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

Changes, page 27 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 27 of 37: changes 5201 to 5400 of 7383, newest first. Dates are when the change was recorded.

  2. HeldBy this siteGary Marcus

    Held record marcus-generative-ais-crippling-and-widespread-2e3047

    The final-text check failed: proposition_faithful.

  3. Assessment publishedBy this siteGary Marcus

    Grok 3 lost to Marcus in four moves at a tic-tac-toe variant that used y's and z's.

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

    Assessment marcus-generative-ais-crippling-and-widespread-1ed5cf-a1.

  4. Assessment publishedBy this siteGary Marcus

    Even the latest video generators, such as Veo 3, are not anchored firmly enough in the world.

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

    Assessment marcus-generative-ais-crippling-and-widespread-1a07e1-a1.

  5. Assessment publishedBy this siteGary Marcus

    As propositions become more numerous and complex, LLMs' encoding of them will also prove unreliable.

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

    Assessment marcus-generative-ais-crippling-and-widespread-137104-a2.

  6. HeldBy this siteGary Marcus

    Held record marcus-generative-ais-crippling-and-widespread-0e033f

    e1 is dated 2025, which is not inside the period described (… to 2025-06-28); it can only be context

  7. Assessment publishedBy this siteGary Marcus

    The AI field should follow the roadmap centred on cognitive world models that Marcus set out in 2020.

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

    Assessment marcus-generative-ais-crippling-and-widespread-0b54d5-a1.

  8. Assessment publishedBy this siteGary Marcus

    The generative AI systems of major AI companies cannot succeed without even more data than the internet-scale data they were already trained on.

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

    Assessment marcus-generative-ai-as-shakespearean-tragedy-e3aad0-a1.

  9. Assessment publishedBy this siteGary Marcus

    Current generative AI systems are unreliable and are unlikely to live up to the hype.

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

    Assessment marcus-generative-ai-as-shakespearean-tragedy-df7cea-a1.

  10. HeldBy this siteGary Marcus

    Held record marcus-generative-ai-as-shakespearean-tragedy-d61fc1

    e5 is dated 2024-04, which is not inside the period described (2024-01-01 to 2024-04-07); it can only be context

  11. HeldBy this siteGary Marcus

    Held record marcus-generative-ai-as-shakespearean-tragedy-b468de

    The anthropic screen reads the claim differently.

  12. Assessment publishedBy this siteGary Marcus

    We must stop funding a losing hand to the infinite degree, which crowds everything else out.

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

    Assessment marcus-generative-ai-as-shakespearean-tragedy-a053f4-a1.

  13. Assessment publishedBy this siteGary Marcus

    The AI field is an intellectual monoculture in which no alternative approach is nearly as well-funded as generative AI.

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

    Assessment marcus-generative-ai-as-shakespearean-tragedy-8533d4-a1.

  14. Assessment publishedBy this siteGary Marcus

    The field must invest heavily in alternatives to generative AI and stop pouring unlimited funding into it, which crowds out everything else.

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

    Assessment marcus-generative-ai-as-shakespearean-tragedy-4e7b38-a1.

  15. Assessment publishedBy this siteGary Marcus

    None of the problems listed (hallucinations, abstraction and reasoning, brittleness, data guzzling, etc.) has been adequately addressed.

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

    Assessment marcus-generative-ai-as-shakespearean-tragedy-317535-a1.

  16. HeldBy this siteGary Marcus

    Held record marcus-generative-ai-as-shakespearean-tragedy-0be214

    e6 is dated 2024-04, which is not inside the period described (2024-01-01 to 2024-04-07); it can only be context

  17. Assessment publishedBy this siteGary Marcus

    Generative AI will never disappear.

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

    Assessment marcus-genais-day-of-reckoning-may-have-f9828c-a2.

  18. HeldBy this siteGary Marcus

    Held record marcus-genais-day-of-reckoning-may-have-f69110

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

  19. Assessment publishedBy this siteGary Marcus

    The LLM industry may ultimately be worth around $30-40B in revenue while costing two to three times that to run.

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

    Assessment marcus-genais-day-of-reckoning-may-have-efb383-a1.

  20. HeldBy this siteGary Marcus

    Held record marcus-genais-day-of-reckoning-may-have-ee3ef0

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

  21. HeldBy this siteGary Marcus

    Held record marcus-genais-day-of-reckoning-may-have-d95d69

    The final-text check failed: citations_correct.

  22. Assessment publishedBy this siteGary Marcus

    Companies like OpenAI may partly supplement revenue by selling private user data to intelligence agencies and similar buyers.

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

    Assessment marcus-genais-day-of-reckoning-may-have-cc5680-a1.

  23. Assessment publishedBy this siteGary Marcus

    Nvidia, the largest company making serious AI profits, is down 17.75% year-to-date in 2025 and clearly facing headwinds.

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

    Assessment marcus-genais-day-of-reckoning-may-have-bd3406-a1.

  24. Assessment publishedBy this siteGary Marcus

    User disillusionment with generative AI is continuing to grow.

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

    Assessment marcus-genais-day-of-reckoning-may-have-b51414-a2.

  25. Assessment publishedBy this siteGary Marcus

    Reaching a trillion-dollar AI market will require a new, reliable and trustworthy kind of AI beyond GenAI.

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

    Assessment marcus-genais-day-of-reckoning-may-have-b0d0f5-a1.

  26. Assessment publishedBy this siteGary Marcus

    GenAI is an enormously popular approach, but one that is neither reliable nor particularly well-grounded in truth.

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

    Assessment marcus-genais-day-of-reckoning-may-have-9ef596-a1.

  27. Assessment publishedBy this siteGary Marcus

    The GenAI bubble may already have begun to deflate as of March 2025.

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

    Assessment marcus-genais-day-of-reckoning-may-have-8609d2-a1.

  28. Assessment publishedBy this siteGary Marcus

    Use of generative AI will likely continue to be held back by hallucinations and silly errors.

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

    Assessment marcus-genais-day-of-reckoning-may-have-83fd5b-a1.

  29. Assessment publishedBy this siteGary Marcus

    LLMs will not vanish but will ultimately become virtually free.

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

    Assessment marcus-genais-day-of-reckoning-may-have-834979-a1.

  30. Assessment publishedBy this siteGary Marcus

    Most GenAI model-building companies are losing money and few have made any profit.

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

    Assessment marcus-genais-day-of-reckoning-may-have-71b55a-a1.

  31. Assessment publishedBy this siteGary Marcus

    The promised major gains in economic productivity from GenAI have not materialized.

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

    Assessment marcus-genais-day-of-reckoning-may-have-609e17-a2.

  32. Assessment publishedBy this siteGary Marcus

    New approaches more connected to symbolic AI, perhaps neurosymbolic hybrids, might prove vastly more valuable than GenAI.

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

    Assessment marcus-genais-day-of-reckoning-may-have-5fd474-a1.

  33. Assessment publishedBy this siteGary Marcus

    So much in GenAI seems to have been built on dreams.

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

    Assessment marcus-genais-day-of-reckoning-may-have-5af829-a1.

  34. Assessment publishedBy this siteGary Marcus

    Most GenAI model-building companies will have meager or no profits.

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

    Assessment marcus-genais-day-of-reckoning-may-have-3ba860-a2.

  35. HeldBy this siteGary Marcus

    Held record marcus-genais-day-of-reckoning-may-have-37ce6a

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

  36. Assessment publishedBy this siteGary Marcus

    AI could someday be a trillion-dollar annual market.

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

    Assessment marcus-genais-day-of-reckoning-may-have-33e312-a1.

  37. Assessment publishedBy this siteGary Marcus

    There has been a massive LLM price war, especially since DeepSeek.

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

    Assessment marcus-genais-day-of-reckoning-may-have-239ebc-a1.

  38. Assessment publishedBy this siteGary Marcus

    AI agents will be a disappointment because of ongoing reliability and hallucination problems.

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

    Assessment marcus-genais-day-of-reckoning-may-have-195254-a1.

  39. HeldBy this siteGary Marcus

    Held record marcus-genais-day-of-reckoning-may-have-0f2f01

    The openai screen reads the claim differently.

  40. Assessment publishedBy this siteGary Marcus

    The economics of generative AI do not make sense and never have, given the technology's actual capabilities.

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

    Assessment marcus-genais-day-of-reckoning-may-have-0a288f-a1.

  41. Assessment publishedBy this siteGary Marcus

    Generative AI's apparent success amounts to 'success theater' rather than real success.

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

    Assessment marcus-genai-in-two-words-success-theater-c9a7ee-a1.

  42. HeldBy this siteGary Marcus

    Held record marcus-genai-in-two-words-success-theater-704680

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

  43. HeldBy this siteGary Marcus

    Held record marcus-gaslighting-and-reality-in-ai-fa52f0

    The final-text check failed: citations_correct.

  44. Assessment publishedBy this siteGary Marcus

    Large language models do not represent robust world models of how events unfold over time, and this was true from 2019 through early 2023.

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

    Assessment marcus-gaslighting-and-reality-in-ai-c17009-a1.

  45. HeldBy this siteGary Marcus

    Held record marcus-gaslighting-and-reality-in-ai-b662c7

    The final-text check failed: rationale_supported_by_evidence, citations_correct.

  46. Assessment publishedBy this siteGary Marcus

    GPT still does not understand what it is talking about.

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

    Assessment marcus-gaslighting-and-reality-in-ai-a0aea6-a1.

  47. Assessment publishedBy this siteGary Marcus

    Even after its January 30, 2023 math-focused update, ChatGPT still has many problems with easy word problems.

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

    Assessment marcus-gaslighting-and-reality-in-ai-9ae28b-a1.

  48. Assessment publishedBy this siteGary Marcus

    The dopey outputs of GPT-style LLMs are caused by a lack of world models of how things unfold over time, which is the central flaw of LLMs.

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

    Assessment marcus-gaslighting-and-reality-in-ai-77c9cb-a1.

  50. HeldBy this siteGary Marcus

    Held record marcus-gary-marcus-vs-sam-altman-on-gpt-adee33

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

  52. HeldBy this siteGary Marcus

    Held record marcus-gary-marcus-vs-sam-altman-on-gpt-adacc7

    The anthropic screen reads the claim differently.

  53. Assessment publishedBy this siteGary Marcus

    LLMs, including OpenAI's new o1, are not really AGI.

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

    Assessment marcus-gary-marcus-vs-sam-altman-on-gpt-871f78-a2.

  54. Assessment publishedBy this siteGary Marcus

    Current AI techniques alone will not be enough to reach AGI.

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

    Assessment marcus-gary-marcus-vs-sam-altman-on-gpt-3723f0-a2.

  55. Assessment publishedBy this siteGary Marcus

    AI research should not center almost entirely on LLMs, as it currently does.

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

    Assessment marcus-gary-marcus-vs-sam-altman-on-gpt-1838d1-a1.

  56. HeldBy this siteGary Marcus

    Held record marcus-game-over-for-pure-llms-even-turing-f0fac5

    The final-text check failed: citations_correct.

  57. HeldBy this siteGary Marcus

    Held record marcus-game-over-for-pure-llms-even-turing-e2f551

    e1 is dated 2025-09, which is not inside the period described (2019-01-01 to 2025-09-26); it can only be context; e2 is dated 2025-09, which is not inside the period described (2019-01-01 to 2025-09-26); it can only be context

  58. Assessment publishedBy this siteGary Marcus

    Marcus has always thought Sutton's Bitter Lesson essay was wrong, or more precisely overstated, and fatally flawed.

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

    Assessment marcus-game-over-for-pure-llms-even-turing-98fc9a-a1.

  59. Assessment publishedBy this siteGary Marcus

    Quite likely, AI needs some of both approaches: Sutton's reinforcement learning and Marcus's neurosymbolic approaches with innate constraints.

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

    Assessment marcus-game-over-for-pure-llms-even-turing-8ce5fe-a1.

  60. HeldBy this siteGary Marcus

    Held record marcus-game-over-for-pure-llms-even-turing-4e7c15

    The final-text check failed: citations_correct.

  61. HeldBy this siteGary Marcus

    Held record marcus-game-over-for-pure-llms-even-turing-495c54

    e3 is dated 2025-09, which is not inside the period described (2025-01-01 to 2025-09-26); it can only be context

  62. Assessment publishedBy this siteGary Marcus

    With Sutton having turned against pure LLMs, it is game over for the pure LLM approach.

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

    Assessment marcus-game-over-for-pure-llms-even-turing-4921c3-a1.

  64. Assessment publishedBy this siteGary Marcus

    Practically the only people still pretending that scaling LLMs is "all you need" are grifters.

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

    Assessment marcus-game-over-for-pure-llms-even-turing-1b2b2f-a1.

  65. HeldBy this siteGary Marcus

    Held record marcus-game-over-for-pure-llms-even-turing-1b1b63

    The anthropic review reads the claim differently.

  66. HeldBy this siteGary Marcus

    Held record marcus-further-trouble-in-hinton-city-fb8aee

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

  67. Assessment publishedBy this siteGary Marcus

    GenAI systems often don't come up with a novel independent design that would meet the requirements.

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

    Assessment marcus-further-trouble-in-hinton-city-ebe684-a2.

  68. Assessment publishedBy this siteGary Marcus

    Generative AI systems still struggle with noncanonical situations because frequency statistics are a poor substitute for deep understanding.

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

    Assessment marcus-further-trouble-in-hinton-city-c753c3-a2.

  69. HeldBy this siteGary Marcus

    Held record marcus-further-trouble-in-hinton-city-b4adab

    The anthropic screen reads the claim differently.

  70. Assessment publishedBy this siteGary Marcus

    GenAI systems effectively store things such as canonical video game controller images.

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

    Assessment marcus-further-trouble-in-hinton-city-860962-a1.

  71. HeldBy this siteGary Marcus

    Held record marcus-further-trouble-in-hinton-city-7f00e5

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

  72. HeldBy this siteGary Marcus

    Held record marcus-further-trouble-in-hinton-city-663092

    The openai screen reads the claim differently.

  73. HeldBy this siteGary Marcus

    Held record marcus-further-trouble-in-hinton-city-4f48d7

    The anthropic screen reads the claim differently.

  74. Assessment publishedBy this siteGary Marcus

    Evidence against Hinton's claims that GenAI understands and doesn't store things keeps accumulating.

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

    Assessment marcus-further-trouble-in-hinton-city-1fb335-a1.

  75. HeldBy this siteGary Marcus

    Held record marcus-further-trouble-in-hinton-city-0be6da

    The final-text check failed: citations_correct; overclaims: Its methods address questions whose answer options would otherwise become invalid when reordered..

  76. HeldBy this siteGary Marcus

    Held record marcus-further-breaking-news-further-vindicating-f53cd2

    The anthropic screen reads the claim differently.

  77. HeldBy this siteGary Marcus

    Held record marcus-further-breaking-news-further-vindicating-b0a779

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

  78. HeldBy this siteGary Marcus

    Held record marcus-further-breaking-news-further-vindicating-4c6540

    The anthropic review does not accept e2 as supports.

  79. Assessment publishedBy this siteGary Marcus

    Marcus thinks the SpaceX-xAI merger is really a kind of bailout, meant to give a lot of cash to xAI, which is otherwise in distress.

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

    Assessment marcus-four-theories-about-the-spacex-xai-dce469-a2.

  80. HeldBy this siteGary Marcus

    Held record marcus-four-theories-about-the-spacex-xai-d52934

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

  81. Assessment publishedBy this siteGary Marcus

    SpaceX should not replace its symbolic navigation algorithms with an LLM trained on X, which would be disastrous.

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

    Assessment marcus-four-theories-about-the-spacex-xai-c50ccf-a1.

  82. HeldBy this siteGary Marcus

    Held record marcus-four-theories-about-the-spacex-xai-ae8679

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

  83. HeldBy this siteGary Marcus

    Held record marcus-four-theories-about-the-spacex-xai-a8d1e8

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

  84. Assessment publishedBy this siteGary Marcus

    Trading xAI at a thousand times revenue seems nuts when it is in an increasingly crowded field where everyone is being forced into price wars.

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

    Assessment marcus-four-theories-about-the-spacex-xai-a8bbf4-a1.

  85. HeldBy this siteGary Marcus

    Held record marcus-four-theories-about-the-spacex-xai-8b2200

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

  86. Assessment publishedBy this siteGary Marcus

    xAI has faced a lot of backlash for being reckless and irresponsible, which is hardly a good brand name.

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

    Assessment marcus-four-theories-about-the-spacex-xai-83bd58-a1.

  88. HeldBy this siteGary Marcus

    Held record marcus-four-theories-about-the-spacex-xai-4d659e

    The openai screen reads the claim differently.

  89. HeldBy this siteGary Marcus

    Held record marcus-four-theories-about-the-spacex-xai-47877f

    The openai screen reads the claim differently.

  90. Assessment publishedBy this siteGary Marcus

    Marcus says that if he owned SpaceX shares he would feel that Musk way overpaid for xAI, his own company, in the merger.

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

    Assessment marcus-four-theories-about-the-spacex-xai-2aa82e-a1.

  91. HeldBy this siteGary Marcus

    Held record marcus-four-theories-about-the-spacex-xai-0d585b

    The anthropic screen reads the claim differently.

  92. HeldBy this siteGary Marcus

    Held record marcus-form-function-and-the-giant-gulf-eb5eae

    The anthropic screen doubts the attribution.

  93. Assessment publishedBy this siteGary Marcus

    Text-to-image systems like Stable Diffusion continue to have problems with negation in prompts.

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

    Assessment marcus-form-function-and-the-giant-gulf-df5e53-a1.

  94. HeldBy this siteGary Marcus

    Held record marcus-form-function-and-the-giant-gulf-b4f3f4

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

  95. Assessment publishedBy this siteGary Marcus

    Stable Diffusion has trouble even with simple part-whole prompts such as a white bicycle with green wheels.

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

    Assessment marcus-form-function-and-the-giant-gulf-a558ad-a1.

  96. HeldBy this siteGary Marcus

    Held record marcus-form-function-and-the-giant-gulf-8eb44f

    The anthropic screen reads the claim differently.

  98. Assessment publishedBy this siteGary Marcus

    Stable Diffusion produced only one very marginal success in nine further attempts at a part-function prompt.

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

    Assessment marcus-form-function-and-the-giant-gulf-67939d-a1.

  99. Assessment publishedBy this siteGary Marcus

    Image synthesis systems only inconsistently match their linguistic input to the images produced, doing well on some prompts and poorly on others.

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

    Assessment marcus-form-function-and-the-giant-gulf-52ece7-a1.

  100. HeldBy this siteGary Marcus

    Held record marcus-form-function-and-the-giant-gulf-2bd72a

    The anthropic screen reads the claim differently.

  101. Assessment publishedBy this siteGary Marcus

    Image synthesis systems generate high quality images that only highly trained human artists could surpass.

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

    Assessment marcus-form-function-and-the-giant-gulf-23167d-a1.

  102. HeldBy this siteGary Marcus

    Held record marcus-form-function-and-the-giant-gulf-0d6191

    The anthropic screen reads the claim differently.

  103. HeldBy this siteGary Marcus

    Held record marcus-five-ways-in-which-the-last-3-months-fe481c

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

  104. Assessment publishedBy this siteGary Marcus

    Newer models such as DeepSeek R1, with a rule-based reward system, partly use neurosymbolic techniques.

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

    Assessment marcus-five-ways-in-which-the-last-3-months-f8bc64-a1.

  105. Assessment publishedBy this siteGary Marcus

    Since 2022 there has been one more huge advance but no system that merits the name GPT-5.

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

    Assessment marcus-five-ways-in-which-the-last-3-months-eee034-a1.

  106. HeldBy this siteGary Marcus

    Held record marcus-five-ways-in-which-the-last-3-months-d7fd9a

    a observation claim cannot be assessed on the causal_assessment axis

  107. HeldBy this siteGary Marcus

    Held record marcus-five-ways-in-which-the-last-3-months-d4ff57

    The anthropic review does not accept e1 as supports.

  108. HeldBy this siteGary Marcus

    Held record marcus-five-ways-in-which-the-last-3-months-cdd98f

    hedged language is not scored, even on a plain reading

  109. HeldBy this siteGary Marcus

    Held record marcus-five-ways-in-which-the-last-3-months-c180be

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

  111. Assessment publishedBy this siteGary Marcus

    Test-time compute systems improve only certain domains like coding and math, not across the board as GPT-4 did over GPT-3.

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

    Assessment marcus-five-ways-in-which-the-last-3-months-b7401a-a1.

  112. HeldBy this siteGary Marcus

    Held record marcus-five-ways-in-which-the-last-3-months-8c9627

    The openai screen reads the claim differently.

  113. Assessment publishedBy this siteGary Marcus

    AI should combine neural networks with classical symbolic techniques such as rules.

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

    Assessment marcus-five-ways-in-which-the-last-3-months-532297-a1.

  114. Assessment publishedBy this siteGary Marcus

    The latest systems such as OpenAI's Deep Research still suffer from hallucinations and reasoning errors.

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

    Assessment marcus-five-ways-in-which-the-last-3-months-471130-a1.

  115. Assessment publishedBy this siteGary Marcus

    LLMs have largely become a commodity.

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

    Assessment marcus-five-ways-in-which-the-last-3-months-30958f-a2.

  116. HeldBy this siteGary Marcus

    Held record marcus-five-ways-in-which-the-last-3-months-27fd37

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

  117. Assessment publishedBy this siteGary Marcus

    Despite hundreds of billions invested since August 2022, pure LLM scaling has not delivered what some expected.

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

    Assessment marcus-five-ways-in-which-the-last-3-months-25cc7d-a1.

  118. Assessment publishedBy this siteGary Marcus

    Because DeepSeek is actually open, unlike some other companies, its openness may help it lure talent from more walled-off operations.

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

    Assessment marcus-five-things-most-people-dont-seem-a104ff-a1.

  119. Assessment publishedBy this siteGary Marcus

    The biggest threat from DeepSeek is not to Nvidia, because people will still need GPUs, albeit fewer higher-end ones.

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

    Assessment marcus-five-things-most-people-dont-seem-9850cd-a1.

  120. Assessment publishedBy this siteGary Marcus

    People will still need GPUs, though fewer higher-end ones, after DeepSeek.

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

    Assessment marcus-five-things-most-people-dont-seem-9850cd-3-a1.

  121. Assessment publishedBy this siteGary Marcus

    The biggest threat from DeepSeek is to OpenAI and Anthropic, because price wars will undercut their hopes of making a profit.

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

    Assessment marcus-five-things-most-people-dont-seem-9850cd-2-a1.

  122. Assessment publishedBy this siteGary Marcus

    DeepSeek r1 does not solve hallucinations or reliability problems.

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

    Assessment marcus-five-things-most-people-dont-seem-570c01-a1.

  123. Assessment publishedBy this siteGary Marcus

    DeepSeek r1 is still expensive to run at inference, especially when made to reason longer like o3.

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

    Assessment marcus-five-things-most-people-dont-seem-507f10-a1.

  124. Assessment publishedBy this siteGary Marcus

    DeepSeek r1 is not smarter than earlier models.

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

    Assessment marcus-five-things-most-people-dont-seem-103fa6-a1.

  125. Assessment publishedBy this siteGary Marcus

    DeepSeek r1 was trained more cheaply than earlier models.

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

    Assessment marcus-five-things-most-people-dont-seem-103fa6-2-a1.

  126. Assessment publishedBy this siteGary Marcus

    DeepSeek is an economic revolution and a geopolitical wake-up call.

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

    Assessment marcus-five-things-most-people-dont-seem-0c6686-a1.

  127. HeldBy this siteGary Marcus

    Held record marcus-five-things-most-people-dont-seem-0c6686-2

    The openai screen reads the claim differently.

  128. Assessment publishedBy this siteGary Marcus

    As of July 2024, users, clients, and venture capitalists have lost faith in generative AI.

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

    Assessment marcus-five-signs-that-the-genai-honeymoon-f4eec6-a1.

  129. Assessment publishedBy this siteGary Marcus

    Meta dropping a major GenAI project is one more GenAI monetization scheme that has failed.

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

    Assessment marcus-five-signs-that-the-genai-honeymoon-8cb84b-a1.

  130. Assessment publishedBy this siteGary Marcus

    Generative AI is fading fast, as of July 2024, and may turn out to be a dud.

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

    Assessment marcus-five-signs-that-the-genai-honeymoon-5cdb09-a1.

  131. Assessment publishedBy this siteGary Marcus

    The GenAI setback could open the way to much greater progress in AI through other approaches.

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

    Assessment marcus-five-signs-that-the-genai-honeymoon-1b9dfc-a1.

  132. Assessment publishedBy this siteGary Marcus

    As of late July 2024, the GenAI bubble has begun to burst.

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

    Assessment marcus-five-signs-that-the-genai-honeymoon-10b5aa-a1.

  133. HeldBy this siteGary Marcus

    Held record marcus-five-signs-that-the-genai-honeymoon-041911

    The anthropic screen reads the claim differently.

  134. HeldBy this siteGary Marcus

    Held record marcus-five-signs-that-the-genai-honeymoon-026405

    The anthropic screen reads the claim differently.

  135. HeldBy this siteGary Marcus

    Held record marcus-five-signs-that-generative-ai-is-eb11a3

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

  136. Assessment publishedBy this siteGary Marcus

    If generative AI collapses, a large part of the cause will be investors naively extrapolating scaling and usage trends.

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

    Assessment marcus-five-signs-that-generative-ai-is-8afb1d-a2.

  137. HeldBy this siteGary Marcus

    Held record marcus-five-signs-that-generative-ai-is-896b72

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

  138. Assessment publishedBy this siteGary Marcus

    The decline in vibe coding usage is quite likely because people are growing disenchanted with the results.

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

    Assessment marcus-five-signs-that-generative-ai-is-896b72-2-a1.

  139. HeldBy this siteGary Marcus

    Held record marcus-five-signs-that-generative-ai-is-0da48a

    The anthropic screen doubts the attribution.

  140. HeldBy this siteGary Marcus

    Held record marcus-five-quick-updates-about-that-apple-fa9f80

    e1 is dated 2025-06, which is not inside the period described (2025-01-01 to 2025-06-17); it can only be context

  141. Assessment publishedBy this siteGary Marcus

    Reaching AGI will require doing much better than current generative AI on distribution shift.

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

    Assessment marcus-five-quick-updates-about-that-apple-f0c3b2-a1.

  142. Assessment publishedBy this siteGary Marcus

    LLMs struggle with distribution shift, and this is the whole ballgame.

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

    Assessment marcus-five-quick-updates-about-that-apple-cc47ad-a1.

  143. Assessment publishedBy this siteGary Marcus

    Generative AI systems do well on familiar problems with no wrinkles, but their outputs on anything else are suspect.

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

    Assessment marcus-five-quick-updates-about-that-apple-b739ae-a1.

  144. Assessment publishedBy this siteGary Marcus

    The rejoinder “The Illusion of the Illusion of Thinking” turned out to be an error-ridden joke.

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

    Assessment marcus-five-quick-updates-about-that-apple-a50a1b-a2.

  145. Assessment publishedBy this siteGary Marcus

    When data contamination is controlled for, LLM performance on hard coding problems drops to zero.

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

    Assessment marcus-five-quick-updates-about-that-apple-977b1e-a1.

  146. Assessment publishedBy this siteGary Marcus

    The Apple reasoning paper's conclusions about LLM reasoning limits still look basically correct.

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

    Assessment marcus-five-quick-updates-about-that-apple-804385-a1.

  147. Assessment publishedBy this siteGary Marcus

    Although neural networks generalize beyond their training data somewhat better than in 1998, distribution shift is still the core challenge for them.

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

    Assessment marcus-five-quick-updates-about-that-apple-4be9a2-a1.

  150. Assessment publishedBy this siteGary Marcus

    AI companies will probably argue their products don't meet the order's serious-risk threshold.

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

    Assessment marcus-first-take-on-the-white-house-executive-e6dae2-a1.

  151. HeldBy this siteGary Marcus

    Held record marcus-first-take-on-the-white-house-executive-dbbe5e

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

  152. Assessment publishedBy this siteGary Marcus

    AI companies will look for and find loopholes in the Executive Order.

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

    Assessment marcus-first-take-on-the-white-house-executive-d4b3d6-a1.

  153. Assessment publishedBy this siteGary Marcus

    The Executive Order's effectiveness depends heavily on exact wording, enforcement, and how much is binding rather than voluntary.

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

    Assessment marcus-first-take-on-the-white-house-executive-c3b9cd-a1.

  154. Assessment publishedBy this siteGary Marcus

    US AI policy should ultimately include more binding rules with enforcement teeth beyond guidance.

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

    Assessment marcus-first-take-on-the-white-house-executive-b6506e-a1.

  155. Assessment publishedBy this siteGary Marcus

    AI governance needs concrete implementation rather than more position papers.

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

    Assessment marcus-first-take-on-the-white-house-executive-abcad3-a1.

  157. Assessment publishedBy this siteGary Marcus

    The AI Executive Order may fall short despite its merits.

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

    Assessment marcus-first-take-on-the-white-house-executive-6fa4ea-a1.

  158. HeldBy this siteGary Marcus

    Held record marcus-first-take-on-the-white-house-executive-69d56c

    The final-text check failed; overclaims: e1 explanation: "Reading section 4.2 confirms that it does not establish an FDA-style approval gate.".

  160. HeldBy this siteGary Marcus

    Held record marcus-first-take-on-the-white-house-executive-63cf10

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

  161. Assessment publishedBy this siteGary Marcus

    The US Executive Order is an important first step toward concrete implementation of AI policy.

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

    Assessment marcus-first-take-on-the-white-house-executive-42ddbc-a1.

  163. HeldBy this siteGary Marcus

    Held record marcus-first-take-on-the-white-house-executive-09599a

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

  165. Assessment publishedBy this siteGary Marcus

    If Sam Altman comes back, the board will have given him more power, which may inherently undermine safety.

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

    Assessment marcus-firing-sam-and-then-trying-to-bring-fb7449-2-a1.

  166. HeldBy this siteGary Marcus

    Held record marcus-firing-sam-and-then-trying-to-bring-ad1d1d

    The openai screen reads the claim differently.

  168. Assessment publishedBy this siteGary Marcus

    If Altman does not return, Mira Murati may rightly be perturbed, and if the rumor is true, the board has undermined her.

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

    Assessment marcus-firing-sam-and-then-trying-to-bring-0d4ba1-a1.

  169. HeldBy this siteGary Marcus

    Held record marcus-facing-facts-684ad6

    a observation claim cannot be assessed on the causal_assessment axis

  170. Assessment publishedBy this siteGary Marcus

    The AI field needs new approaches based on ideas other than LLMs to address hallucinations.

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

    Assessment marcus-facing-facts-684ad6-2-a1.

  171. HeldBy this siteGary Marcus

    Held record marcus-face-it-self-driving-cars-still-havent-fac0bc

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

  172. HeldBy this siteGary Marcus

    Held record marcus-face-it-self-driving-cars-still-havent-f9acb7

    The final-text check failed: citations_correct.

  173. Assessment publishedBy this siteGary Marcus

    Marcus would not trust a self-driving plane, and he does not think any quasi-self-driving car has yet earned its stripes.

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

    Assessment marcus-face-it-self-driving-cars-still-havent-f872c4-a1.

  175. HeldBy this siteGary Marcus

    Held record marcus-face-it-self-driving-cars-still-havent-cc3ad7

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

  176. HeldBy this siteGary Marcus

    Held record marcus-face-it-self-driving-cars-still-havent-a927df

    The anthropic screen doubts the attribution.

  178. Assessment publishedBy this siteGary Marcus

    Approving 24/7 driverless operation across San Francisco without a vetted solution to edge cases was reckless and bound to lead to accidents.

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

    Assessment marcus-face-it-self-driving-cars-still-havent-6d901d-a1.

  179. Assessment publishedBy this siteGary Marcus

    Marcus suggests that the way driverless-car safety depends on data rather than reasoning should remind attentive readers of LLMs.

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

    Assessment marcus-face-it-self-driving-cars-still-havent-67c16c-a1.

  181. Assessment publishedBy this siteGary Marcus

    Nobody has yet figured out how to build a driverless car that can deal with the real world's endless crazy and out-of-the-ordinary circumstances.

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

    Assessment marcus-face-it-self-driving-cars-still-havent-205e76-a1.

  182. HeldBy this siteGary Marcus

    Held record marcus-face-it-self-driving-cars-still-havent-066600

    The anthropic review does not accept e1 as supports.

  183. HeldBy this siteGary Marcus

    Held record marcus-f-cancer-fa9717

    a causal claim cannot be assessed on the forecast_outcome axis

  184. Assessment publishedBy this siteGary Marcus

    Over and over, drugs discovered with AI work in animal trials but are not safe enough in humans, because we just don't understand biology well enough.

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

    Assessment marcus-f-cancer-f93b9b-a2.

  185. HeldBy this siteGary Marcus

    Held record marcus-f-cancer-f63107

    The anthropic screen reads the claim differently.

  186. Assessment publishedBy this siteGary Marcus

    Actually beating cancer will also require AI to sweat the details a whole lot better.

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

    Assessment marcus-f-cancer-bfb5d1-a1.

  187. Assessment publishedBy this siteGary Marcus

    In the near term, AI's biggest contributions to fighting disease will probably come from giving humans better tools, not from radical new scientific discoveries.

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

    Assessment marcus-f-cancer-906d6b-a1.

  188. Assessment publishedBy this siteGary Marcus

    Getting the right AI algorithm matters a great deal for beating disease, and current AI is far from that point.

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

    Assessment marcus-f-cancer-87c4c1-a1.

  189. Assessment publishedBy this siteGary Marcus

    In the short term, AI will plausibly bring productivity gains to medical research through mundane tasks, even if more dramatic breakthroughs do not arrive.

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

    Assessment marcus-f-cancer-72beae-a1.

  190. Assessment publishedBy this siteGary Marcus

    Dario Amodei's claims that AI will double human life span in the next decade are deluded technosolutionism.

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

    Assessment marcus-f-cancer-286fed-a1.

  191. Assessment publishedBy this siteGary Marcus

    So far, AI has not changed the toll that cancer takes on people.

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

    Assessment marcus-f-cancer-1e2f0f-a1.