. Assessment publishedBy this siteGary Marcus

    GPT cannot filter out falsehood and creates and spreads falsehoods it fails to verify, making it dangerous.

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

    Assessment marcus-gpt-5-and-irrational-exuberance-c5623d-a2.

  81. HeldBy this siteGary Marcus

    Held record marcus-gpt-5-and-irrational-exuberance-c12267

    The anthropic screen reads the claim differently.

  82. HeldBy this siteGary Marcus

    Held record marcus-gpt-5-and-irrational-exuberance-bb3ad3

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

  83. Assessment publishedBy this siteGary Marcus

    Marcus says he is afraid not because GPT is too smart but because it is too stupid: too dumb to comprehend an op-ed and too dumb to keep its mouth shut.

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

    Assessment marcus-gpt-5-and-irrational-exuberance-b55123-a1.

  84. Assessment publishedBy this siteGary Marcus

    GPT-5 will not give China, the US, or anyone else interstellar starships, and it will not give anyone an insuperable military advantage.

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

    Assessment marcus-gpt-5-and-irrational-exuberance-b288bf-a1.

  85. HeldBy this siteGary Marcus

    Held record marcus-gpt-5-and-irrational-exuberance-a052cc

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

  86. HeldBy this siteGary Marcus

    Held record marcus-gpt-5-and-irrational-exuberance-9da352

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

  87. HeldBy this siteGary Marcus

    Held record marcus-gpt-5-and-irrational-exuberance-763ec4

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

  88. Assessment publishedBy this siteGary Marcus

    The West should be concerned about China catching up over the next decade but should not base AI policy on fantasies about GPT-5.

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

    Assessment marcus-gpt-5-and-irrational-exuberance-6477c2-a1.

  90. HeldBy this siteGary Marcus

    Held record marcus-gpt-5-and-irrational-exuberance-3c2796

    The anthropic screen reads the claim differently.

  91. Assessment publishedBy this siteGary Marcus

    GPT-5 will be better than GPT-4 at creating convincing-sounding prose.

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

    Assessment marcus-gpt-5-and-irrational-exuberance-32fe5d-a1.

  92. Assessment publishedBy this siteGary Marcus

    The claim that large language models have mastered theory of mind is a myth.

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

    Assessment marcus-gpt-5-and-irrational-exuberance-310b61-a2.

  93. Assessment publishedBy this siteGary Marcus

    Essentially none of the advanced cognitive abilities attributed to GPT-4 hold up under scrutiny.

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

    Assessment marcus-gpt-5-and-irrational-exuberance-28c621-a1.

  94. Assessment publishedBy this siteGary Marcus

    People are wildly overattributing intelligence to GPT-4 and extrapolating that into fantasies about GPT-5.

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

    Assessment marcus-gpt-5-and-irrational-exuberance-27d0c5-a1.

  95. Assessment publishedBy this siteGary Marcus

    For GPT models on their own, making things up has been and always will be their forte, and they cannot check facts.

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

    Assessment marcus-gpt-5-and-irrational-exuberance-187146-a1.

  96. Assessment publishedBy this siteGary Marcus

    We should worry a great deal about how hard-to-steer LLMs are changing the world, but not about an immediate transition to a singularity.

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

    Assessment marcus-gpt-5-and-irrational-exuberance-0dc277-a1.

  97. Assessment publishedBy this siteGary Marcus

    GPT-4's release is a step backwards for science because it sets a new precedent for pretending to be scientific while revealing absolutely nothing.

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

    Assessment marcus-gpt-4s-successes-and-gpt-4s-failures-f23c67-a1.

  98. HeldBy this siteGary Marcus

    Held record marcus-gpt-4s-successes-and-gpt-4s-failures-ef6e9f

    The final-text check failed: citations_correct.

  100. Assessment publishedBy this siteGary Marcus

    GPT-4's alignment remains shaky and it cannot be used reliably to guide robots or scientific discovery.

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

    Assessment marcus-gpt-4s-successes-and-gpt-4s-failures-d8a265-a1.

  101. Assessment publishedBy this siteGary Marcus

    GPT-4 has vision capability and performs astonishingly well on standardized tests such as the LSAT, GRE, and SAT.

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

    Assessment marcus-gpt-4s-successes-and-gpt-4s-failures-ce06c5-a2.

  102. HeldBy this siteGary Marcus

    Held record marcus-gpt-4s-successes-and-gpt-4s-failures-caabeb

    The openai screen reads the claim differently.

  104. Assessment publishedBy this siteGary Marcus

    GPT-4 is a big step for hype but not necessarily for science, AGI, or humanity.

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

    Assessment marcus-gpt-4s-successes-and-gpt-4s-failures-b380f5-a1.

  106. Assessment publishedBy this siteGary Marcus

    On its own, GPT-4 still requires frequent massive retraining to stay current with the news.

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

    Assessment marcus-gpt-4s-successes-and-gpt-4s-failures-a647e7-a1.

  107. Assessment publishedBy this siteGary Marcus

    As of March 2023, massive scaling has not led to a revolution in AI.

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

    Assessment marcus-gpt-4s-successes-and-gpt-4s-failures-87d0e3-a1.

  109. Assessment publishedBy this siteGary Marcus

    GPT-4 does not solve any of the core truthfulness and reliability problems identified in Marcus's March 2022 essay.

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

    Assessment marcus-gpt-4s-successes-and-gpt-4s-failures-73f773-a1.

  110. HeldBy this siteGary Marcus

    Held record marcus-gpt-4s-successes-and-gpt-4s-failures-3ed47b

    The anthropic review does not accept e2 as supports.

  111. HeldBy this siteGary Marcus

    Held record marcus-gpt-4s-successes-and-gpt-4s-failures-3d3262

    The final-text check failed: rationale_supported_by_evidence, citations_correct; overclaims: OpenAI's March 14 launch page also identified Stripe as using GPT-4 [e3]..

  112. Assessment publishedBy this siteGary Marcus

    OpenAI has not disclosed GPT-4's size, architecture, energy use, processor count, or training data.

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

    Assessment marcus-gpt-4s-successes-and-gpt-4s-failures-17c3ce-a2.

  113. HeldBy this siteGary Marcus

    Held record marcus-gpt-4s-successes-and-gpt-4s-failures-05acd8

    The openai screen reads the claim differently.

  114. Assessment publishedBy this siteGary Marcus

    GPT 4.5 likely shows mixed results, with reasoning and hallucination problems unresolved and gains in only some domains.

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

    Assessment marcus-gpt-45-is-no-gpt-5-fd21a1-a1.

  115. Assessment publishedBy this siteGary Marcus

    GPT 4.5 is unlikely to meet high expectations.

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

    Assessment marcus-gpt-45-is-no-gpt-5-bf9e34-a1.

  116. Assessment publishedBy this siteGary Marcus

    Problems with reasoning and hallucination in deep learning systems have not gone away as a function of scaling.

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

    Assessment marcus-gpt-45-is-no-gpt-5-b2ee0b-a1.

  117. Assessment publishedBy this siteGary Marcus

    GPT 4.5 is unlikely to revolutionize AI.

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

    Assessment marcus-gpt-45-is-no-gpt-5-b28163-a1.

  118. Assessment publishedBy this siteGary Marcus

    Nvidia investors should not look only at quarterly reports.

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

    Assessment marcus-gpt-45-is-no-gpt-5-ac3c3f-a1.

  119. Assessment publishedBy this siteGary Marcus

    Investors should weigh the prospects for solving AI through pure scaling, which underpins Nvidia's pitch.

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

    Assessment marcus-gpt-45-is-no-gpt-5-a70f4e-a1.

  121. Assessment publishedBy this siteGary Marcus

    Scaling is no longer producing the gains it once did.

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

    Assessment marcus-gpt-45-is-no-gpt-5-9d42b9-a1.

  122. HeldBy this siteGary Marcus

    Held record marcus-gpt-45-is-no-gpt-5-746c82

    The final-text check failed: citations_correct.

  123. Assessment publishedBy this siteGary Marcus

    GPT 4.5 will not live up to the expectations that had been set for GPT-5.

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

    Assessment marcus-gpt-45-is-no-gpt-5-6a1ee1-a1.

  124. HeldBy this siteGary Marcus

    Held record marcus-gpt-45-is-no-gpt-5-5e8b20

    The openai screen reads the claim differently.

  125. HeldBy this siteGary Marcus

    Held record marcus-gpt-45-is-no-gpt-5-3e4ea6

    e1 is dated 2025-02, which is not inside the period described (2025-01-01 to 2025-02-26); it can only be context; e2 is dated 2025-02, which is not inside the period described (2025-01-01 to 2025-02-26); it can only be context

  126. Assessment publishedBy this siteGary Marcus

    Diminishing returns from scaling affect the whole AI field, not just OpenAI.

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

    Assessment marcus-gpt-45-is-no-gpt-5-20ad04-a1.

  127. Assessment publishedBy this siteGary Marcus

    The math for generative AI profitability currently doesn't work.

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

    Assessment marcus-gpt-45-is-no-gpt-5-1d00e5-a1.

  128. Assessment publishedBy this siteGary Marcus

    Prospects for pure scaling of AI are not looking great.

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

    Assessment marcus-gpt-45-is-no-gpt-5-1d00e5-3-a1.

  129. Assessment publishedBy this siteGary Marcus

    Generative AI's influence on productivity has been minimal.

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

    Assessment marcus-gpt-45-is-no-gpt-5-1d00e5-2-a1.

  130. HeldBy this siteGary Marcus

    Held record marcus-gpt-45-is-no-gpt-5-13c1cf

    e1 is dated 2025-02, which is not inside the period described (2024-02-26 to 2025-02-26); it can only be context

  131. Assessment publishedBy this siteGary Marcus

    Despite enormous investments, recent AI releases appear to be only minor variations on earlier ones.

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

    Assessment marcus-gpt-45-is-no-gpt-5-12cc4e-a1.

  132. Assessment publishedBy this siteGary Marcus

    Google's Bard image generation software will likely generate more IP legal work in 2024.

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

    Assessment marcus-google-bard-fast-gorgeous-and-out-e9f5f3-a1.

  133. Assessment publishedBy this siteGary Marcus

    2024 will be the year of generative AI litigation.

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

    Assessment marcus-google-bard-fast-gorgeous-and-out-ab5350-a2.

  135. Assessment publishedBy this siteGary Marcus

    Judging by its earnings call omissions, Google's AI arms race spending has not paid off so far.

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

    Assessment marcus-google-2001-dont-be-evil-b72dde-a1.

  136. HeldBy this siteGary Marcus

    Held record marcus-google-2001-dont-be-evil-b346c6

    The anthropic screen doubts the attribution.

  137. Assessment publishedBy this siteGary Marcus

    Pressure to recoup AI investment is driving Google to abandon its motto and pursue military and surveillance applications.

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

    Assessment marcus-google-2001-dont-be-evil-ab2236-a2.

  138. Assessment publishedBy this siteGary Marcus

    Google is presumably not already making zillions from AI (or it would have said so on the investor call), and that situation cannot continue indefinitely.

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

    Assessment marcus-google-2001-dont-be-evil-a11dcc-a2.

  139. Assessment publishedBy this siteGary Marcus

    The news of Google's change confirms Marcus's long-held fear that, no matter how much you must trust a company or its management, things can eventually change.

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

    Assessment marcus-google-2001-dont-be-evil-8beeac-a1.

  140. Assessment publishedBy this siteGary Marcus

    Google's shift toward military and surveillance AI was caused by financial pressure in the changed AI landscape.

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

    Assessment marcus-google-2001-dont-be-evil-68bca0-a1.

  141. HeldBy this siteGary Marcus

    Held record marcus-google-2001-dont-be-evil-0f5109

    The final-text check failed: citations_correct.

  142. Assessment publishedBy this siteGary Marcus

    The EU AI Act should adopt a tiered approach putting the greatest regulatory burden on the largest foundation model companies.

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

    Assessment marcus-give-me-compromise-or-give-me-chaos-e35335-a1.

  143. Assessment publishedBy this siteGary Marcus

    Generative AI's combined total revenue so far is only a few billion dollars, a tiny fraction of the economy.

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

    Assessment marcus-give-me-compromise-or-give-me-chaos-d372bb-2-a1.

  144. Assessment publishedBy this siteGary Marcus

    The EU needs to reach a compromise deal on its AI Act.

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

    Assessment marcus-give-me-compromise-or-give-me-chaos-c53e42-a1.

  145. HeldBy this siteGary Marcus

    Held record marcus-give-me-compromise-or-give-me-chaos-be08da

    The openai screen doubts the attribution.

  146. Assessment publishedBy this siteGary Marcus

    In the worst case the EU AI Act negotiations could produce nothing, leaving citizens largely unprotected from generative AI's negative externalities.

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

    Assessment marcus-give-me-compromise-or-give-me-chaos-aa75b4-a1.

  147. HeldBy this siteGary Marcus

    Held record marcus-give-me-compromise-or-give-me-chaos-93c9d0

    The anthropic screen doubts the attribution.

  148. Assessment publishedBy this siteGary Marcus

    As of November 2023, the EU AI Act is in danger of failing.

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

    Assessment marcus-give-me-compromise-or-give-me-chaos-9283e2-a1.

  149. Assessment publishedBy this siteGary Marcus

    Society should not allow big tech to exempt generative AI from regulation.

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

    Assessment marcus-give-me-compromise-or-give-me-chaos-218578-a1.

  150. Assessment publishedBy this siteGary Marcus

    Virtually all existing laws were written before AI was really on the scene, and so very few of them really envision how AI changes things.

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

    Assessment marcus-give-me-compromise-or-give-me-chaos-180bf6-a1.

  151. Assessment publishedBy this siteGary Marcus

    If the EU fails to reach a compromise on the AI Act, everyone will be in trouble and only the largest AI companies will benefit.

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

    Assessment marcus-give-me-compromise-or-give-me-chaos-13bd65-a1.

  153. Assessment publishedBy this siteGary Marcus

    The AI industry should not be exempted from regulation; regulation should be crafted wisely instead.

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

    Assessment marcus-give-me-compromise-or-give-me-chaos-0b74dc-a1.

  154. HeldBy this siteGary Marcus

    Held record marcus-getting-gpt-to-work-with-external-f8363f

    The anthropic screen reads the claim differently.

  155. HeldBy this siteGary Marcus

    Held record marcus-getting-gpt-to-work-with-external-f0df96

    The final-text check failed: rationale_supported_by_evidence; overclaims: In its instructed human baseline, 39 of 50 participants produced valid plans [e1]..

  156. Assessment publishedBy this siteGary Marcus

    True understanding of elementary math requires more than LLM token prediction or Wolfram-style symbol manipulation.

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

    Assessment marcus-getting-gpt-to-work-with-external-e9001a-a1.

  158. Assessment publishedBy this siteGary Marcus

    GPT-4 and similar current language-based AI programs struggle with mathematical calculations.

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

    Assessment marcus-getting-gpt-to-work-with-external-c70b85-a1.

  159. Assessment publishedBy this siteGary Marcus

    Neither GPT-4 nor its plug-ins have the commonsense knowledge and reasoning needed to reliably translate word problems into calculations.

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

    Assessment marcus-getting-gpt-to-work-with-external-c5d27d-a1.

  160. Assessment publishedBy this siteGary Marcus

    Getting LLMs to reliably use external tools may require major innovation, though it is not necessarily unsolvable.

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

    Assessment marcus-getting-gpt-to-work-with-external-c1535d-a1.

  161. Assessment publishedBy this siteGary Marcus

    Until fundamental problems are addressed, AI's ability to apply mathematics to real-world situations will remain limited.

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

    Assessment marcus-getting-gpt-to-work-with-external-903974-a1.

  162. HeldBy this siteGary Marcus

    Held record marcus-getting-gpt-to-work-with-external-8575ae

    The anthropic review does not accept e1 as contradicts.

  163. HeldBy this siteGary Marcus

    Held record marcus-getting-gpt-to-work-with-external-5f020c

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

  164. Assessment publishedBy this siteGary Marcus

    Adding the Code Interpreter and Wolfram Alpha plug-ins improves GPT-4's math performance.

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

    Assessment marcus-getting-gpt-to-work-with-external-53ff44-a1.

  165. Assessment publishedBy this siteGary Marcus

    So far, the authors have seen consistently unreliable results from LLMs using external tools, from math to fact-checking.

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

    Assessment marcus-getting-gpt-to-work-with-external-44d1f1-a1.

  166. Assessment publishedBy this siteGary Marcus

    With enough resources, OpenAI or Wolfram Alpha could fix some of the interface problems.

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

    Assessment marcus-getting-gpt-to-work-with-external-2c9d9f-a1.

  167. Assessment publishedBy this siteGary Marcus

    Getting LLMs to reliably use external tools is far from trivial.

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

    Assessment marcus-getting-gpt-to-work-with-external-19406b-a1.

  172. Assessment publishedBy this siteGary Marcus

    Mechanistic interpretability has had very limited success at finding world models inside LLMs.

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

    Assessment marcus-generative-ais-crippling-and-widespread-f56877-a1.

  173. HeldBy this siteGary Marcus

    Held record marcus-generative-ais-crippling-and-widespread-eb4d81

    The final-text check failed: proposition_faithful.

  174. Assessment publishedBy this siteGary Marcus

    The lack of world models explains a huge number of peculiar LLM errors.

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

    Assessment marcus-generative-ais-crippling-and-widespread-ea1d48-a1.

  175. HeldBy this siteGary Marcus

    Held record marcus-generative-ais-crippling-and-widespread-e8844f

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

  176. HeldBy this siteGary Marcus

    Held record marcus-generative-ais-crippling-and-widespread-e03987

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

  177. HeldBy this siteGary Marcus

    Held record marcus-generative-ais-crippling-and-widespread-d6ba15

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

  178. Assessment publishedBy this siteGary Marcus

    The idea that the military would use such obtuse video-understanding tools in the fog of war is halfway between alarming and preposterous.

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

    Assessment marcus-generative-ais-crippling-and-widespread-d6962e-a1.

  179. HeldBy this siteGary Marcus

    Held record marcus-generative-ais-crippling-and-widespread-d23c04

    The final-text check failed: citations_correct.

  180. HeldBy this siteGary Marcus

    Held record marcus-generative-ais-crippling-and-widespread-cb47d7

    The openai screen reads the claim differently.

  181. HeldBy this siteGary Marcus

    Held record marcus-generative-ais-crippling-and-widespread-ca71d4

    The anthropic review does not accept e3 as contradicts.

  182. HeldBy this siteGary Marcus

    Held record marcus-generative-ais-crippling-and-widespread-c71158

    The final-text check failed: citations_correct.

  183. HeldBy this siteGary Marcus

    Held record marcus-generative-ais-crippling-and-widespread-c1a5d9

    e1 is dated 2025, which is not inside the period described (… to 2025-06-27); it can only be context

  184. Assessment publishedBy this siteGary Marcus

    LLMs hallucinate often because they rebuild facts stochastically from fragments of text instead of using structured databases.

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

    Assessment marcus-generative-ais-crippling-and-widespread-c02a12-a1.

  185. HeldBy this siteGary Marcus

    Held record marcus-generative-ais-crippling-and-widespread-aef52f

    e2 is dated 2025-06, which is not inside the period described (2025-06-01 to 2025-06-28); it can only be context

  187. Assessment publishedBy this siteGary Marcus

    Parroting words has limited value for understanding human relationships.

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

    Assessment marcus-generative-ais-crippling-and-widespread-a66b70-a1.

  188. Assessment publishedBy this siteGary Marcus

    Much of what goes wrong with LLMs comes from the design choice to do without explicit world models.

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

    Assessment marcus-generative-ais-crippling-and-widespread-801c25-a1.

  189. Assessment publishedBy this siteGary Marcus

    Any system that counts as AGI should, at a minimum, be able to induce the rules of chess.

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

    Assessment marcus-generative-ais-crippling-and-widespread-7d2790-a1.

  190. Assessment publishedBy this siteGary Marcus

    ChatGPT cannot play chess reliably because it never induces a proper world model of the board and the rules.

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

    Assessment marcus-generative-ais-crippling-and-widespread-74e1e3-a1.

  191. Assessment publishedBy this siteGary Marcus

    LLMs try to function without classical world models, and this harms them.

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

    Assessment marcus-generative-ais-crippling-and-widespread-74b4af-a1.

  192. HeldBy this siteGary Marcus

    Held record marcus-generative-ais-crippling-and-widespread-72d3d3

    The final-text check failed: citations_correct.

  193. HeldBy this siteGary Marcus

    Held record marcus-generative-ais-crippling-and-widespread-72a40d

    The openai screen reads the claim differently.

  194. Assessment publishedBy this siteGary Marcus

    Video comprehension models show world-model failures, especially in unusual situations that stored cases cannot cover.

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

    Assessment marcus-generative-ais-crippling-and-widespread-601f36-a1.

  195. Assessment publishedBy this siteGary Marcus

    Marcus's intuition is that world models defined as environment transition functions will not be enough for AGI.

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

    Assessment marcus-generative-ais-crippling-and-widespread-5e7b02-a1.

  196. HeldBy this siteGary Marcus

    Held record marcus-generative-ais-crippling-and-widespread-54c7de

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

  197. HeldBy this siteGary Marcus

    Held record marcus-generative-ais-crippling-and-widespread-4bb5f4

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

  198. HeldBy this siteGary Marcus

    Held record marcus-generative-ais-crippling-and-widespread-4b8e3a

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

  199. Assessment publishedBy this siteGary Marcus

    As long as LLMs lack robust cognitive models of the world, they should never be fully trusted.

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

    Assessment marcus-generative-ais-crippling-and-widespread-48e2eb-a1.

  200. HeldBy this siteGary Marcus

    Held record marcus-generative-ais-crippling-and-widespread-3d27a5

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