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

Changes, page 18 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 18 of 37: changes 3401 to 3600 of 7383, newest first. Dates are when the change was recorded.

  1. Assessment publishedBy this siteGary Marcus

    The dominant, heavily funded current AI approach is unlikely to work.

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

    Assessment marcus-technoptimism-ted-and-the-road-to-1ce1cb-2-a1.

  2. Assessment publishedBy this siteGary Marcus

    Marcus thinks venture capitalists will have a huge role in whether or not we get to the promised land of world-improving technology.

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

    Assessment marcus-technoptimism-ted-and-the-road-to-19dce2-a1.

  3. Assessment publishedBy this siteGary Marcus

    Deploying a billion robots with only superficial understanding of the physical world would lead to disaster.

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

    Assessment marcus-technoptimism-ted-and-the-road-to-05ff15-a2.

  4. HeldBy this siteGary Marcus

    Held record marcus-technology-review-jumps-the-shark-927767

    The final-text check failed: citations_correct.

  5. Assessment publishedBy this siteGary Marcus

    AI is unlikely by end of 2025 to reliably drive a car in novel locations with unusual circumstances without human assistance.

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

    Assessment marcus-superhuman-agi-is-not-nigh-fade4b-a1.

  6. HeldBy this siteGary Marcus

    Held record marcus-superhuman-agi-is-not-nigh-f76683

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

  7. Assessment publishedBy this siteGary Marcus

    AI is unlikely by end of 2025 to tend to the physical and psychological needs of an elderly or infirm person.

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

    Assessment marcus-superhuman-agi-is-not-nigh-eec063-a1.

  9. HeldBy this siteGary Marcus

    Held record marcus-superhuman-agi-is-not-nigh-9e0473

    The anthropic screen doubts the attribution.

  10. Assessment publishedBy this siteGary Marcus

    AI is unlikely by end of 2025 to learn to ride a mountain bike off-road through forest trails.

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

    Assessment marcus-superhuman-agi-is-not-nigh-9d0704-a1.

  11. Assessment publishedBy this siteGary Marcus

    AI is unlikely by end of 2025 to clean unfamiliar houses and apartments.

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

    Assessment marcus-superhuman-agi-is-not-nigh-8efb7e-a1.

  12. Assessment publishedBy this siteGary Marcus

    AI is unlikely by end of 2025 to drive off-road vehicles without maps across streams and around obstacles.

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

    Assessment marcus-superhuman-agi-is-not-nigh-83d446-a1.

  13. Assessment publishedBy this siteGary Marcus

    AI is unlikely by end of 2025 to babysit children in unfamiliar homes and keep them safe.

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

    Assessment marcus-superhuman-agi-is-not-nigh-6f6d5a-a1.

  14. Assessment publishedBy this siteGary Marcus

    AI is unlikely by end of 2025 to walk into random kitchens and prepare coffee or breakfast.

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

    Assessment marcus-superhuman-agi-is-not-nigh-68a824-a1.

  15. Assessment publishedBy this siteGary Marcus

    As of April 2024 there are likely hundreds or more pieces of intellectual work that the best humans can do and AI cannot.

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

    Assessment marcus-superhuman-agi-is-not-nigh-5a5723-a1.

  16. Assessment publishedBy this siteGary Marcus

    None of the listed tasks seem within immediate reach of AI.

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

    Assessment marcus-superhuman-agi-is-not-nigh-59f186-a1.

  17. Assessment publishedBy this siteGary Marcus

    AI is unlikely by end of 2025 to watch an unseen mainstream movie, follow plot twists, know when to laugh, and summarize it without spoilers or fabrications.

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

    Assessment marcus-superhuman-agi-is-not-nigh-4116ad-a1.

  18. Assessment publishedBy this siteGary Marcus

    AI is unlikely by end of 2025 to write cogent, persuasive legal briefs without hallucinating any cases.

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

    Assessment marcus-superhuman-agi-is-not-nigh-047414-a1.

  19. HeldBy this siteGary Marcus

    Held record marcus-sub-optimal-fc985f

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

  20. Assessment publishedBy this siteGary Marcus

    With enough investment, Tesla might in principle catch up to Boston Dynamics in humanoid robotics.

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

    Assessment marcus-sub-optimal-fc14d8-a1.

  21. HeldBy this siteGary Marcus

    Held record marcus-sub-optimal-fa1170

    The anthropic screen reads the claim differently.

  22. HeldBy this siteGary Marcus

    Held record marcus-sub-optimal-f89b27

    The final-text check failed: citations_correct.

  23. Assessment publishedBy this siteGary Marcus

    Humanoid robots pose a set of challenges well beyond current technology's grasp.

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

    Assessment marcus-sub-optimal-f89b27-2-a1.

  24. HeldBy this siteGary Marcus

    Held record marcus-sub-optimal-d58e88

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

  25. Assessment publishedBy this siteGary Marcus

    Meta has recognized that data and compute alone won't suffice, while Musk has not.

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

    Assessment marcus-sub-optimal-d00e95-a1.

  26. Assessment publishedBy this siteGary Marcus

    Musk clearly has not, so far, invested nearly enough to win the robotics race.

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

    Assessment marcus-sub-optimal-bccac7-a1.

  27. HeldBy this siteGary Marcus

    Held record marcus-sub-optimal-b97db5

    The openai screen reads the claim differently.

  28. HeldBy this siteGary Marcus

    Held record marcus-sub-optimal-8375e8

    hedged language is not scored, even on a plain reading

  29. Assessment publishedBy this siteGary Marcus

    Tesla might eventually exceed Boston Dynamics in compute, but very likely could not outgun Google if Google went all in on humanoid robotics.

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

    Assessment marcus-sub-optimal-5aaaa1-a1.

  30. Assessment publishedBy this siteGary Marcus

    Tesla showed very little vision for building the cognitive AI Optimus needs beyond motor control, and little recognition of why real-world robotics is hard.

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

    Assessment marcus-sub-optimal-46aebc-a1.

  31. Assessment publishedBy this siteGary Marcus

    Having the most data and compute and building the biggest model is not inherently a contribution to the innovations needed for AGI.

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

    Assessment marcus-sub-optimal-425b86-a1.

  32. Assessment publishedBy this siteGary Marcus

    The Optimus presentation contained bluster about productivity but no road map.

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

    Assessment marcus-sub-optimal-267fd0-a2.

  33. HeldBy this siteGary Marcus

    Held record marcus-sub-optimal-2501dc

    The anthropic screen reads the claim differently.

  34. Assessment publishedBy this siteGary Marcus

    AI needs paradigm-shift innovation; simply scaling neural networks will not be enough.

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

    Assessment marcus-sub-optimal-1af790-a1.

  35. Assessment publishedBy this siteGary Marcus

    Musk barely acknowledged the common-sense problem and laid out no credible strategy for addressing it.

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

    Assessment marcus-sub-optimal-0f1c2b-a1.

  36. Assessment publishedBy this siteGary Marcus

    AI has a long way to go, and data and compute alone will not be enough.

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

    Assessment marcus-sub-optimal-0d9437-a1.

  37. Assessment publishedBy this siteGary Marcus

    People must stop treating AI models like people.

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

    Assessment marcus-stop-treating-ai-models-like-people-c4efe4-a1.

  38. Assessment publishedBy this siteGary Marcus

    People should not treat chatbots as friends.

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

    Assessment marcus-stop-treating-ai-models-like-people-ae0ed0-a1.

  39. Assessment publishedBy this siteGary Marcus

    New tools like watermarks and detectors, plus policy measures limiting how and where AI models can be used, are needed.

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

    Assessment marcus-stop-treating-ai-models-like-people-a600cd-a1.

  40. Assessment publishedBy this siteGary Marcus

    Educating people to overcome overattribution bias is vital, and senators and AI community members should not worsen it.

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

    Assessment marcus-stop-treating-ai-models-like-people-8e07da-a1.

  41. Assessment publishedBy this siteGary Marcus

    The more false agency people ascribe to chatbots, the more they can be exploited via fraud, catfishing, chatbot therapy or flawed financial advice.

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

    Assessment marcus-stop-treating-ai-models-like-people-531a05-a1.

  42. Assessment publishedBy this siteGary Marcus

    Language models powering chatbots compute word-sequence probabilities without deep or human-like comprehension of what they say.

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

    Assessment marcus-stop-treating-ai-models-like-people-43c07d-a1.

  43. HeldBy this siteGary Marcus

    Held record marcus-stop-treating-ai-models-like-people-40ed31

    The anthropic screen reads the claim differently.

  45. Assessment publishedBy this siteGary Marcus

    GPT-3 has no idea what grape juice or cranberry juice is, what pouring, sniffing, smelling or drinking are, or what it means to be dead.

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

    Assessment marcus-statistics-versus-understanding-the-dc72d7-a1.

  46. HeldBy this siteGary Marcus

    Held record marcus-statistics-versus-understanding-the-9b6676

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

  48. Assessment publishedBy this siteGary Marcus

    Generative AI systems have always tried to use statistics as a proxy for deeper understanding, and it has never entirely worked.

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

    Assessment marcus-statistics-versus-understanding-the-881ba8-a1.

  49. Assessment publishedBy this siteGary Marcus

    In Marcus and Davis's experiments, GPT-3 produced outputs showing a complete failure to understand human biology.

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

    Assessment marcus-statistics-versus-understanding-the-542efb-a1.

  50. HeldBy this siteGary Marcus

    Held record marcus-statistics-versus-understanding-the-325195

    The final-text check failed: citations_correct.

  51. Assessment publishedBy this siteGary Marcus

    Foundation models (generative AI) have never provided the firm, reliable foundation their name implies.

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

    Assessment marcus-statistics-versus-understanding-the-19ea09-a1.

  52. HeldBy this siteGary Marcus

    Held record marcus-statistics-versus-understanding-the-1787b1

    The openai screen reads the claim differently.

  53. Assessment publishedBy this siteGary Marcus

    Grok's generated video failed to keep objects consistent, with the bass tuners appearing and disappearing.

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

    Assessment marcus-spinal-tap-meets-john-malkovich-f1c9ad-a2.

  54. HeldBy this siteGary Marcus

    Held record marcus-spinal-tap-meets-john-malkovich-a8e949

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

  55. Assessment publishedBy this siteGary Marcus

    Generative image systems can produce world-like images but do not understand what things are or how they work outside their training data.

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

    Assessment marcus-spinal-tap-meets-john-malkovich-71a62b-a1.

  57. HeldBy this siteGary Marcus

    Held record marcus-spinal-tap-meets-john-malkovich-49e2f3

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

  58. HeldBy this siteGary Marcus

    Held record marcus-spinal-tap-meets-john-malkovich-486745

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

  59. Assessment publishedBy this siteGary Marcus

    Image generation graphics quality has improved vastly since 2022, to a level no 2022 model could reach.

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

    Assessment marcus-spinal-tap-meets-john-malkovich-0500bc-a2.

  60. Assessment publishedBy this siteGary Marcus

    OpenAI's ordinary guardrail classifiers would have prevented the HuggingFace hack, so it does not show uncontrollable AI.

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

    Assessment marcus-sorry-sam-and-elon-we-have-not-reached-fe0ce9-a1.

  61. Assessment publishedBy this siteGary Marcus

    Current AI has not reached a level beyond AGI that can do anything an expert human can and much more.

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

    Assessment marcus-sorry-sam-and-elon-we-have-not-reached-fc6d75-a1.

  62. Assessment publishedBy this siteGary Marcus

    A widely viewed AI projection from May 2026 has very little chance of coming true.

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

    Assessment marcus-sorry-sam-and-elon-we-have-not-reached-f4e55d-a1.

  63. Assessment publishedBy this siteGary Marcus

    Humans have not merged with AI via brain-computer interfaces to become much smarter, so Kurzweil's singularity has not arrived.

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

    Assessment marcus-sorry-sam-and-elon-we-have-not-reached-af42f6-a1.

  64. Assessment publishedBy this siteGary Marcus

    People should not take seriously the CEO claims (Altman, Hassabis, Musk, Huang) that we are in the singularity or have achieved AGI.

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

    Assessment marcus-sorry-sam-and-elon-we-have-not-reached-9cdd4b-a1.

  65. Assessment publishedBy this siteGary Marcus

    CEO claims that the singularity has arrived are not true just because CEOs make them.

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

    Assessment marcus-sorry-sam-and-elon-we-have-not-reached-9649e3-a1.

  67. Assessment publishedBy this siteGary Marcus

    If AI safety is not radically improved, a future superintelligence could be out of human control.

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

    Assessment marcus-sorry-sam-and-elon-we-have-not-reached-6c798c-a1.

  68. Assessment publishedBy this siteGary Marcus

    AI-driven technological growth has not accelerated beyond human control as of July 2026.

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

    Assessment marcus-sorry-sam-and-elon-we-have-not-reached-66636b-a1.

  69. HeldBy this siteGary Marcus

    Held record marcus-sorry-sam-and-elon-we-have-not-reached-5c58ab

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

  70. Assessment publishedBy this siteGary Marcus

    Humans can still shut off current AI systems, and the main reason they don't is greed.

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

    Assessment marcus-sorry-sam-and-elon-we-have-not-reached-4541ce-a1.

  71. Assessment publishedBy this siteGary Marcus

    Superintelligence may arrive someday but most likely not before the end of this decade (2029).

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

    Assessment marcus-sorry-sam-and-elon-we-have-not-reached-33f5ab-a1.

  72. Assessment publishedBy this siteGary Marcus

    Intelligence is multidimensional, and AI may have progressed a lot on some dimensions but much less on others.

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

    Assessment marcus-sorry-sam-and-elon-we-have-not-reached-2a1abe-a1.

  73. HeldBy this siteGary Marcus

    Held record marcus-sorry-sam-and-elon-we-have-not-reached-19360e

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

  74. HeldBy this siteGary Marcus

    Held record marcus-sorry-sam-and-elon-we-have-not-reached-106fee

    e2 is dated 2026-07, which is not inside the period described (… to 2026-07-28); it can only be context

  75. Assessment publishedBy this siteGary Marcus

    Sam Altman's singularity claim merely copies Elon Musk, and Altman's earlier blog on it never defined the term.

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

    Assessment marcus-sorry-sam-and-elon-we-have-not-reached-0d73a7-a1.

  76. Assessment publishedBy this siteGary Marcus

    Programmers should use generative AI to type faster, not as a substitute for clear thinking about algorithms and data structures.

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

    Assessment marcus-sorry-genai-is-not-going-to-10x-computer-dd1890-a1.

  77. Assessment publishedBy this siteGary Marcus

    Generating code without conceptual understanding provides only limited productivity help.

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

    Assessment marcus-sorry-genai-is-not-going-to-10x-computer-d7a038-a1.

  78. Assessment publishedBy this siteGary Marcus

    Less secure code written with AI coding assistants could lead to a net loss of productivity in the long term.

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

    Assessment marcus-sorry-genai-is-not-going-to-10x-computer-a54861-a1.

  79. Assessment publishedBy this siteGary Marcus

    Generative AI cannot 10x programming because doing so requires deep conceptual understanding, which GenAI lacks.

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

    Assessment marcus-sorry-genai-is-not-going-to-10x-computer-8679ce-a1.

  81. Assessment publishedBy this siteGary Marcus

    The FunSearch solution probably isn't all that general.

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

    Assessment marcus-sorry-but-funsearch-probably-isnt-d75615-a1.

  82. HeldBy this siteGary Marcus

    Held record marcus-sorry-but-funsearch-probably-isnt-84c5e6

    The anthropic screen reads the claim differently.

  83. Assessment publishedBy this siteGary Marcus

    FunSearch required human mathematicians to write problem-specific code for each separate problem.

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

    Assessment marcus-sorry-but-funsearch-probably-isnt-63781a-a1.

  84. Assessment publishedBy this siteGary Marcus

    Google DeepMind's PR team oversold the FunSearch paper, ending with an exuberant but immensely speculative set of claims.

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

    Assessment marcus-sorry-but-funsearch-probably-isnt-623aaa-a1.

  85. Assessment publishedBy this siteGary Marcus

    FunSearch is unlikely to be a major factor in solving cancer or making lightweight batteries.

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

    Assessment marcus-sorry-but-funsearch-probably-isnt-4492c9-a1.

  86. Assessment publishedBy this siteGary Marcus

    There is no evidence that the problem FunSearch addressed was previously unsolvable.

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

    Assessment marcus-sorry-but-funsearch-probably-isnt-2de8a6-a1.

  87. Assessment publishedBy this siteGary Marcus

    The claim that FunSearch went 'beyond human knowledge' is a bit oversold.

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

    Assessment marcus-sorry-but-funsearch-probably-isnt-2bb753-a1.

  88. Assessment publishedBy this siteGary Marcus

    FunSearch is not the first time AI has helped with mathematics.

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

    Assessment marcus-sorry-but-funsearch-probably-isnt-2a3cc5-a1.

  90. Assessment publishedBy this siteGary Marcus

    Claims that Sora amounts to AGI, vindicates scaling laws, or constitutes a world model are hyperbolic and unlikely to be true.

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

    Assessment marcus-soras-surreal-physics-f68c83-a1.

  91. Assessment publishedBy this siteGary Marcus

    Many of Sora's physics glitches will be hard to fix.

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

    Assessment marcus-soras-surreal-physics-dff3dd-a1.

  92. HeldBy this siteGary Marcus

    Held record marcus-soras-surreal-physics-da5e54

    The anthropic screen reads the claim differently.

  93. Assessment publishedBy this siteGary Marcus

    More training data will not solve Sora's hallucination-like physics glitches.

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

    Assessment marcus-soras-surreal-physics-cb9e51-a2.

  94. Assessment publishedBy this siteGary Marcus

    Sora may well have been trained on lots of YouTube video and various copyrighted materials.

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

    Assessment marcus-soras-surreal-physics-c6bb6d-a1.

  95. HeldBy this siteGary Marcus

    Held record marcus-soras-surreal-physics-77a2b8

    The anthropic screen reads the claim differently.

  96. Assessment publishedBy this siteGary Marcus

    If a system can't handle the permanence of objects at all, Marcus is not sure it should even be called a world model.

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

    Assessment marcus-soras-surreal-physics-6d2684-a1.

  97. HeldBy this siteGary Marcus

    Held record marcus-soras-surreal-physics-5e107a

    The openai screen reads the claim differently.

  98. Assessment publishedBy this siteGary Marcus

    Sora is akin to morphing and splicing of images, not a path to the physical reasoning needed for AGI.

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

    Assessment marcus-soras-surreal-physics-50bc9e-a1.

  99. Assessment publishedBy this siteGary Marcus

    In Marcus's view, Sora is a distraction as a solution to artificial general intelligence.

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

    Assessment marcus-soras-surreal-physics-4b64dd-a1.

  100. Assessment publishedBy this siteGary Marcus

    Artists are probably being badly harmed by the way Sora was trained.

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

    Assessment marcus-soras-surreal-physics-3c8088-a1.

  101. Assessment publishedBy this siteGary Marcus

    As more people gain access to Sora, more systemic glitches will be found.

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

    Assessment marcus-soras-surreal-physics-18217a-a1.

  102. HeldBy this siteGary Marcus

    Held record marcus-soras-surreal-physics-12448e

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

  103. HeldBy this siteGary Marcus

    Held record marcus-soras-surreal-physics-0677aa

    The anthropic screen reads the claim differently.

  104. Assessment publishedBy this siteGary Marcus

    Sora will never really grasp object permanence, even with vast amounts of data.

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

    Assessment marcus-sora-still-appears-to-have-trouble-ed695f-a1.

  105. Assessment publishedBy this siteGary Marcus

    The author guesses that Sora's failed attempt to learn physics from data may have cost around a billion dollars.

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

    Assessment marcus-sora-still-appears-to-have-trouble-e78401-2-a1.

  106. Assessment publishedBy this siteGary Marcus

    Sora's failures of object permanence and spatiotemporal continuity are general rather than isolated.

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

    Assessment marcus-sora-still-appears-to-have-trouble-e0f584-a1.

  107. Assessment publishedBy this siteGary Marcus

    Sora does not really understand physics and will not come to understand it.

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

    Assessment marcus-sora-still-appears-to-have-trouble-c207f4-a1.

  108. HeldBy this siteGary Marcus

    Held record marcus-sora-cant-handle-the-truth-ffbb46

    The openai screen reads the claim differently.

  109. Assessment publishedBy this siteGary Marcus

    Sora fundamentally works by reconstructing nearby pixels in space and time rather than perceiving whole objects.

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

    Assessment marcus-sora-cant-handle-the-truth-f817ae-a1.

  110. HeldBy this siteGary Marcus

    Held record marcus-sora-cant-handle-the-truth-d0cf91

    The openai screen reads the claim differently.

  111. Assessment publishedBy this siteGary Marcus

    Generative AI has struggled mightily with world models, high-level reasoning and factuality, even with ever more data.

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

    Assessment marcus-sora-cant-handle-the-truth-cfa367-a1.

  112. Assessment publishedBy this siteGary Marcus

    Sora arguably shows some generalization, but in the monkey-chess case it is drawing the wrong generalization.

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

    Assessment marcus-sora-cant-handle-the-truth-b53b7f-a2.

  113. Assessment publishedBy this siteGary Marcus

    The AI industry and the world should recognize soon that generative AI shares a persistent factuality and comprehension problem.

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

    Assessment marcus-sora-cant-handle-the-truth-7bee6a-a1.

  114. HeldBy this siteGary Marcus

    Held record marcus-sora-cant-handle-the-truth-65d869

    The anthropic screen reads the claim differently.

  115. HeldBy this siteGary Marcus

    Held record marcus-sora-cant-handle-the-truth-3a6dd1

    The anthropic screen reads the claim differently.

  116. Assessment publishedBy this siteGary Marcus

    From an AGI perspective, Sora's fixation on the local at the expense of a deep representation of the global is an immense warning sign.

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

    Assessment marcus-sora-cant-handle-the-truth-2d31d7-a1.

  117. Assessment publishedBy this siteGary Marcus

    Sora's erroneous chessboard is probably not due to a lack of training data.

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

    Assessment marcus-sora-cant-handle-the-truth-0d9666-a1.

  118. Assessment publishedBy this siteGary Marcus

    Meta's Cicero deserves more attention than it has received, in part because it can plan.

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

    Assessment marcus-some-things-garymarcus-might-say-11c45f-a1.

  120. Assessment publishedBy this siteGary Marcus

    Some form of AI will eventually yield the returns people anticipated from generative AI.

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

    Assessment marcus-some-disconcerting-facts-about-ai-babe3d-a2.

  121. Assessment publishedBy this siteGary Marcus

    Generative AI has some clear value, but its costs are high and its revenue has been modest relative to those costs.

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

    Assessment marcus-some-disconcerting-facts-about-ai-b94b50-a1.

  122. HeldBy this siteGary Marcus

    Held record marcus-some-disconcerting-facts-about-ai-92546b

    hedged language is not scored, even on a plain reading

  123. Assessment publishedBy this siteGary Marcus

    Generative AI is probably going to lose a lot of money.

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

    Assessment marcus-some-disconcerting-facts-about-ai-6db64c-a1.

  125. Assessment publishedBy this siteGary Marcus

    Generative AI, being unreliable statistical mimicry, is unlikely to deliver the anticipated massive productivity gains.

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

    Assessment marcus-some-disconcerting-facts-about-ai-34b503-a1.

  126. Assessment publishedBy this siteGary Marcus

    A trillion dollars or more invested in AI might be lost, with possible considerable economic and other collateral damage to society.

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

    Assessment marcus-some-disconcerting-facts-about-ai-160801-a1.

  127. Assessment publishedBy this siteGary Marcus

    AI needs new approaches beyond big data and deep learning.

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

    Assessment marcus-smells-a-little-bit-like-ai-winter-e3097c-a1.

  128. HeldBy this siteGary Marcus

    Held record marcus-smells-a-little-bit-like-ai-winter-64336e

    The anthropic screen doubts the attribution.

  129. Assessment publishedBy this siteGary Marcus

    AI that works in the real world cannot be built from big data and deep learning alone.

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

    Assessment marcus-smells-a-little-bit-like-ai-winter-5b54e6-a1.

  130. HeldBy this siteGary Marcus

    Held record marcus-smells-a-little-bit-like-ai-winter-5600ee

    The final-text check failed: citations_correct.

  131. Assessment publishedBy this siteGary Marcus

    The problem of AI slop/enshittification is serious and ubiquitous.

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

    Assessment marcus-slopocalypse-now-6e1233-a2.

  133. Assessment publishedBy this siteGary Marcus

    It is very likely that math research is being flooded with low-quality AI-generated work.

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

    Assessment marcus-slop-productivity-and-why-the-ai-aa79c3-a1.

  134. Assessment publishedBy this siteGary Marcus

    It is hard even to say how the chip companies are doing once circular financing and shrinking cash flow are factored out.

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

    Assessment marcus-slop-productivity-and-why-the-ai-91f873-a1.

  135. Assessment publishedBy this siteGary Marcus

    The AI spending and losses he has just described amount to digging holes without creating real value.

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

    Assessment marcus-slop-productivity-and-why-the-ai-8fc8ec-a1.

  137. Assessment publishedBy this siteGary Marcus

    AI coding, the one possible major productivity success, is also the use case burning the most cash.

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

    Assessment marcus-slop-productivity-and-why-the-ai-57557e-a1.

  138. Assessment publishedBy this siteGary Marcus

    Coding may be the biggest exception to AI's low value, but it is not yet clear how many systems built with agentic coding will endure.

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

    Assessment marcus-slop-productivity-and-why-the-ai-412bfa-a1.

  139. Assessment publishedBy this siteGary Marcus

    Most of the AI-generated content is slop, and in some cases work slop.

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

    Assessment marcus-slop-productivity-and-why-the-ai-3600d5-a1.

  142. Assessment publishedBy this siteGary Marcus

    In many domains AI produces lots of nominal productivity but little real-world impact.

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

    Assessment marcus-slop-productivity-and-why-the-ai-245333-a1.

  143. Assessment publishedBy this siteGary Marcus

    No country will take a decisive lead in the generative AI race in 2026.

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

    Assessment marcus-six-or-seven-predictions-for-ai-2026-ecc887-a1.

  144. HeldBy this siteGary Marcus

    Held record marcus-six-or-seven-predictions-for-ai-2026-e53987

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

  145. HeldBy this siteGary Marcus

    Held record marcus-six-or-seven-predictions-for-ai-2026-db962b

    The final-text check failed: citations_correct.

  146. Assessment publishedBy this siteGary Marcus

    The AI field's intellectual climate has shifted from LLM-only and AGI-is-nigh thinking to a more fluid, realistic, and open-minded one.

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

    Assessment marcus-six-or-seven-predictions-for-ai-2026-ce8b5f-a1.

  147. Assessment publishedBy this siteGary Marcus

    Marcus's 2026 predictions will be less accurate than his 2025 ones because the field has become more fluid, realistic, and open-minded.

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

    Assessment marcus-six-or-seven-predictions-for-ai-2026-ccd8c5-a1.

  148. Assessment publishedBy this siteGary Marcus

    There will be many lab demos of humanoid robots in 2026, but getting them to work in homes will be very hard.

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

    Assessment marcus-six-or-seven-predictions-for-ai-2026-b7460d-a2.

  149. Assessment publishedBy this siteGary Marcus

    AGI will not be achieved in 2026 or 2027.

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

    Assessment marcus-six-or-seven-predictions-for-ai-2026-b34135-a2.

  151. Assessment publishedBy this siteGary Marcus

    In 2026, work on new approaches such as world models and neurosymbolic AI will escalate.

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

    Assessment marcus-six-or-seven-predictions-for-ai-2026-93f081-a1.

  152. HeldBy this siteGary Marcus

    Held record marcus-six-or-seven-predictions-for-ai-2026-763cad

    The final-text check failed: citations_correct.

  153. Assessment publishedBy this siteGary Marcus

    Few AI companies other than Nvidia are profitable, and no AI company has much of a technical moat.

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

    Assessment marcus-six-or-seven-predictions-for-ai-2026-679c99-a1.

  154. Assessment publishedBy this siteGary Marcus

    Sentiment about near-term AGI has shifted markedly in the past few months, with figures like Sutskever and Sutton voicing concerns.

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

    Assessment marcus-six-or-seven-predictions-for-ai-2026-474844-a1.

  155. Assessment publishedBy this siteGary Marcus

    2025 will come to be seen as the peak of the generative AI bubble and when Wall Street began losing confidence in generative AI.

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

    Assessment marcus-six-or-seven-predictions-for-ai-2026-30f525-a1.

  156. HeldBy this siteGary Marcus

    Held record marcus-six-or-seven-predictions-for-ai-2026-1632dd

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

  157. Assessment publishedBy this siteGary Marcus

    In 2026, humanoid domestic robots such as Optimus and Figure will be mostly demos with very little actual product.

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

    Assessment marcus-six-or-seven-predictions-for-ai-2026-148444-a2.

  158. HeldBy this siteGary Marcus

    Held record marcus-six-or-seven-predictions-for-ai-2026-110fa7

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

  159. Assessment publishedBy this siteGary Marcus

    Many would agree that scaling has reached a point of diminishing returns, and faith in scaling as a route to AGI has dissipated.

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

    Assessment marcus-six-or-seven-predictions-for-ai-2026-06fcce-a1.

  160. HeldBy this siteGary Marcus

    Held record marcus-six-or-seven-predictions-for-ai-2026-04a549

    The openai review abstained.

  161. Assessment publishedBy this siteGary Marcus

    Trump may eventually distance himself from AI due to public backlash.

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

    Assessment marcus-six-or-seven-predictions-for-ai-2026-012e4d-a1.

  162. HeldBy this siteGary Marcus

    Held record marcus-six-months-after-the-pause-letter-e5537c

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

  164. HeldBy this siteGary Marcus

    Held record marcus-six-months-after-the-pause-letter-dd5dcf

    The openai screen doubts the attribution.

  165. HeldBy this siteGary Marcus

    Held record marcus-six-months-after-the-pause-letter-daac24

    The openai screen reads the claim differently.

  166. HeldBy this siteGary Marcus

    Held record marcus-six-months-after-the-pause-letter-d31de2

    The anthropic screen reads the claim differently.

  167. HeldBy this siteGary Marcus

    Held record marcus-six-months-after-the-pause-letter-c607b1

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

  168. HeldBy this siteGary Marcus

    Held record marcus-six-months-after-the-pause-letter-c27fc4

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

  169. Assessment publishedBy this siteGary Marcus

    LLMs still hallucinate and their reasoning cannot be trusted.

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

    Assessment marcus-six-months-after-the-pause-letter-aec4cd-a1.

  170. Assessment publishedBy this siteGary Marcus

    Global attention on AI policy has vastly increased since March 2023, and the pause letter deserves some credit for that.

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

    Assessment marcus-six-months-after-the-pause-letter-95b913-a1.

  171. Assessment publishedBy this siteGary Marcus

    There will be no moratorium on AI development in the near term.

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

    Assessment marcus-six-months-after-the-pause-letter-72caee-a1.

  172. HeldBy this siteGary Marcus

    Held record marcus-six-months-after-the-pause-letter-68a174

    The openai screen reads the claim differently.

  175. Assessment publishedBy this siteGary Marcus

    US AI policy remains largely notional rather than actual.

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

    Assessment marcus-six-months-after-the-pause-letter-44e8a8-a1.

  177. Assessment publishedBy this siteGary Marcus

    At least in the United States, voluntary AI guidelines are not laws or enforcement mechanisms, and companies can still do as they please.

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

    Assessment marcus-six-months-after-the-pause-letter-22f558-a1.

  178. HeldBy this siteGary Marcus

    Held record marcus-six-months-after-the-pause-letter-168ffe

    hedged language is not scored, even on a plain reading

  179. Assessment publishedBy this siteGary Marcus

    AGI, defined as a system with all human cognitive capabilities, will not be achieved this decade (by the end of 2029).

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

    Assessment marcus-sir-demis-hassabis-vs-sir-demis-hassabis-9626a8-a1.

  180. Assessment publishedBy this siteGary Marcus

    Hassabis predicted AGI around 2030 in mid-2026 but gave a slower 2031-2036 timeline at Davos in January 2026.

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

    Assessment marcus-sir-demis-hassabis-vs-sir-demis-hassabis-03bd9a-a1.

  181. Assessment publishedBy this siteGary Marcus

    Many machine learning researchers worry about very bad outcomes from AI, yet most people carry on as if they didn't.

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

    Assessment marcus-should-we-worry-more-about-short-f915b5-a1.

  182. HeldBy this siteGary Marcus

    Held record marcus-should-we-worry-more-about-short-ec324a

    The anthropic screen reads the claim differently.

  183. Assessment publishedBy this siteGary Marcus

    The field is a long way from knowing how to build machines with consensus human values.

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

    Assessment marcus-should-we-worry-more-about-short-ec324a-2-a1.

  184. Assessment publishedBy this siteGary Marcus

    Society must not ignore AI risks or assume it has unlimited time to decide how to respond.

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

    Assessment marcus-should-we-worry-more-about-short-e81d0a-a1.

  185. HeldBy this siteGary Marcus

    Held record marcus-should-we-worry-more-about-short-e246ea

    The openai screen reads the claim differently.

  186. Assessment publishedBy this siteGary Marcus

    Society should seriously raise questions about pausing or regulating AI, sooner rather than later.

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

    Assessment marcus-should-we-worry-more-about-short-c304e2-a1.

  187. Assessment publishedBy this siteGary Marcus

    Because of AI-generated misinformation, the 2024 elections could be unlike any elections seen before.

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

    Assessment marcus-should-we-worry-more-about-short-b10a34-a1.

  188. Assessment publishedBy this siteGary Marcus

    LLM technology has already partly escaped containment.

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

    Assessment marcus-should-we-worry-more-about-short-8ae777-a1.

  189. Assessment publishedBy this siteGary Marcus

    Fighting AI-generated misinformation will be an uphill, ongoing arms race of moves and countermoves.

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

    Assessment marcus-should-we-worry-more-about-short-8ad9b7-a1.

  190. Assessment publishedBy this siteGary Marcus

    If we don't start fighting misinformation now, democracy may well be overwhelmed by misinformation and consequent polarization, perhaps quite soon.

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

    Assessment marcus-should-we-worry-more-about-short-5aa65e-a1.

  191. Assessment publishedBy this siteGary Marcus

    ChatGPT is only dimly related to artificial general intelligence, but it serves as a dress rehearsal for it.

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

    Assessment marcus-should-we-worry-more-about-short-563a80-a1.

  192. HeldBy this siteGary Marcus

    Held record marcus-should-we-worry-more-about-short-54ba15

    The openai screen reads the claim differently.

  193. Assessment publishedBy this siteGary Marcus

    AI deployment should be governed by stricter regulations, rather than a full pause on AI.

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

    Assessment marcus-should-we-worry-more-about-short-5456c7-a1.

  195. Assessment publishedBy this siteGary Marcus

    Google repeating the same kind of LLM error two years apart mocks claims of exponential progress in AI.

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

    Assessment marcus-shame-on-google-twice-e8ac0e-a1.

  196. HeldBy this siteGary Marcus

    Held record marcus-shame-on-google-twice-686a00

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

  197. HeldBy this siteGary Marcus

    Held record marcus-sex-violence-and-ethics-trying-to-0aadea

    The final-text check failed: citations_correct.

  198. Assessment publishedBy this siteGary Marcus

    Current AI does not yet integrate neural and symbolic methods reliably and generally.

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

    Assessment marcus-seven-replies-to-the-viral-apple-ff7c91-a1.

  199. Assessment publishedBy this siteGary Marcus

    None of the roughly seven rejoinders to the Apple paper reviewed in this post are compelling.

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

    Assessment marcus-seven-replies-to-the-viral-apple-f8d3d1-a1.

  200. Assessment publishedBy this siteGary Marcus

    The output length limit of LLMs is a bug, and most certainly not a feature.

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

    Assessment marcus-seven-replies-to-the-viral-apple-f885d0-a1.