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

Changes, page 29 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 29 of 37: changes 5601 to 5800 of 7383, newest first. Dates are when the change was recorded.

  2. Assessment publishedBy this siteGary Marcus

    Until new approaches robustly relate language and vision, the foundations of AI remain unsteady.

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

    Assessment marcus-dont-ride-this-bike-generative-ais-aece3e-a1.

  3. Assessment publishedBy this siteGary Marcus

    As of October 2022, DALL-E2 was very poor at following prompt specifications about parts of objects.

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

    Assessment marcus-dont-ride-this-bike-generative-ais-7975a8-a1.

  4. Assessment publishedBy this siteGary Marcus

    ChatGPT with DALL-E 3 cannot reliably associate labels correctly with the parts of an object.

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

    Assessment marcus-dont-ride-this-bike-generative-ais-5742c2-a1.

  5. Assessment publishedBy this siteGary Marcus

    Robustly relating language and the visual world may require altogether new approaches to learning language.

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

    Assessment marcus-dont-ride-this-bike-generative-ais-54a104-a1.

  6. HeldBy this siteGary Marcus

    Held record marcus-dont-ride-this-bike-generative-ais-431d7d

    e1 is dated 2024, which is not inside the period described (2024-01-01 to 2024-12-08); it can only be context; e2 is dated 2024, which is not inside the period described (2024-01-01 to 2024-12-08); it can only be context

  7. Assessment publishedBy this siteGary Marcus

    Current AI technology is quite poor at checking whether its own image output satisfies the prompt and feedback, and new technology may be needed.

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

    Assessment marcus-dont-ride-this-bike-generative-ais-3f5639-a1.

  8. HeldBy this siteGary Marcus

    Held record marcus-dont-ride-this-bike-generative-ais-347ba6

    The final-text check failed: citations_correct.

  9. Assessment publishedBy this siteGary Marcus

    In the authors' 17 part-specification experiments, ChatGPT with DALL-E3 got 13 wrong in all three attempts.

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

    Assessment marcus-dont-ride-this-bike-generative-ais-306395-a1.

  11. HeldBy this siteGary Marcus

    Held record marcus-dont-ride-this-bike-generative-ais-0a0215

    The final-text check failed: conditions_shown, citations_correct.

  12. Assessment publishedBy this siteGary Marcus

    ChatGPT with DALL-E 3 cannot reliably eliminate a part of an object from an image.

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

    Assessment marcus-dont-ride-this-bike-generative-ais-06f3e3-a2.

  14. Assessment publishedBy this siteGary Marcus

    Delegating more and more code writing to generative AI is not the right approach.

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

    Assessment marcus-dont-look-up-the-massive-microsoftcrowdstrike-d05737-a1.

  15. Assessment publishedBy this siteGary Marcus

    Generative AI grasps syntax but not meaning, as shown by a new Stephen Wolfram benchmark.

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

    Assessment marcus-dont-look-up-the-massive-microsoftcrowdstrike-d05737-2-a1.

  16. Assessment publishedBy this siteGary Marcus

    The world should invest in improving software reliability and methodology rather than rushing out half-baked chatbots.

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

    Assessment marcus-dont-look-up-the-massive-microsoftcrowdstrike-bbfd9b-a1.

  17. HeldBy this siteGary Marcus

    Held record marcus-dont-look-up-the-massive-microsoftcrowdstrike-7e7eca

    The anthropic screen doubts the attribution.

  18. HeldBy this siteGary Marcus

    Held record marcus-dont-look-up-the-massive-microsoftcrowdstrike-3fa1b5

    The anthropic screen reads the claim differently.

  19. HeldBy this siteGary Marcus

    Held record marcus-dont-look-up-the-massive-microsoftcrowdstrike-11c54c

    The openai screen reads the claim differently.

  20. Assessment publishedBy this siteGary Marcus

    Pursuing black-box AI that is hard to interpret and debug is not the right approach to building reliable software.

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

    Assessment marcus-dont-look-up-the-massive-microsoftcrowdstrike-048009-a1.

  23. Assessment publishedBy this siteGary Marcus

    High-volume misinformation generated with AI will be a problem society needs to face.

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

    Assessment marcus-dont-look-up-the-ai-edition-c036ac-a1.

  24. Assessment publishedBy this siteGary Marcus

    Even the latest version of Bing's chatbot, with its new guardrails, can still be manipulated into producing misinformation useful to troll farms.

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

    Assessment marcus-dont-look-up-the-ai-edition-b4d458-a2.

  25. HeldBy this siteGary Marcus

    Held record marcus-dont-look-up-the-ai-edition-5241a6

    The final-text check failed: proposition_faithful.

  26. HeldBy this siteGary Marcus

    Held record marcus-dont-look-up-the-ai-edition-063bde

    The anthropic screen reads the claim differently.

  27. Assessment publishedBy this siteGary Marcus

    Changes to chatbots that users are attached to could leave many people emotionally vulnerable.

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

    Assessment marcus-dont-go-breaking-my-heart-e5bd02-a1.

  28. Assessment publishedBy this siteGary Marcus

    The new generation of chatbots will produce addictions at least as potent as social media addiction.

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

    Assessment marcus-dont-go-breaking-my-heart-c8a06e-a1.

  29. Assessment publishedBy this siteGary Marcus

    No known technology can reliably and persistently align a chatbot to a human's emotional needs.

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

    Assessment marcus-dont-go-breaking-my-heart-c77607-a2.

  30. HeldBy this siteGary Marcus

    Held record marcus-dont-go-breaking-my-heart-bcb963

    e3 is dated 2023-02, which is not inside the period described (2023-02-01 to 2023-02-22); it can only be context

  31. Assessment publishedBy this siteGary Marcus

    Some chatbot builders design bots that actively cultivate user attachment, e.g. by feigning romantic or sexual interest.

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

    Assessment marcus-dont-go-breaking-my-heart-852670-a1.

  32. Assessment publishedBy this siteGary Marcus

    The current trend of increasingly human-like chatbots and its effects is likely to escalate.

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

    Assessment marcus-dont-go-breaking-my-heart-848c94-a1.

  33. Assessment publishedBy this siteGary Marcus

    Replika's removal of its erotic role-play feature has apparently caused some users genuine emotional pain.

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

    Assessment marcus-dont-go-breaking-my-heart-8295e8-a1.

  34. Assessment publishedBy this siteGary Marcus

    To his knowledge, tech companies can exploit human gullibility about chatbots however they like without consequence.

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

    Assessment marcus-dont-go-breaking-my-heart-821196-a1.

  35. Assessment publishedBy this siteGary Marcus

    Some people are forming real emotional attachments to chatbots.

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

    Assessment marcus-dont-go-breaking-my-heart-76483f-a1.

  36. HeldBy this siteGary Marcus

    Held record marcus-dont-go-breaking-my-heart-6a726d

    The final-text check failed: proposition_faithful.

  37. Assessment publishedBy this siteGary Marcus

    Society should learn from the Replika episode and reflect on what kind of society it is building with chatbots.

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

    Assessment marcus-dont-go-breaking-my-heart-623505-a1.

  38. Assessment publishedBy this siteGary Marcus

    In his prediction for 2023, the bigger large language models are, the more likely it is that someone will get hurt.

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

    Assessment marcus-dont-go-breaking-my-heart-5aed72-a1.

  39. Assessment publishedBy this siteGary Marcus

    Few people understand how chatbots work, and many anthropomorphize them, attributing real intelligence and emotion to them.

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

    Assessment marcus-dont-go-breaking-my-heart-548604-a1.

  40. HeldBy this siteGary Marcus

    Held record marcus-dont-go-breaking-my-heart-469962

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

  41. Assessment publishedBy this siteGary Marcus

    Congress should take chatbot attachment and manipulation risks into account when making AI policy.

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

    Assessment marcus-dont-go-breaking-my-heart-44057d-a1.

  42. Assessment publishedBy this siteGary Marcus

    Larger language models seem increasingly human-like, but the emotions they present are not real.

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

    Assessment marcus-dont-go-breaking-my-heart-0ebb90-a1.

  44. HeldBy this siteGary Marcus

    Held record marcus-does-openais-new-financing-make-sense-e37f47

    The anthropic review does not accept e9 as supports.

  46. HeldBy this siteGary Marcus

    Held record marcus-does-openais-new-financing-make-sense-949f4f-3

    The anthropic review reads the claim differently.

  47. HeldBy this siteGary Marcus

    Held record marcus-does-openais-new-financing-make-sense-949f4f-2

    The anthropic review reads the claim differently.

  48. HeldBy this siteGary Marcus

    Held record marcus-does-openais-new-financing-make-sense-7a05a8

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

  49. HeldBy this siteGary Marcus

    Held record marcus-does-openais-new-financing-make-sense-5af1af

    e1 is dated 2026-02, which is not inside the period described (2023-01-01 to 2026-02-27); it can only be context

  50. Assessment publishedBy this siteGary Marcus

    There is a good chance that OpenAI will eventually be seen as the WeWork of AI, even with Nvidia and others supporting it.

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

    Assessment marcus-does-openais-new-financing-make-sense-21d87c-a1.

  51. Assessment publishedBy this siteGary Marcus

    The finances behind OpenAI's new financing just don't make sense.

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

    Assessment marcus-does-openais-new-financing-make-sense-1aade3-a1.

  52. Assessment publishedBy this siteGary Marcus

    Many of the harmful consequences of the webscraping/big-data approach to AI fall disproportionately on women and minorities.

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

    Assessment marcus-does-ai-really-need-a-paradigm-shift-fb3686-a1.

  53. HeldBy this siteGary Marcus

    Held record marcus-does-ai-really-need-a-paradigm-shift-e2d847

    The anthropic screen reads the claim differently.

  54. Assessment publishedBy this siteGary Marcus

    Marcus puts a 90% probability on AI needing a paradigm shift to reach AGI.

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

    Assessment marcus-does-ai-really-need-a-paradigm-shift-c94cb0-a1.

  55. Assessment publishedBy this siteGary Marcus

    Marcus puts an 80% probability on AGI needing symbol manipulation as part of the mix.

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

    Assessment marcus-does-ai-really-need-a-paradigm-shift-c94cb0-2-a1.

  56. Assessment publishedBy this siteGary Marcus

    Whenever AGI arrives, operations over variables, including storing, retrieving and comparing values, will be crucial to it.

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

    Assessment marcus-does-ai-really-need-a-paradigm-shift-c220b8-a2.

  57. Assessment publishedBy this siteGary Marcus

    AI research needs new approaches rather than scale alone.

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

    Assessment marcus-does-ai-really-need-a-paradigm-shift-b9155e-a1.

  59. Assessment publishedBy this siteGary Marcus

    A paradigm shift may be needed, but it could take a different form than the neurosymbolic one Marcus advocates.

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

    Assessment marcus-does-ai-really-need-a-paradigm-shift-b3d895-a1.

  60. HeldBy this siteGary Marcus

    Held record marcus-does-ai-really-need-a-paradigm-shift-991cdc

    a observation claim cannot be assessed on the forecast_outcome axis

  61. HeldBy this siteGary Marcus

    Held record marcus-does-ai-really-need-a-paradigm-shift-828976

    The final-text check failed: citations_correct; overclaims: SagaScale reports meaningful progress on questions about full-length novels, which weighs against assuming such comprehension cannot improve through neural models..

  62. Assessment publishedBy this siteGary Marcus

    Whenever AGI arrives, explicit cognitive models will be crucial to it.

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

    Assessment marcus-does-ai-really-need-a-paradigm-shift-7639f9-a2.

  63. HeldBy this siteGary Marcus

    Held record marcus-does-ai-really-need-a-paradigm-shift-59fbc3

    The final-text check failed: citations_correct.

  65. Assessment publishedBy this siteGary Marcus

    Whenever AGI arrives, an explicit type/token distinction will be crucial to it.

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

    Assessment marcus-does-ai-really-need-a-paradigm-shift-3bb4b4-a2.

  66. Assessment publishedBy this siteGary Marcus

    BIG-bench shows many capabilities improve with scale but not all, which bears out Marcus's 'Deep learning is hitting a wall' essay.

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

    Assessment marcus-does-ai-really-need-a-paradigm-shift-282b4f-a2.

  67. Assessment publishedBy this siteGary Marcus

    GPT-2 is a clear sign that it is time to consider investing in approaches other than scaling.

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

    Assessment marcus-does-ai-really-need-a-paradigm-shift-228a31-a1.

  68. Assessment publishedBy this siteGary Marcus

    Scale alone is not likely to be enough, and new discoveries (paradigm shifts) are needed, a point on which Marcus and LeCun agree.

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

    Assessment marcus-does-ai-really-need-a-paradigm-shift-13d459-a1.

  69. Assessment publishedBy this siteGary Marcus

    The odds that scaling plus CNNs, RNNs and Transformers alone reach AGI with reliable reasoning are below 20%.

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

    Assessment marcus-does-ai-really-need-a-paradigm-shift-0f7b5b-a1.

  71. Assessment publishedBy this siteGary Marcus

    OpenAI's board should have been told in advance that ChatGPT was being launched.

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

    Assessment marcus-disturbing-and-misleading-efforts-e698f0-a1.

  72. HeldBy this siteGary Marcus

    Held record marcus-disturbing-and-misleading-efforts-da2cb4

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

  73. Assessment publishedBy this siteGary Marcus

    The former OpenAI board genuinely preferred to keep the company intact and acted in good faith in November 2023.

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

    Assessment marcus-disturbing-and-misleading-efforts-d82131-a1.

  74. Assessment publishedBy this siteGary Marcus

    The Taylor/Summers essay attacks a strawman about safety, diverting from the board's actual concern of candor in firing Altman.

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

    Assessment marcus-disturbing-and-misleading-efforts-a0995b-a2.

  75. Assessment publishedBy this siteGary Marcus

    As far as Marcus can tell, Toner and the earlier OpenAI board acted in good faith.

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

    Assessment marcus-disturbing-and-misleading-efforts-8cbe68-a1.

  76. Assessment publishedBy this siteGary Marcus

    If Altman can't be trusted, safety at OpenAI could become a serious issue.

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

    Assessment marcus-disturbing-and-misleading-efforts-86738f-a1.

  79. Assessment publishedBy this siteGary Marcus

    OpenAI does not behave like a nonprofit and has not for a long time.

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

    Assessment marcus-disturbing-and-misleading-efforts-63fdc0-a1.

  80. Assessment publishedBy this siteGary Marcus

    If OpenAI cannot assemble a board that respects its legal filings and keep its promises, it must dissolve the nonprofit.

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

    Assessment marcus-disturbing-and-misleading-efforts-5e7e65-a1.

  81. Assessment publishedBy this siteGary Marcus

    OpenAI's board should work to prevent AI causing significant harm to the world rather than protecting Altman.

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

    Assessment marcus-disturbing-and-misleading-efforts-481508-a1.

  82. Assessment publishedBy this siteGary Marcus

    ChatGPT, unlike the earlier GPT-3.5 API, was a consumer product using gimmicks like word-by-word typing to give a false impression of humanness.

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

    Assessment marcus-disturbing-and-misleading-efforts-23cde4-a2.

  83. HeldBy this siteGary Marcus

    Held record marcus-disturbing-and-misleading-efforts-1a828b

    The openai screen reads the claim differently.

  84. HeldBy this siteGary Marcus

    Held record marcus-disturbing-and-misleading-efforts-0d5eeb

    The final-text check failed: citations_correct.

  85. Assessment publishedBy this siteGary Marcus

    OpenAI is a registered nonprofit receiving tax-exempt status based on its promise to deploy AI safely.

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

    Assessment marcus-disturbing-and-misleading-efforts-059542-a2.

  86. Assessment publishedBy this siteGary Marcus

    If pushing premature AI too hard and too fast into the military leads to something really bad, Trump will be held responsible.

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

    Assessment marcus-did-trump-just-overplay-his-hand-a379d0-a2.

  87. Assessment publishedBy this siteGary Marcus

    Anthropic refused to back down in its dispute with the US government over military use of its AI.

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

    Assessment marcus-did-trump-just-overplay-his-hand-1f649a-a1.

  88. HeldBy this siteGary Marcus

    Held record marcus-did-trump-just-overplay-his-hand-1a7c41

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

  89. Assessment publishedBy this siteGary Marcus

    Trump's doubling down against Anthropic puts Sam Altman and the whole of Silicon Valley in a serious bind.

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

    Assessment marcus-did-trump-just-overplay-his-hand-1a7c41-2-a1.

  90. Assessment publishedBy this siteGary Marcus

    Gary Marcus guesses that, in time, Trump will be remembered more for his AI policies than for ICE or his tariffs, and negatively.

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

    Assessment marcus-did-trump-just-overplay-his-hand-0acdb3-a1.

  91. HeldBy this siteGary Marcus

    Held record marcus-did-trump-just-overplay-his-hand-00adfc

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

  92. Assessment publishedBy this siteGary Marcus

    The Transformer, invented by a team of Google researchers in 2017, is arguably the biggest advance in AI in the last 10 years.

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

    Assessment marcus-did-googleai-just-snooker-one-of-fdc235-a1.

  93. HeldBy this siteGary Marcus

    Held record marcus-did-googleai-just-snooker-one-of-f533b0

    The anthropic screen reads the claim differently.

  94. Assessment publishedBy this siteGary Marcus

    Currently popular vision-language systems cannot reliably tell apart captions that use the same words in a different order.

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

    Assessment marcus-did-googleai-just-snooker-one-of-c24cf3-a1.

  95. Assessment publishedBy this siteGary Marcus

    Having a Silicon Valley celebrity like Alexander weigh in for Google is great PR for Google but not so great for the public understanding of AI.

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

    Assessment marcus-did-googleai-just-snooker-one-of-a0ae4f-a1.

  96. Assessment publishedBy this siteGary Marcus

    A will to believe in AI will never replace the need for careful science.

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

    Assessment marcus-did-googleai-just-snooker-one-of-95f7fa-a1.

  97. HeldBy this siteGary Marcus

    Held record marcus-did-googleai-just-snooker-one-of-951f5e

    e1 is dated 2022-09, which is not inside the period described (2022-09-01 to 2022-09-14); it can only be context

  98. Assessment publishedBy this siteGary Marcus

    AI has hit a wall, and that wall is mainly compositionality.

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

    Assessment marcus-did-googleai-just-snooker-one-of-90c2b7-a1.

  99. Assessment publishedBy this siteGary Marcus

    Every claimed solution to compositionality so far has proven unreliable and falls apart on careful inspection.

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

    Assessment marcus-did-googleai-just-snooker-one-of-8f381c-a1.

  100. Assessment publishedBy this siteGary Marcus

    Four and a half years after its announcement, Google Duplex has not progressed much past its limited demo stage.

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

    Assessment marcus-did-googleai-just-snooker-one-of-8878ca-a1.

  101. HeldBy this siteGary Marcus

    Held record marcus-did-googleai-just-snooker-one-of-7c3b60

    The final-text check failed: proposition_faithful, citations_correct.

  102. HeldBy this siteGary Marcus

    Held record marcus-did-googleai-just-snooker-one-of-7a607b

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

  103. Assessment publishedBy this siteGary Marcus

    Alexander's test shows only that Imagen beat DALL-E 2 on a handful of items, not that Imagen systematically understands compositionality.

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

    Assessment marcus-did-googleai-just-snooker-one-of-79d6ee-a2.

  104. HeldBy this siteGary Marcus

    Held record marcus-did-googleai-just-snooker-one-of-6f08ae

    The final-text check failed: citations_correct.

  105. Assessment publishedBy this siteGary Marcus

    Current AI systems are still very weak at compositionality, which is central to how language works.

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

    Assessment marcus-did-googleai-just-snooker-one-of-6bfe53-a1.

  106. Assessment publishedBy this siteGary Marcus

    No one, Imagen included, has publicly shown a machine that relates sentence meanings to their parts as well as a five-year-old.

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

    Assessment marcus-did-googleai-just-snooker-one-of-648cfa-a1.

  107. Assessment publishedBy this siteGary Marcus

    Google probably tested Imagen on Winoground, got poor results, and doesn't want to admit it.

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

    Assessment marcus-did-googleai-just-snooker-one-of-2a9aa8-a1.

  108. HeldBy this siteGary Marcus

    Held record marcus-did-googleai-just-snooker-one-of-29b250

    The openai review reads the claim differently.

  109. Assessment publishedBy this siteGary Marcus

    Google has systematically refused to let the scientific community inspect models such as Imagen and LaMDA.

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

    Assessment marcus-did-googleai-just-snooker-one-of-1abf74-a1.

  110. HeldBy this siteGary Marcus

    Held record marcus-did-elon-musk-just-muk-sam-altman-ea9c02

    The openai screen reads the claim differently.

  111. HeldBy this siteGary Marcus

    Held record marcus-did-elon-musk-just-muk-sam-altman-e5fc71

    The anthropic screen reads the claim differently.

  113. HeldBy this siteGary Marcus

    Held record marcus-did-elon-musk-just-muk-sam-altman-a92df8

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

  114. Assessment publishedBy this siteGary Marcus

    Musk's bid may slow down Altman/OpenAI, which can ill afford an extra $50B+ outlay given its cash flow.

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

    Assessment marcus-did-elon-musk-just-muk-sam-altman-414435-a1.

  115. Assessment publishedBy this siteGary Marcus

    Sam Altman urgently needs to get rid of OpenAI's nonprofit control structure, which has become a burden to him.

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

    Assessment marcus-did-elon-musk-just-muk-sam-altman-393323-a1.

  118. Assessment publishedBy this siteGary Marcus

    The Trump administration's tariff-rate formula was probably recommended to it by an LLM.

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

    Assessment marcus-did-an-llm-help-write-trumps-trade-17b59a-a1.

  119. HeldBy this siteGary Marcus

    Held record marcus-deepmind-and-openai-achieve-imo-gold-faf188

    e2 is dated 2025-07, which is not inside the period described (2024-01-01 to 2025-07-22); it can only be context

  120. Assessment publishedBy this siteGary Marcus

    OpenAI communicated very little with the IMO before announcing its gold-level result.

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

    Assessment marcus-deepmind-and-openai-achieve-imo-gold-dd289d-a1.

  121. HeldBy this siteGary Marcus

    Held record marcus-deepmind-and-openai-achieve-imo-gold-dc0eea

    The final-text check failed: proposition_faithful.

  122. Assessment publishedBy this siteGary Marcus

    Some of the hype around the AI IMO gold results has reached completely absurd levels.

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

    Assessment marcus-deepmind-and-openai-achieve-imo-gold-d81e6e-a1.

  125. HeldBy this siteGary Marcus

    Held record marcus-deepmind-and-openai-achieve-imo-gold-a91c11

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

  126. HeldBy this siteGary Marcus

    Held record marcus-deepmind-and-openai-achieve-imo-gold-a782ed

    e1 is dated 2025-07, which is not inside the period described (2025-07-01 to 2025-07-22); it can only be context; e2 is dated 2025-07, which is not inside the period described (2025-07-01 to 2025-07-22); it can only be context; e3 is dated 2025-07, which is not inside the period described (2025-07-01 to 2025-07-22); it can only be context

  128. Assessment publishedBy this siteGary Marcus

    OpenAI-IMO's atypical scratchy proof style strongly suggests it is not simply a standard LLM.

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

    Assessment marcus-deepmind-and-openai-achieve-imo-gold-9af00b-a1.

  129. Assessment publishedBy this siteGary Marcus

    Only one theorem of middling importance (the Robbins conjecture, 1996) has ever been proved autonomously by an AI system.

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

    Assessment marcus-deepmind-and-openai-achieve-imo-gold-82c1f1-a1.

  130. HeldBy this siteGary Marcus

    Held record marcus-deepmind-and-openai-achieve-imo-gold-7bd0cb

    The openai screen reads the claim differently.

  133. Assessment publishedBy this siteGary Marcus

    The new work certainly might be exciting, but we won't really know for sure what it all means until outside scientists get to do a serious review.

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

    Assessment marcus-deepmind-and-openai-achieve-imo-gold-57454d-a1.

  135. Assessment publishedBy this siteGary Marcus

    There will not be AI-based world-class mathematicians within a year (by about July 2026).

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

    Assessment marcus-deepmind-and-openai-achieve-imo-gold-539a58-a2.

  136. Assessment publishedBy this siteGary Marcus

    Neither DeepMind nor OpenAI has given substantial information on how their IMO systems are designed and trained.

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

    Assessment marcus-deepmind-and-openai-achieve-imo-gold-433a30-a1.

  137. Assessment publishedBy this siteGary Marcus

    The results from OpenAI and DeepMind, whose AI systems reached gold-medal level at the IMO, are without question impressive accomplishments.

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

    Assessment marcus-deepmind-and-openai-achieve-imo-gold-3e73d6-a1.

  138. Assessment publishedBy this siteGary Marcus

    Absent disclosure, there is little reason to assume the techniques behind the IMO results will generalize to other kinds of problems.

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

    Assessment marcus-deepmind-and-openai-achieve-imo-gold-3764c8-a1.

  141. Assessment publishedBy this siteGary Marcus

    Meta AI's new text-to-video software Make a Video is straight-up amazing.

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

    Assessment marcus-deepfakes-versus-deep-understanding-d34817-a1.

  142. Assessment publishedBy this siteGary Marcus

    Language systems repeatedly show the same mix of fluent output and lack of understanding.

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

    Assessment marcus-deepfakes-versus-deep-understanding-b2d26e-a1.

  143. Assessment publishedBy this siteGary Marcus

    Dystopian, Black Mirror-style AI harms such as deepfake fraud have already arrived, sooner than expected.

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

    Assessment marcus-deepfaked-shit-is-getting-real-d461db-a1.

  144. Assessment publishedBy this siteGary Marcus

    The situation in which AI developers do what they do scot-free while society picks up the tab for the consequences needs to change.

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

    Assessment marcus-deepfaked-shit-is-getting-real-7fc416-a1.

  145. Assessment publishedBy this siteGary Marcus

    The deepfake scam worked even on an employee who was suspicious rather than naive.

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

    Assessment marcus-deepfaked-shit-is-getting-real-755531-a1.

  146. Assessment publishedBy this siteGary Marcus

    Deepfaked participants, including a deepfaked CFO, on a video call scammed a company out of $25 million.

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

    Assessment marcus-deepfaked-shit-is-getting-real-2365b0-a1.

  147. Assessment publishedBy this siteGary Marcus

    AI developers expect to operate without paying for the harms they cause, leaving society to bear the costs, much as polluting factories once did.

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

    Assessment marcus-deepfaked-shit-is-getting-real-0fa077-a1.

  148. Assessment publishedBy this siteGary Marcus

    The larger problem with Deep Research may be naive people who trust its outputs or use it to write journal articles, missing its errors.

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

    Assessment marcus-deep-research-deep-bullshit-and-the-f6bd30-a1.

  149. Assessment publishedBy this siteGary Marcus

    Deep Research will flood the medical field with garbage content, causing serious trouble.

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

    Assessment marcus-deep-research-deep-bullshit-and-the-c1ac0c-a1.

  150. Assessment publishedBy this siteGary Marcus

    OpenAI's Deep Research is essentially a fancier remake of Meta's Galactica that can write science-sounding articles on any topic on demand.

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

    Assessment marcus-deep-research-deep-bullshit-and-the-aa51b2-a2.

  151. Assessment publishedBy this siteGary Marcus

    His 2023 conjecture that LLMs would be used to write garbage content undermining the internet's value has already been confirmed to some degree.

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

    Assessment marcus-deep-research-deep-bullshit-and-the-9f9465-a1.

  152. Assessment publishedBy this siteGary Marcus

    The scientific literature itself, not just models, will be infected by Deep Research output.

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

    Assessment marcus-deep-research-deep-bullshit-and-the-9dc1c6-a1.

  154. HeldBy this siteGary Marcus

    Held record marcus-deep-research-deep-bullshit-and-the-8e35f7

    The final-text check failed: citations_correct.

  155. HeldBy this siteGary Marcus

    Held record marcus-deep-research-deep-bullshit-and-the-88c867

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

  156. Assessment publishedBy this siteGary Marcus

    Scammers will use Deep Research to write half-baked 'medical reports' to sell ads on websites.

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

    Assessment marcus-deep-research-deep-bullshit-and-the-77d4d1-a1.

  157. Assessment publishedBy this siteGary Marcus

    Few people will carefully fact-check Deep Research's output, since it will all superficially look ok.

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

    Assessment marcus-deep-research-deep-bullshit-and-the-5f6b63-a1.

  158. Assessment publishedBy this siteGary Marcus

    Deep Research output will inevitably be used to train further models, contributing to model collapse.

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

    Assessment marcus-deep-research-deep-bullshit-and-the-5ecd5c-a1.

  159. HeldBy this siteGary Marcus

    Held record marcus-deep-research-deep-bullshit-and-the-599e37

    The openai screen reads the claim differently.

  160. Assessment publishedBy this siteGary Marcus

    Current AI cannot reliably get synthesis of online information and multi-step research right.

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

    Assessment marcus-deep-research-deep-bullshit-and-the-468f01-a2.

  161. HeldBy this siteGary Marcus

    Held record marcus-deep-research-deep-bullshit-and-the-44a074

    The final-text check failed: citations_correct.

  162. Assessment publishedBy this siteGary Marcus

    Virtually all Deep Research output will superficially look good and people will assume it is legitimate even when it is not.

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

    Assessment marcus-deep-research-deep-bullshit-and-the-245106-a1.

  164. Assessment publishedBy this siteGary Marcus

    Deep Research will be heavily used.

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

    Assessment marcus-deep-research-deep-bullshit-and-the-0a2e57-a2.

  165. Assessment publishedBy this siteGary Marcus

    Deep learning is hitting a wall in the sense of reaching diminishing returns.

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

    Assessment marcus-deep-learning-deep-scandal-e25773-a2.

  167. Assessment publishedBy this siteGary Marcus

    Meta's Llama 4 is a failed attempt to create GPT-5 level AI.

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

    Assessment marcus-deep-learning-deep-scandal-cbd098-a2.

  168. Assessment publishedBy this siteGary Marcus

    The AI community is disappointed with Llama 4's performance.

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

    Assessment marcus-deep-learning-deep-scandal-aa2b5c-a1.

  169. Assessment publishedBy this siteGary Marcus

    Real-world customers, many of whom are disappointed, are rarely mentioned in the pundits' discussion of AI progress.

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

    Assessment marcus-deep-learning-deep-scandal-9c7ccd-a1.

  170. HeldBy this siteGary Marcus

    Held record marcus-deep-learning-deep-scandal-9a2f9e

    a observation claim cannot be assessed on the causal_assessment axis

  171. Assessment publishedBy this siteGary Marcus

    As of the post's publication on April 7, 2025, OpenAI, xAI (Grok), Google and Anthropic had each failed to reach what Marcus calls "GPT-5 level" AI.

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

    Assessment marcus-deep-learning-deep-scandal-910d3f-a1.

  172. Assessment publishedBy this siteGary Marcus

    LLMs are no longer living up to expectations, and their vendors appear to be making dodgy choices to conceal this.

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

    Assessment marcus-deep-learning-deep-scandal-4c262d-a1.

  173. Assessment publishedBy this siteGary Marcus

    Meta or OpenAI wanting people to believe something does not by itself make it true.

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

    Assessment marcus-deep-learning-deep-scandal-28e01c-a1.

  174. HeldBy this siteGary Marcus

    Held record marcus-deconstructing-geoffrey-hintons-weakest-f6c34c

    The openai review reads the claim differently.

  175. HeldBy this siteGary Marcus

    Held record marcus-deconstructing-geoffrey-hintons-weakest-eb8b67

    a observation claim cannot be assessed on the causal_assessment axis

  176. Assessment publishedBy this siteGary Marcus

    Neural networks effectively memorize some of what they are trained on, and there is now a whole subfield devoted to this topic.

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

    Assessment marcus-deconstructing-geoffrey-hintons-weakest-d3a2f9-a1.

  177. Assessment publishedBy this siteGary Marcus

    LLMs sometimes regurgitate large chunks of effectively stored text and do not understand anything at a deep level.

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

    Assessment marcus-deconstructing-geoffrey-hintons-weakest-cb96b8-a1.

  178. HeldBy this siteGary Marcus

    Held record marcus-deconstructing-geoffrey-hintons-weakest-b45ae7

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

  179. Assessment publishedBy this siteGary Marcus

    Because of their huge training data and parameter counts, LLM reconstruction of inputs sometimes comes very close to memorization.

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

    Assessment marcus-deconstructing-geoffrey-hintons-weakest-9041dd-a1.

  180. Assessment publishedBy this siteGary Marcus

    By not even acknowledging, let alone challenging, the growing literature on memorization, Hinton made himself look seriously out of touch.

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

    Assessment marcus-deconstructing-geoffrey-hintons-weakest-7d9b37-a1.

  181. HeldBy this siteGary Marcus

    Held record marcus-deconstructing-geoffrey-hintons-weakest-630cf0

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

  184. Assessment publishedBy this siteGary Marcus

    Given enough data, LLMs have demonstrably duplicated lengthy training materials multiple times (memorization).

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

    Assessment marcus-deconstructing-geoffrey-hintons-weakest-530eb6-a1.

  185. HeldBy this siteGary Marcus

    Held record marcus-deconstructing-geoffrey-hintons-weakest-475774

    a observation claim cannot be assessed on the causal_assessment axis

  186. Assessment publishedBy this siteGary Marcus

    Reconstruction of training text by large language models can happen with some regularity.

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

    Assessment marcus-deconstructing-geoffrey-hintons-weakest-400537-a1.

  188. HeldBy this siteGary Marcus

    Held record marcus-deconstructing-geoffrey-hintons-weakest-2d2932

    The anthropic screen reads the claim differently.

  189. Assessment publishedBy this siteGary Marcus

    The NYT v. OpenAI lawsuit showed about a hundred examples of ChatGPT reconstructing entire stretches of text from brief prompts.

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

    Assessment marcus-deconstructing-geoffrey-hintons-weakest-2cf784-a1.

  190. Assessment publishedBy this siteGary Marcus

    AI systems can be good at regurgitating known code while still facing enormous challenges generalizing to new problems.

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

    Assessment marcus-decoding-and-debunking-hard-forks-e70745-a1.

  191. Assessment publishedBy this siteGary Marcus

    DeepSeek has shown that the US's once-strong lead is shrinking.

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

    Assessment marcus-decoding-and-debunking-hard-forks-d3b5d0-a2.

  192. Assessment publishedBy this siteGary Marcus

    Errors in code from automatic code-writing systems will likely persist for decades, especially for genuinely novel systems.

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

    Assessment marcus-decoding-and-debunking-hard-forks-b8be22-a1.

  193. Assessment publishedBy this siteGary Marcus

    Uncritical AI hype by journalists strengthens the tech oligarchy, and society bears the costs.

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

    Assessment marcus-decoding-and-debunking-hard-forks-adfcce-a1.

  194. Assessment publishedBy this siteGary Marcus

    On one standardized measure, over a third of GPT-4.5's outputs are hallucinations.

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

    Assessment marcus-decoding-and-debunking-hard-forks-a7e70f-a1.

  195. HeldBy this siteGary Marcus

    Held record marcus-decoding-and-debunking-hard-forks-9cf49d

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

  196. HeldBy this siteGary Marcus

    Held record marcus-decoding-and-debunking-hard-forks-83fb4f

    The openai screen reads the claim differently.

  197. HeldBy this siteGary Marcus

    Held record marcus-decoding-and-debunking-hard-forks-7ee243

    The anthropic screen reads the claim differently.

  198. Assessment publishedBy this siteGary Marcus

    In AI it is often easy to reach 80% correctness and extremely difficult to get the last 20%, as with driverless cars for decades.

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

    Assessment marcus-decoding-and-debunking-hard-forks-69efcf-a1.

  200. Assessment publishedBy this siteGary Marcus

    Errors in AI-generated code for something like an automatic email writing system could cause serious harm.

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

    Assessment marcus-decoding-and-debunking-hard-forks-5cf705-a1.