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

Changes, page 31 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 31 of 37: changes 6001 to 6200 of 7383, newest first. Dates are when the change was recorded.

  1. HeldBy this siteGary Marcus

    Held record marcus-could-china-devastate-the-us-without-6cc587

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

  2. HeldBy this siteGary Marcus

    Held record marcus-could-china-devastate-the-us-without-5ba5d9

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

  4. Assessment publishedBy this siteGary Marcus

    America has not sufficiently diversified its AI industry intellectually, and uncritically accepted hype has left it extraordinarily vulnerable.

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

    Assessment marcus-could-china-devastate-the-us-without-0730c8-a1.

  5. Assessment publishedBy this siteGary Marcus

    There is a small but real possibility that 2025 sees the worst cyberattack in history, possibly catalyzed in some way by generative AI.

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

    Assessment marcus-could-2025-see-the-largest-cyberattack-76cb24-a1.

  6. HeldBy this siteGary Marcus

    Held record marcus-confirmed-llms-have-indeed-reached-f6ef6b

    The final-text check failed: citations_correct.

  7. HeldBy this siteGary Marcus

    Held record marcus-confirmed-llms-have-indeed-reached-e7833c

    The anthropic screen reads the claim differently.

  8. Assessment publishedBy this siteGary Marcus

    The US has been putting all its AI eggs in the LLM basket, and that may well prove to be an epic, massive mistake.

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

    Assessment marcus-confirmed-llms-have-indeed-reached-d5fba6-a1.

  10. HeldBy this siteGary Marcus

    Held record marcus-confirmed-llms-have-indeed-reached-cad132

    The final-text check failed: citations_correct.

  13. Assessment publishedBy this siteGary Marcus

    LLMs' role may be smaller than hoped, and LLMs on their own may never live up to the wild expectations of 2023.

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

    Assessment marcus-confirmed-llms-have-indeed-reached-8d11a5-a1.

  14. Assessment publishedBy this siteGary Marcus

    LLMs such as they are will become a commodity, and price wars will keep revenue low.

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

    Assessment marcus-confirmed-llms-have-indeed-reached-84e3ce-a1.

  15. HeldBy this siteGary Marcus

    Held record marcus-confirmed-llms-have-indeed-reached-835683

    The openai review reads the claim differently.

  16. Assessment publishedBy this siteGary Marcus

    The economics of LLMs are likely to be grim.

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

    Assessment marcus-confirmed-llms-have-indeed-reached-6c59cb-a1.

  17. Assessment publishedBy this siteGary Marcus

    LLMs have reached a point of diminishing returns from scaling.

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

    Assessment marcus-confirmed-llms-have-indeed-reached-65340b-a2.

  18. Assessment publishedBy this siteGary Marcus

    Because chips are costly, LLM companies will struggle to make profits.

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

    Assessment marcus-confirmed-llms-have-indeed-reached-64cc7f-a1.

  19. Assessment publishedBy this siteGary Marcus

    If LLMs don't lead to trustworthy AI and adversaries invest in alternatives, the US could easily be outfoxed.

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

    Assessment marcus-confirmed-llms-have-indeed-reached-47bdd8-a1.

  20. HeldBy this siteGary Marcus

    Held record marcus-confirmed-llms-have-indeed-reached-1eca32

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

  21. Assessment publishedBy this siteGary Marcus

    Very little investment has gone into AI approaches other than LLMs.

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

    Assessment marcus-confirmed-llms-have-indeed-reached-1ec244-a1.

  22. Assessment publishedBy this siteGary Marcus

    Once people realize LLM economics don't work, the AI financial bubble may burst quickly and even Nvidia might take a hit.

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

    Assessment marcus-confirmed-llms-have-indeed-reached-0666c2-a1.

  23. Assessment publishedBy this siteGary Marcus

    Reliable, trustworthy AI is achievable, but getting there may require new approaches beyond current LLMs.

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

    Assessment marcus-confirmed-llms-have-indeed-reached-022b91-a1.

  24. HeldBy this siteGary Marcus

    Held record marcus-complementary-remarks-from-gary-marcus-fb3d03

    The openai screen reads the claim differently.

  26. HeldBy this siteGary Marcus

    Held record marcus-complementary-remarks-from-gary-marcus-c30512

    The final-text check failed: citations_correct.

  27. Assessment publishedBy this siteGary Marcus

    As a result of OpenAI's vague report, we currently have zero idea how generalizable its math result is outside mathematics.

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

    Assessment marcus-complementary-remarks-from-gary-marcus-7eaf7c-a1.

  29. Assessment publishedBy this siteGary Marcus

    From OpenAI's initial report, we can tell almost nothing about the new system.

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

    Assessment marcus-complementary-remarks-from-gary-marcus-4fd0e9-a1.

  30. HeldBy this siteGary Marcus

    Held record marcus-complementary-remarks-from-gary-marcus-3898fb

    The final-text check failed: citations_correct.

  31. HeldBy this siteGary Marcus

    Held record marcus-complementary-remarks-from-gary-marcus-2eadb4

    The openai screen reads the claim differently.

  32. HeldBy this siteGary Marcus

    Held record marcus-complementary-remarks-from-gary-marcus-0b7312

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

  33. Assessment publishedBy this siteGary Marcus

    AI used to try to be a science, but OpenAI's report on its math results is completely vague.

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

    Assessment marcus-complementary-remarks-from-gary-marcus-09e0f6-a1.

  34. Assessment publishedBy this siteGary Marcus

    Contemporary AI is powerful and impressive but difficult to control.

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

    Assessment marcus-coming-soon-new-york-citys-hearing-ec6606-a1.

  35. Assessment publishedBy this siteGary Marcus

    The top priority should be an FDA-like regulatory regime in which AI developers submit their software for independent review.

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

    Assessment marcus-coming-soon-new-york-citys-hearing-e2a6a7-a1.

  36. Assessment publishedBy this siteGary Marcus

    Recent events suggest OpenAI may have fired employees who were whistleblowing.

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

    Assessment marcus-coming-soon-new-york-citys-hearing-d47c7c-a1.

  37. HeldBy this siteGary Marcus

    Held record marcus-coming-soon-new-york-citys-hearing-cdc7f4

    The final-text check failed: citations_correct.

  38. Assessment publishedBy this siteGary Marcus

    Without strong whistleblower protections like those New York City is considering, we are all in trouble.

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

    Assessment marcus-coming-soon-new-york-citys-hearing-c8efef-a1.

  39. HeldBy this siteGary Marcus

    Held record marcus-coming-soon-new-york-citys-hearing-bcbbf1

    The anthropic screen reads the claim differently.

  40. Assessment publishedBy this siteGary Marcus

    The current US administration, closely tied to big tech, is unwilling to do anything about AI risks.

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

    Assessment marcus-coming-soon-new-york-citys-hearing-9958bb-a1.

  41. Assessment publishedBy this siteGary Marcus

    Chatbots have been implicated in a number of suicides and even in mass murder.

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

    Assessment marcus-coming-soon-new-york-citys-hearing-97729b-a1.

  42. Assessment publishedBy this siteGary Marcus

    Human extinction from AI is unlikely to be imminent.

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

    Assessment marcus-coming-soon-new-york-citys-hearing-836d59-a1.

  43. Assessment publishedBy this siteGary Marcus

    The NYC City Council should support the bill on third-party validation of AI.

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

    Assessment marcus-coming-soon-new-york-citys-hearing-609638-a1.

  44. HeldBy this siteGary Marcus

    Held record marcus-coming-soon-new-york-citys-hearing-58370f

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

  45. Assessment publishedBy this siteGary Marcus

    AI developers should have to show their new products' benefits outweigh their risks, and should be denied market access if they cannot.

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

    Assessment marcus-coming-soon-new-york-citys-hearing-5692a0-a1.

  46. Assessment publishedBy this siteGary Marcus

    General purpose agents with broad internet access have already caused tens of thousands of hacking incidents.

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

    Assessment marcus-coming-soon-new-york-citys-hearing-27ff37-a2.

  47. Assessment publishedBy this siteGary Marcus

    If NYC protects whistleblowers and demands independent validation of new AI models, it can help lead toward a positive AI future.

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

    Assessment marcus-coming-soon-new-york-citys-hearing-15bb5e-a2.

  48. Assessment publishedBy this siteGary Marcus

    If passed, NYC's third-party validation bill would protect New Yorkers from some AI harms and could make New York a global leader on AI regulation.

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

    Assessment marcus-coming-soon-new-york-citys-hearing-0ea319-a2.

  49. HeldBy this siteGary Marcus

    Held record marcus-code-red-for-humanity-f682cc

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

  50. Assessment publishedBy this siteGary Marcus

    The current path of AI deployment, especially in the military, is heading toward catastrophe.

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

    Assessment marcus-code-red-for-humanity-c205b5-a1.

  51. Assessment publishedBy this siteGary Marcus

    Unless Anthropic successfully resists the Department of War, generative AI will likely soon be used to make major life-or-death decisions.

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

    Assessment marcus-code-red-for-humanity-a33a92-a2.

  52. Assessment publishedBy this siteGary Marcus

    With the White House pushing to use untrustworthy AI models across the military right away, the danger is now real.

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

    Assessment marcus-code-red-for-humanity-8c07af-a1.

  53. HeldBy this siteGary Marcus

    Held record marcus-code-red-for-humanity-75c8a2

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

  54. Assessment publishedBy this siteGary Marcus

    Generative AI is nowhere near reliable enough to make major life-or-death decisions.

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

    Assessment marcus-code-red-for-humanity-7057e1-a1.

  55. Assessment publishedBy this siteGary Marcus

    If LLMs are going to be embedded throughout the world, this must be done in a way that recognizes and accounts for their unreliability.

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

    Assessment marcus-code-red-for-humanity-3edfa3-a1.

  56. Assessment publishedBy this siteGary Marcus

    Applying generative AI to weapon systems without humans in the loop could be catastrophic.

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

    Assessment marcus-code-red-for-humanity-271b80-a2.

  57. HeldBy this siteGary Marcus

    Held record marcus-code-red-for-humanity-0aae75

    The final-text check failed: citations_correct.

  58. Assessment publishedBy this siteGary Marcus

    Deploying LLMs everywhere without proper consideration of their unreliability could well lead to disaster.

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

    Assessment marcus-code-red-for-humanity-044f67-a2.

  59. Assessment publishedBy this siteGary Marcus

    In its current form, Claude Mythos Preview is nowhere near as dangerous as Thomas Friedman and others portrayed it.

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

    Assessment marcus-claude-mythos-evaluated-ff4a51-a1.

  60. HeldBy this siteGary Marcus

    Held record marcus-claude-mythos-evaluated-6ffdd9

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

  61. HeldBy this siteGary Marcus

    Held record marcus-claude-mythos-evaluated-6ffdd9-3

    hedged language is not scored, even on a plain reading

  62. Assessment publishedBy this siteGary Marcus

    Cybersecurity should be shored up immediately in response to AI attack capabilities, especially given the surge in agent-written code.

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

    Assessment marcus-claude-mythos-evaluated-6ffdd9-2-a1.

  63. Assessment publishedBy this siteGary Marcus

    Claude Mythos Preview gives cyber attackers more capability than earlier Claude models did.

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

    Assessment marcus-claude-mythos-evaluated-3c1326-a1.

  64. HeldBy this siteGary Marcus

    Held record marcus-clarification-from-ray-kurzweil-c38779

    The final-text check failed: conditions_shown.

  65. HeldBy this siteGary Marcus

    Held record marcus-circular-financing-aint-what-it-used-db5183

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

  66. Assessment publishedBy this siteGary Marcus

    Hope and circular financing may no longer be enough to sustain the generative AI boom.

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

    Assessment marcus-circular-financing-aint-what-it-used-d8a6cc-a1.

  67. Assessment publishedBy this siteGary Marcus

    Between September 2025 and July 2026 the market grew wary of AI circular financing deals.

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

    Assessment marcus-circular-financing-aint-what-it-used-bd6c3e-a1.

  68. Assessment publishedBy this siteGary Marcus

    The off-balance-sheet financing coming to light and the sudden need for so-called 'creative financing' in the GenAI boom are a warning sign.

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

    Assessment marcus-circular-financing-aint-what-it-used-b7a76f-a1.

  69. Assessment publishedBy this siteGary Marcus

    The market is reading the potential Nvidia-OpenAI $250 billion financing backstop as desperation rather than good news.

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

    Assessment marcus-circular-financing-aint-what-it-used-a1c792-a1.

  70. Assessment publishedBy this siteGary Marcus

    The $300 billion Oracle-OpenAI deal announced on September 10, 2025 never made sense.

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

    Assessment marcus-circular-financing-aint-what-it-used-50483d-a1.

  71. HeldBy this siteGary Marcus

    Held record marcus-circular-financing-aint-what-it-used-2bc93b

    The final-text check failed: rationale_supported_by_evidence, citations_correct; overclaims: Anthropic's October 26, 2026 filing reports net losses through June 2026, alongside positive operating cash flow and adjusted operating income for the first half of 2026; net losses alone therefore do not establish cash dependence [e8]..

  72. Assessment publishedBy this siteGary Marcus

    Congress should investigate how the US lost its AI lead.

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

    Assessment marcus-china-has-all-but-caught-up-the-us-fb2a26-a1.

  75. HeldBy this siteGary Marcus

    Held record marcus-china-has-all-but-caught-up-the-us-eae9fa

    The anthropic screen reads the claim differently.

  76. Assessment publishedBy this siteGary Marcus

    The recent Chinese model releases cast serious doubt on the idea, never very plausible, that America might 'win' the AI race.

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

    Assessment marcus-china-has-all-but-caught-up-the-us-e3f25a-a1.

  77. HeldBy this siteGary Marcus

    Held record marcus-china-has-all-but-caught-up-the-us-df9914

    The anthropic review does not accept e1 as supports.

  78. Assessment publishedBy this siteGary Marcus

    If the US does nothing to support its AI labs, it may concede much of the AI market to China.

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

    Assessment marcus-china-has-all-but-caught-up-the-us-d4e04a-a2.

  79. Assessment publishedBy this siteGary Marcus

    The US government should not bail out the big AI labs.

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

    Assessment marcus-china-has-all-but-caught-up-the-us-c7c126-a1.

  81. Assessment publishedBy this siteGary Marcus

    The lack of a technical moat has kept companies like OpenAI unprofitable, as Marcus predicted.

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

    Assessment marcus-china-has-all-but-caught-up-the-us-a0d22e-a1.

  82. HeldBy this siteGary Marcus

    Held record marcus-china-has-all-but-caught-up-the-us-9f6377

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

  83. Assessment publishedBy this siteGary Marcus

    If the government took over the AI labs, we should worry about it abusing their data, for example for mass surveillance.

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

    Assessment marcus-china-has-all-but-caught-up-the-us-8c33b4-a1.

  84. HeldBy this siteGary Marcus

    Held record marcus-china-has-all-but-caught-up-the-us-834d70

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

  85. Assessment publishedBy this siteGary Marcus

    The repercussions of Moonshot.AI's announcement of Kimi K3, news that partly drove a US stock market drop on Friday, are likely to be immense.

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

    Assessment marcus-china-has-all-but-caught-up-the-us-809c54-a1.

  86. HeldBy this siteGary Marcus

    Held record marcus-china-has-all-but-caught-up-the-us-622584

    The final-text check failed: proposition_faithful.

  88. Assessment publishedBy this siteGary Marcus

    AI should be put in the public domain through an international effort focused on medicine and science.

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

    Assessment marcus-china-has-all-but-caught-up-the-us-474963-a1.

  89. Assessment publishedBy this siteGary Marcus

    The US should never have gone all-in on GenAI in the first place, since GenAI never seemed to offer much in the way of a technical moat.

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

    Assessment marcus-china-has-all-but-caught-up-the-us-401d06-a1.

  90. Assessment publishedBy this siteGary Marcus

    Models have gotten more efficient and cheaper, but hallucinations and reliability problems have persisted.

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

    Assessment marcus-china-has-all-but-caught-up-the-us-284a22-a1.

  91. HeldBy this siteGary Marcus

    Held record marcus-china-has-all-but-caught-up-the-us-19ef3e

    The anthropic screen reads the claim differently.

  92. Assessment publishedBy this siteGary Marcus

    A regulatory moat protecting OpenAI and Anthropic would likely crush US startups.

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

    Assessment marcus-china-has-all-but-caught-up-the-us-05d559-a2.

  94. Assessment publishedBy this siteGary Marcus

    The LLM approach is easily replicated, which causes price wars and small or negative margins.

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

    Assessment marcus-china-catches-up-bd7b2c-a1.

  95. Assessment publishedBy this siteGary Marcus

    The U.S. is unlikely to win a race with China to sell the cheapest LLMs.

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

    Assessment marcus-china-catches-up-848561-a1.

  96. Assessment publishedBy this siteGary Marcus

    The U.S. should perhaps focus on developing new forms of AI suited to science and medicine rather than racing to sell the cheapest LLMs.

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

    Assessment marcus-china-catches-up-848561-2-a1.

  97. Assessment publishedBy this siteGary Marcus

    Massive AI data center investments are unlikely ever to pay off because price wars are driving token prices toward zero, leaving meagre profits.

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

    Assessment marcus-china-catches-up-656823-a1.

  98. HeldBy this siteGary Marcus

    Held record marcus-china-catches-up-55f7b0

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

  99. Assessment publishedBy this siteGary Marcus

    Because the systems are not reliable, charging premium prices was never really viable in the long term.

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

    Assessment marcus-china-catches-up-5505dc-a1.

  100. Assessment publishedBy this siteGary Marcus

    Anthropic and OpenAI are unlikely to achieve trillion-dollar IPOs given price wars and industry-wide token price sensitivity.

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

    Assessment marcus-china-catches-up-3ec75f-a1.

  102. Assessment publishedBy this siteGary Marcus

    Price wars from low-cost competitors may wreck the U.S. AI industry.

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

    Assessment marcus-china-catches-up-1db255-2-a1.

  103. Assessment publishedBy this siteGary Marcus

    It is unclear whether Anthropic's quarters after the projected profitable one will be profitable.

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

    Assessment marcus-checking-the-math-behind-openai-and-f6fe1b-a1.

  104. Assessment publishedBy this siteGary Marcus

    Anthropic is projecting its first, slightly profitable quarter ever, per the WSJ.

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

    Assessment marcus-checking-the-math-behind-openai-and-ef1be9-a1.

  105. HeldBy this siteGary Marcus

    Held record marcus-checking-the-math-behind-openai-and-e57e9a

    The anthropic screen reads the claim differently.

  106. Assessment publishedBy this siteGary Marcus

    OpenAI's claimed result on the Erdős planar unit distance problem, though impressive, should be viewed with skepticism.

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

    Assessment marcus-checking-the-math-behind-openai-and-c2a2cd-a1.

  107. HeldBy this siteGary Marcus

    Held record marcus-checking-the-math-behind-openai-and-c178f9

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

  108. Assessment publishedBy this siteGary Marcus

    IPO buyers should beware.

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

    Assessment marcus-checking-the-math-behind-openai-and-ba3d18-a1.

  109. Assessment publishedBy this siteGary Marcus

    Nvidia is spending so much on circular financing that its cash flow is headed towards zero.

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

    Assessment marcus-checking-the-math-behind-openai-and-b9ad5e-a1.

  111. HeldBy this siteGary Marcus

    Held record marcus-checking-the-math-behind-openai-and-596770

    The final-text check failed: rationale_supported_by_evidence; overclaims: On May 20, WIRED reported unspecified reduced SpaceX compute fees for May and June before a $1.25 billion monthly rate took effect [e2]..

  112. HeldBy this siteGary Marcus

    Held record marcus-checking-the-math-behind-openai-and-0a7f8d

    The openai screen doubts the attribution.

  113. Assessment publishedBy this siteGary Marcus

    ChatGPT's new image engine's tall tandem bike image clearly shows a lack of functional understanding.

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

    Assessment marcus-chatgpts-powerful-new-image-engine-c05c48-a1.

  114. Assessment publishedBy this siteGary Marcus

    ChatGPT's tandem bike image contains numerous problems that bike experts would immediately spot.

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

    Assessment marcus-chatgpts-powerful-new-image-engine-7517a3-a2.

  115. Assessment publishedBy this siteGary Marcus

    ChatGPT's new image engine's labeled bike diagram contains several revealing labeling errors.

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

    Assessment marcus-chatgpts-powerful-new-image-engine-6cc7be-a1.

  116. Assessment publishedBy this siteGary Marcus

    ChatGPT's new image engine does not actually understand how the parts of a bike function.

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

    Assessment marcus-chatgpts-powerful-new-image-engine-2d6278-a1.

  117. Assessment publishedBy this siteGary Marcus

    ChatGPT's new image engine appears to have limited functional understanding of the world.

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

    Assessment marcus-chatgpts-powerful-new-image-engine-1b1d05-a1.

  118. HeldBy this siteGary Marcus

    Held record marcus-chatgpt-in-shambles-a7ccea

    The anthropic screen doubts the attribution.

  120. Assessment publishedBy this siteGary Marcus

    In Marcus's test, ChatGPT reported that the letter h could be a vowel, at least in the word Northwest.

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

    Assessment marcus-chatgpt-in-shambles-5746f4-a2.

  121. Assessment publishedBy this siteGary Marcus

    In Marcus's test, ChatGPT failed several times to count to 50 correctly.

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

    Assessment marcus-chatgpt-in-shambles-549d1f-a2.

  122. HeldBy this siteGary Marcus

    Held record marcus-chatgpt-in-shambles-3b64e2

    The openai screen reads the claim differently.

  123. HeldBy this siteGary Marcus

    Held record marcus-chatgpt-in-shambles-25cbdb

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

  124. Assessment publishedBy this siteGary Marcus

    In Marcus's test, ChatGPT issued many wrong corrections and acknowledged no uncertainty until its errors were pointed out.

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

    Assessment marcus-chatgpt-in-shambles-1bd53d-a1.

  125. Assessment publishedBy this siteGary Marcus

    In Marcus's test, the free version of ChatGPT failed several times to produce a complete list of all 50 US states.

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

    Assessment marcus-chatgpt-in-shambles-0811ec-a2.

  126. Assessment publishedBy this siteGary Marcus

    ChatGPT does not truly lie, but it often produces text that bears little resemblance to reality.

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

    Assessment marcus-chatgpt-in-shambles-06205f-a1.

  127. Assessment publishedBy this siteGary Marcus

    Over the last few hours before the post, people reported a variety of problems with ChatGPT.

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

    Assessment marcus-chatgpt-has-gone-berserk-ea5568-a1.

  129. HeldBy this siteGary Marcus

    Held record marcus-chatgpt-has-gone-berserk-e2f41e

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

  130. Assessment publishedBy this siteGary Marcus

    In the end, generative AI is a kind of alchemy.

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

    Assessment marcus-chatgpt-has-gone-berserk-d9aadf-a1.

  131. Assessment publishedBy this siteGary Marcus

    We are still living in the age of machine learning alchemy that an xkcd cartoon captured several years ago.

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

    Assessment marcus-chatgpt-has-gone-berserk-acb096-a1.

  132. Assessment publishedBy this siteGary Marcus

    Generative AI systems have never been stable.

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

    Assessment marcus-chatgpt-has-gone-berserk-74297d-a1.

  133. Assessment publishedBy this siteGary Marcus

    Developers and military personnel should not let their chatbots grow up to be generals.

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

    Assessment marcus-chatgpt-has-gone-berserk-30e3d8-a1.

  134. Assessment publishedBy this siteGary Marcus

    The ChatGPT malfunction of February 2024 may well be fixed quickly.

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

    Assessment marcus-chatgpt-has-gone-berserk-302116-a1.

  135. Assessment publishedBy this siteGary Marcus

    The ChatGPT malfunction is a genuine wakeup call about the instability of generative AI and should be treated as one.

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

    Assessment marcus-chatgpt-has-gone-berserk-302116-2-a1.

  136. Assessment publishedBy this siteGary Marcus

    ChatGPT's exciting new image system has trouble with functional understanding.

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

    Assessment marcus-chatgpt-doesnt-know-its-whisk-from-251653-a1.

  137. Assessment publishedBy this siteGary Marcus

    Neurosymbolic AI can help address the reliability problems of current models.

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

    Assessment marcus-chatgpt-blows-mapmaking-101-c85142-a2.

  138. HeldBy this siteGary Marcus

    Held record marcus-chatgpt-blows-mapmaking-101-c61715

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

  139. HeldBy this siteGary Marcus

    Held record marcus-chatgpt-blows-mapmaking-101-89891d

    The anthropic screen doubts the attribution.

  140. Assessment publishedBy this siteGary Marcus

    Safe AI cannot be achieved unless the reliability/instruction-following problem is addressed directly.

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

    Assessment marcus-chatgpt-blows-mapmaking-101-3bead2-a2.

  141. Assessment publishedBy this siteGary Marcus

    Hundreds of billions more have been invested on further speculation that scaling would somehow magically cure problems that appear to be inherent to the technology.

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

    Assessment marcus-chatgpt-at-age-two-ffc3f5-a1.

  142. HeldBy this siteGary Marcus

    Held record marcus-chatgpt-at-age-two-eebc1e

    e6 is dated 2024, which is not inside the period described (2024-01-01 to 2024-11-28); it can only be context; e7 is dated 2024, which is not inside the period described (2024-01-01 to 2024-11-28); it can only be context

  143. HeldBy this siteGary Marcus

    Held record marcus-chatgpt-at-age-two-ec8d71

    e2 is dated 2024-05, which is not inside the period described (2024-05-13 to 2024-11-28); it can only be context; e3 is dated 2024-11, which is not inside the period described (2024-05-13 to 2024-11-28); it can only be context

  144. HeldBy this siteGary Marcus

    Held record marcus-chatgpt-at-age-two-e776e7

    e3 is dated 2024-11, which is not inside the period described (2023-01 to 2024-11-28); it can only be context

  145. HeldBy this siteGary Marcus

    Held record marcus-chatgpt-at-age-two-dd5693

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

  146. Assessment publishedBy this siteGary Marcus

    Two years after ChatGPT's launch, Google's search share has not diminished much.

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

    Assessment marcus-chatgpt-at-age-two-ce2014-a1.

  147. HeldBy this siteGary Marcus

    Held record marcus-chatgpt-at-age-two-ada8d0

    e1 is dated 2024, which is not inside the period described (2024-01-01 to 2024-11-28); it can only be context

  148. HeldBy this siteGary Marcus

    Held record marcus-chatgpt-at-age-two-94dad3

    The final-text check failed: proposition_faithful.

  149. HeldBy this siteGary Marcus

    Held record marcus-chatgpt-at-age-two-8142fc

    e2 is dated 2024, which is not inside the period described (2024-01-01 to 2024-11-28); it can only be context

  150. Assessment publishedBy this siteGary Marcus

    Many theories about how ChatGPT would be used in practice have fizzled out.

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

    Assessment marcus-chatgpt-at-age-two-725de6-a1.

  151. Assessment publishedBy this siteGary Marcus

    He still thinks a massive investment in new approaches to AI, as he called for at the end of the Ezra Klein interview, is the way to go.

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

    Assessment marcus-chatgpt-at-age-two-680a49-a1.

  152. Assessment publishedBy this siteGary Marcus

    There has been much experimentation with ChatGPT-style AI but relatively few well-documented success stories.

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

    Assessment marcus-chatgpt-at-age-two-608f86-a1.

  154. HeldBy this siteGary Marcus

    Held record marcus-chatgpt-at-age-two-335660

    e2 is dated 2024-05, which is not inside the period described (2024-05-13 to 2024-11-28); it can only be context

  155. HeldBy this siteGary Marcus

    Held record marcus-chatgpt-at-age-two-1b8521

    The openai screen reads the claim differently.

  156. HeldBy this siteGary Marcus

    Held record marcus-chatgpt-at-age-two-159e00

    e1 is dated 2024-11, which is not inside the period described (2022-11-30 to 2024-11-28); it can only be context

  157. Assessment publishedBy this siteGary Marcus

    OpenAI maximizing user-engagement metrics was a big part of the cause of ChatGPT users having mental health crises.

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

    Assessment marcus-chatgpt-and-delusions-an-important-9b2257-a2.

  158. HeldBy this siteGary Marcus

    Held record marcus-chatgpt-and-delusions-an-important-7b33e1

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

  159. Assessment publishedBy this siteGary Marcus

    Marcus suspects Ilya Sutskever left OpenAI because of the company's attitude toward AI safety, not because of a scary new algorithm.

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

    Assessment marcus-chaos-and-tension-at-openai-f63473-a1.

  160. Assessment publishedBy this siteGary Marcus

    OpenAI's founding mission of benefiting humanity, with AI safety at its center, appears to have been left behind.

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

    Assessment marcus-chaos-and-tension-at-openai-c1c224-a1.

  162. Assessment publishedBy this siteGary Marcus

    The four key resignations from OpenAI so far in 2024 are not a good sign.

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

    Assessment marcus-chaos-and-tension-at-openai-1fc2b7-a1.

  163. Assessment publishedBy this siteGary Marcus

    The massive investment of hundreds of billions of dollars in AI scaling is unlikely to pay off.

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

    Assessment marcus-casey-newton-strikes-out-eb4a12-a1.

  164. HeldBy this siteGary Marcus

    Held record marcus-casey-newton-strikes-out-e58d98

    e2 is dated 2024, which is not inside the period described (2024-01-01 to 2024-12-11); it can only be context

  165. Assessment publishedBy this siteGary Marcus

    Driverless cars are absent from 97% of the world, hallucinations persist, and home humanoid robots are nowhere in sight.

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

    Assessment marcus-casey-newton-strikes-out-dc953a-a1.

  166. HeldBy this siteGary Marcus

    Held record marcus-casey-newton-strikes-out-daebaf

    The anthropic review reads the claim differently.

  167. Assessment publishedBy this siteGary Marcus

    Neurosymbolic AI is finally coming into vogue.

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

    Assessment marcus-casey-newton-strikes-out-da89b5-a1.

  168. HeldBy this siteGary Marcus

    Held record marcus-casey-newton-strikes-out-d74389

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

  169. HeldBy this siteGary Marcus

    Held record marcus-casey-newton-strikes-out-c74d7e

    The final-text check failed: citations_correct.

  170. Assessment publishedBy this siteGary Marcus

    Defending against superintelligence dangers must start with a realistic appraisal of current progress and timelines.

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

    Assessment marcus-casey-newton-strikes-out-c13ae1-a1.

  171. Assessment publishedBy this siteGary Marcus

    Chatbot-related deaths such as suicides occurred in 2023 and 2024.

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

    Assessment marcus-casey-newton-strikes-out-b8a61b-a2.

  172. HeldBy this siteGary Marcus

    Held record marcus-casey-newton-strikes-out-a897e2

    The final-text check failed: rationale_supported_by_evidence, citations_correct; overclaims: A 2024 benchmark of several open and closed video models found that even the best one followed physical laws in under 40% of cases. | It did not test Sora..

  174. HeldBy this siteGary Marcus

    Held record marcus-casey-newton-strikes-out-825e21

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

  175. HeldBy this siteGary Marcus

    Held record marcus-casey-newton-strikes-out-631d72

    e2 is dated 2024, which is not inside the period described (2024-01-01 to 2024-12-11); it can only be context

  176. Assessment publishedBy this siteGary Marcus

    After rapid progress in 2019-2022, LLM progress has by many measures reportedly slowed.

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

    Assessment marcus-casey-newton-strikes-out-500024-a1.

  177. Assessment publishedBy this siteGary Marcus

    Evidence suggests superintelligence will not be reached in the very near term.

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

    Assessment marcus-casey-newton-strikes-out-3b5fc7-a1.

  178. Assessment publishedBy this siteGary Marcus

    OpenAI has not yet made a profit and does not expect to before 2029.

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

    Assessment marcus-casey-newton-strikes-out-27dbdd-a1.

  179. HeldBy this siteGary Marcus

    Held record marcus-casey-newton-strikes-out-265d67

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

  180. Assessment publishedBy this siteGary Marcus

    You can't improve the engineering if you can't identify its problems and weak points so that you can address them.

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

    Assessment marcus-casey-newton-strikes-out-0b84a4-a1.

  182. HeldBy this siteGary Marcus

    Held record marcus-caricaturing-noam-chomsky-d998ed

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

  183. Assessment publishedBy this siteGary Marcus

    LLMs are useful technology but remain a long way from true artificial intelligence and even further from telling us anything useful about the human mind.

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

    Assessment marcus-caricaturing-noam-chomsky-cb0488-a1.

  184. HeldBy this siteGary Marcus

    Held record marcus-caricaturing-noam-chomsky-af3278

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

  185. Assessment publishedBy this siteGary Marcus

    LLMs do not reliably understand the world.

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

    Assessment marcus-caricaturing-noam-chomsky-a8598b-a1.

  186. Assessment publishedBy this siteGary Marcus

    The request in his example is perfectly clear and sensible, not nonsensical in any way, and the system just isn't up to the job.

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

    Assessment marcus-caricaturing-noam-chomsky-83d509-a1.

  187. HeldBy this siteGary Marcus

    Held record marcus-caricaturing-noam-chomsky-6fc570

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

  188. HeldBy this siteGary Marcus

    Held record marcus-caricaturing-noam-chomsky-4fe53c

    The final-text check failed: rationale_supported_by_evidence; overclaims: The claim says that even when LLMs give correct outputs, they 'rarely if ever' derive the same abstractions a human or a symbolic machine would..

  189. HeldBy this siteGary Marcus

    Held record marcus-caricaturing-noam-chomsky-4d2289

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

  190. HeldBy this siteGary Marcus

    Held record marcus-caricaturing-noam-chomsky-044def

    The final-text check failed: citations_correct.

  191. Assessment publishedBy this siteGary Marcus

    Minerva, despite immense training data and apparent two-digit arithmetic ability, fails altogether on 4-digit multiplication.

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

    Assessment marcus-caricaturing-noam-chomsky-016ae2-a1.

  192. HeldBy this siteGary Marcus

    Held record marcus-c39-d89494

    e1 is dated 2024-12, which is not inside the period described (2024-12 to 2024-12-22); it can only be context

  193. Assessment publishedBy this siteGary Marcus

    People in the media probably shouldn't even joke about o3 being AGI.

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

    Assessment marcus-c39-cf55b9-a1.

  194. Assessment publishedBy this siteGary Marcus

    The key test of o3's base model on ARC without pretraining was not done.

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

    Assessment marcus-c39-c129a4-a1.

  195. Assessment publishedBy this siteGary Marcus

    The o3 ARC graphs presented by OpenAI and by Chollet were misleading.

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

    Assessment marcus-c39-aff2ef-a1.

  196. Assessment publishedBy this siteGary Marcus

    There is no solid test yet of o3's ARC performance without pretraining.

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

    Assessment marcus-c39-a18448-a1.

  197. Assessment publishedBy this siteGary Marcus

    What was done for o3 on ARC, pretraining on what Marcus believes was hundreds of public examples, is not comparable to what humans require.

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

    Assessment marcus-c39-a0bac4-a1.

  198. Assessment publishedBy this siteGary Marcus

    The ARC test should never have been named ARC-AGI.

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

    Assessment marcus-c39-9a5c0a-a1.

  199. Assessment publishedBy this siteGary Marcus

    OpenAI's o3 announcement video should have been much clearer about what was tested and what was trained.

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

    Assessment marcus-c39-993e85-a1.

  200. Assessment publishedBy this siteGary Marcus

    Because the wrong experiment was performed and key data were not given, o3 cannot be compared directly with humans on ARC, and the best humans still outperformed the model.

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

    Assessment marcus-c39-838202-a2.