Changes, page 18 of 55

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 (10984)

Page 18 of 55: changes 3401 to 3600 of 10984, newest first. Dates are when the change was recorded.

2026-10-29
HeldBy this siteEd Zitron
Held record wyea-ai-is-already-in-dangerous-hands-f6a29d
e1 is dated 2026, which is not inside the period described (2025-12-01 to 2026-08-31); it can only be context
2026-10-29
Assessment publishedBy this siteEd Zitron
The AI industry's current path is not the right one, and every further step down it makes the bubble's eventual collapse worse and multiplies the dangerous, reckless experiments these companies can run thanks to near-unlimited compute.
First assessment published after both reviews and the final-text check passed.
Assessment wyea-ai-is-already-in-dangerous-hands-f00297-a2.
2026-10-29
HeldBy this siteEd Zitron
The agents in the Hugging Face attack were not trying to understand their world; they executed a task using actions from their training material.
The final-text check failed: citations_correct.
2026-10-29
HeldBy this siteEd Zitron
Held record wyea-ai-is-already-in-dangerous-hands-d5a07e
The final-text check failed: excerpt_matches.
2026-10-29
HeldBy this siteEd Zitron
Held record wyea-ai-is-already-in-dangerous-hands-c7e9c3
The anthropic review reads the claim differently.
2026-10-29
Assessment publishedBy this siteEd Zitron
Coxon's warnings are specious and in many cases amount to little more than him saying he heard colleagues say such things, yet they have made a lot of people really scared of AI, even if AI can't really do the things he describes.
First assessment published after both reviews and the final-text check passed.
Assessment wyea-ai-is-already-in-dangerous-hands-bc9bad-a1.
2026-10-29
Assessment publishedBy this siteEd Zitron
Even though most of the media mostly overreacted, the sheer volume of traffic to Coxon's tweet and his endless media interviews have moved the idea of a needed 'global AI slowdown' into the global zeitgeist.
First assessment published after both reviews and the final-text check passed.
Assessment wyea-ai-is-already-in-dangerous-hands-aacbee-a1.
2026-10-29
HeldBy this siteEd Zitron
Held record wyea-ai-is-already-in-dangerous-hands-9b8ba5
The anthropic review reads the claim differently.
2026-10-29
Assessment publishedBy this siteEd Zitron
Because AI labs and hyperscalers intentionally sold their products on a theoretical future version that would fix all the problems, their 'powerful AI', graded on preferential anecdotes and rigged benchmarks rather than actual outcomes, was always described as 'on the frontier' and 'getting smarter every day'.
First assessment published after both reviews and the final-text check passed.
Assessment wyea-ai-is-already-in-dangerous-hands-93f7e3-a1.
2026-10-29
HeldBy this siteEd Zitron
Held record wyea-ai-is-already-in-dangerous-hands-84d36e
The final-text check failed: proposition_faithful.
2026-10-29
Assessment publishedBy this siteEd Zitron
For years the AI industry has engaged in vague safety theater while selling harmful, unreliable software.
First assessment published after both reviews and the final-text check passed.
Assessment wyea-ai-is-already-in-dangerous-hands-7aa4d6-a1.
2026-10-29
Assessment publishedBy this siteEd Zitron
The Hugging Face attack was software doing what it was asked, not a rogue agent, and OpenAI appeared to lack any monitoring of thousands of LLMs running.
First assessment published after both reviews and the final-text check passed.
Assessment wyea-ai-is-already-in-dangerous-hands-74595c-a1.
2026-10-29
HeldBy this siteEd Zitron
Held record wyea-ai-is-already-in-dangerous-hands-62f621
The anthropic review found the criteria unfair to the speaker's words.
2026-10-29
Assessment publishedBy this siteEd Zitron
Altman's early-2023 warning that AI could be used for offensive cyber-attacks proved right, but only because Altman made sure it was the case.
First assessment published after both reviews and the final-text check passed.
Assessment wyea-ai-is-already-in-dangerous-hands-57d845-a1.
2026-10-29
HeldBy this siteEd Zitron
Held record wyea-ai-is-already-in-dangerous-hands-572ba3
hedged language is not scored, even on a plain reading
2026-10-29
HeldBy this siteEd Zitron
Held record wyea-ai-is-already-in-dangerous-hands-508f9e
The final-text check failed: proposition_faithful.
2026-10-29
Assessment publishedBy this siteEd Zitron
The Hugging Face attack was made possible not so much by 'powerful AI' as by the weaponization of hundreds of billions of dollars' worth of GPU infrastructure owned by Microsoft, Google, Amazon, Oracle, and CoreWeave.
First assessment published after both reviews and the final-text check passed.
Assessment wyea-ai-is-already-in-dangerous-hands-4ae1a6-a2.
2026-10-29
Assessment publishedBy this siteEd Zitron
Coxon never mentioned that ChatGPT was used as a 'suicide coach', directly caused a murder-suicide, or aided and abetted mass shootings in Florida and Canada, or the gas turbines poisoning black communities.
First assessment published after both reviews and the final-text check passed.
Assessment wyea-ai-is-already-in-dangerous-hands-434030-a2.
2026-10-29
Assessment publishedBy this siteEd Zitron
The AI involved in the Hugging Face attack did not go rogue; it solved the task it was given.
First assessment published after both reviews and the final-text check passed.
Assessment wyea-ai-is-already-in-dangerous-hands-312aaf-a2.
2026-10-29
HeldBy this siteEd Zitron
Held record wyea-ai-is-already-in-dangerous-hands-28c75d
The final-text check failed: citations_correct.
2026-10-29
HeldBy this siteEd Zitron
Held record wyea-ai-is-already-in-dangerous-hands-1bb6b0
The final-text check failed: conditions_shown, citations_correct.
2026-10-29
Assessment publishedBy this siteEd Zitron
AI labs' 'AI safety' talk, including Amodei's, never addresses actual harms of the products they build today.
First assessment published after both reviews and the final-text check passed.
Assessment wyea-ai-is-already-in-dangerous-hands-1bb409-a1.
2026-10-29
Assessment publishedBy this siteEd Zitron
Hype around OpenAI and Anthropic has created massive concentration risk in world debt markets, and an economic downturn is coming.
First assessment published after both reviews and the final-text check passed.
Assessment wyea-ai-is-already-in-dangerous-hands-15a3b8-a1.
2026-10-29
Assessment publishedBy this siteEd Zitron
Systemic risk from private credit financing of AI infrastructure is building.
First assessment published after both reviews and the final-text check passed.
Assessment wyea-ai-is-a-money-trap-f33804-a2.
2026-10-29
HeldBy this siteEd Zitron
Held record wyea-ai-is-a-money-trap-ef6e16
hedged language is not scored, even on a plain reading
2026-10-29
HeldBy this siteEd Zitron
Held record wyea-ai-is-a-money-trap-e48b5b
The anthropic review found the criteria unfair to the speaker's words.
2026-10-29
Assessment publishedBy this siteEd Zitron
Generative AI businesses do not have the same economics or the same total addressable markets as earlier cash-burning companies such as Uber.
First assessment published after both reviews and the final-text check passed.
Assessment wyea-ai-is-a-money-trap-d44a7c-a1.
2026-10-29
Assessment publishedBy this siteEd Zitron
Cursor's product is now worse, so people will cancel their subscriptions, its annualized revenue will drop, and as a direct result its ability to raise capital will suffer.
First assessment published after both reviews and the final-text check passed.
Assessment wyea-ai-is-a-money-trap-cef5f9-a1.
2026-10-29
HeldBy this siteEd Zitron
Held record wyea-ai-is-a-money-trap-c701ba
The openai review reads the claim differently.
2026-10-29
Assessment publishedBy this siteEd Zitron
When the AI bubble pops, VC funding will dry up and pension funds will take a hit that affects how much people have to spend in retirement, and it will be grim.
First assessment published after both reviews and the final-text check passed.
Assessment wyea-ai-is-a-money-trap-c698eb-a2.
2026-10-29
Assessment publishedBy this siteEd Zitron
Three years in, there is no profitable generative AI company other than training-data firms and Nvidia.
First assessment published after both reviews and the final-text check passed.
Assessment wyea-ai-is-a-money-trap-b85ba7-a1.
2026-10-29
Assessment publishedBy this siteEd Zitron
The US government will not give generative AI companies money forever.
First assessment published after both reviews and the final-text check passed.
Assessment wyea-ai-is-a-money-trap-abcafb-a1.
2026-10-29
HeldBy this siteEd Zitron
Held record wyea-ai-is-a-money-trap-896d82
The openai review found the criteria unfair to the speaker's words.
2026-10-29
HeldBy this siteEd Zitron
Held record wyea-ai-is-a-money-trap-884dae
The anthropic review reads the claim differently.
2026-10-29
HeldBy this siteEd Zitron
Held record wyea-ai-is-a-money-trap-82a167
the conclusive event e1 is not outcome evidence in the packet
2026-10-29
HeldBy this siteEd Zitron
Held record wyea-ai-is-a-money-trap-81a0dd
The openai review found the criteria unfair to the speaker's words.
2026-10-29
HeldBy this siteEd Zitron
Held record wyea-ai-is-a-money-trap-6bc0cc
The anthropic review found the criteria unfair to the speaker's words.
2026-10-29
HeldBy this siteEd Zitron
Held record wyea-ai-is-a-money-trap-6b88f9
a observation claim cannot be assessed on the forecast_outcome axis; the horizon on 2026-10-28 is open_horizon, not elapsed
2026-10-29
HeldBy this siteEd Zitron
Held record wyea-ai-is-a-money-trap-5d0403
hedged language is not scored, even on a plain reading
2026-10-29
Assessment publishedBy this siteEd Zitron
If Microsoft's capex dips, it will directly hurt the US economy and stock market.
First assessment published after both reviews and the final-text check passed.
Assessment wyea-ai-is-a-money-trap-5a530f-a1.
2026-10-29
Assessment publishedBy this siteEd Zitron
He believes Anthropic's new weekly limits on Claude Code accounts will create some of the most gnarly churn in SaaS history.
First assessment published after both reviews and the final-text check passed.
Assessment wyea-ai-is-a-money-trap-5400fd-a1.
2026-10-29
HeldBy this siteEd Zitron
Held record wyea-ai-is-a-money-trap-53dba6
The anthropic review found the criteria unfair to the speaker's words.
2026-10-29
HeldBy this siteEd Zitron
Held record wyea-ai-is-a-money-trap-4e80eb
"unlikely" is hedged or probabilistic; one outcome does not score it; hedged language is not scored, even on a plain reading
2026-10-29
Assessment publishedBy this siteEd Zitron
Generative AI startups have had no exits and no IPOs.
First assessment published after both reviews and the final-text check passed.
Assessment wyea-ai-is-a-money-trap-4dd359-a2.
2026-10-29
HeldBy this siteEd Zitron
Held record wyea-ai-is-a-money-trap-4c9fb8
The openai review found the criteria unfair to the speaker's words.
2026-10-29
HeldBy this siteEd Zitron
Held record wyea-ai-is-a-money-trap-4464db
The openai review found the criteria unfair to the speaker's words.
2026-10-29
Assessment publishedBy this siteEd Zitron
Most AI capex comes from big tech, which is a major systemic weakness in the US economy.
First assessment published after both reviews and the final-text check passed.
Assessment wyea-ai-is-a-money-trap-38273f-a1.
2026-10-29
Assessment publishedBy this siteEd Zitron
He thinks the future of LLMs will be client-side on egregiously expensive personal setups for enthusiasts and in a handful of niche enterprise roles.
First assessment published after both reviews and the final-text check passed.
Assessment wyea-ai-is-a-money-trap-365cd7-a2.
2026-10-29
Assessment publishedBy this siteEd Zitron
When the AI bubble pops and capex falls, layoffs and offshoring will accelerate, driven partly by tech companies needing to show fiscal restraint for the first time in living memory.
First assessment published after both reviews and the final-text check passed.
Assessment wyea-ai-is-a-money-trap-2a7e5c-a2.
2026-10-29
Assessment publishedBy this siteEd Zitron
LLMs don't scale profitably and their functionality doesn't justify their running costs.
First assessment published after both reviews and the final-text check passed.
Assessment wyea-ai-is-a-money-trap-1d3658-a1.
2026-10-29
HeldBy this siteEd Zitron
Held record wyea-ai-is-a-money-trap-1908a9
The two reviews differ: anthropic causal_assessment/open_horizon/unresolved, openai causal_assessment/not_applicable/unresolved.
2026-10-29
Assessment publishedBy this siteEd Zitron
Generative AI companies are unprofitable with no end in sight, in most cases do not make much money, are valued at more than anyone would ever buy them for, do not have much valuable intellectual property, and the two biggest players burn billions of dollars more than they make.
First assessment published after both reviews and the final-text check passed.
Assessment wyea-ai-is-a-money-trap-18cb1b-a3.
2026-10-29
Assessment publishedBy this siteEd Zitron
Reduced AI capex will cut NVIDIA GPU spending, hit the US stock market, and the AI bubble's popping will damage the wider economy like the dot-com bust.
First assessment published after both reviews and the final-text check passed.
Assessment wyea-ai-is-a-money-trap-177d7a-a3.
2026-10-29
HeldBy this siteEd Zitron
Held record wyea-ai-is-a-money-trap-0962f1
The anthropic review abstained.
2026-10-29
Assessment publishedBy this siteEd Zitron
OpenAI is not following standard methods for calculating annualized revenue.
First assessment published after both reviews and the final-text check passed.
Assessment wyea-ai-is-a-money-trap-06cd69-a1.
2026-10-29
Assessment publishedBy this siteEd Zitron
AI capex will eventually slow, reducing NVIDIA GPU purchases and slowing US economic growth.
First assessment published after both reviews and the final-text check passed.
Assessment wyea-ai-is-a-money-trap-064b60-a1.
2026-10-29
HeldBy this siteEd Zitron
Held record wyea-ai-is-a-money-trap-027a73
e2 measures something else (period 2026-04..2026-04 is not ?..?); it can only be context unless it is derived with stated inputs
2026-10-29
Assessment publishedBy this siteEd Zitron
Generative AI startups do not make enough money to survive without a continuous flow of venture capital, and they do not seem to make impressive sums even when allowed to burn as much as they like.
First assessment published after both reviews and the final-text check passed.
Assessment wyea-ai-is-a-money-trap-02700e-a2.
2026-10-29
HeldBy this siteEd Zitron
Held record wyea-ai-had-better-work-out-fe4384
The final-text check failed: citations_correct.
2026-10-29
Assessment publishedBy this siteEd Zitron
Borrowing costs for AI data centers are skyrocketing, making their economics increasingly problematic.
First assessment published after both reviews and the final-text check passed.
Assessment wyea-ai-had-better-work-out-f3e4c9-a2.
2026-10-29
HeldBy this siteEd Zitron
Held record wyea-ai-had-better-work-out-e23af3
The final-text check failed: citations_correct.
2026-10-29
HeldBy this siteEd Zitron
Held record wyea-ai-had-better-work-out-dbbd2a
a observation claim cannot be assessed on the causal_assessment axis
2026-10-29
HeldBy this siteEd Zitron
Held record wyea-ai-had-better-work-out-c8bc6c
The final-text check failed: excerpt_matches, citations_correct.
2026-10-29
Assessment publishedBy this siteEd Zitron
Blackstone has staked its entire future on AI working out.
First assessment published after both reviews and the final-text check passed.
Assessment wyea-ai-had-better-work-out-c854ec-a2.
2026-10-29
Assessment publishedBy this siteEd Zitron
The higher prices of RAM and consumer electronics were caused by an artificial 'supply chain crisis' that Jensen Huang created by pushing NVIDIA's customers to take early delivery of GPUs.
First assessment published after both reviews and the final-text check passed.
Assessment wyea-ai-had-better-work-out-c54db0-a2.
2026-10-29
Assessment publishedBy this siteEd Zitron
With endless delays and hundreds of billions in chips sitting in storage, AI data centers have become a nightmare for fixed-income investors.
First assessment published after both reviews and the final-text check passed.
Assessment wyea-ai-had-better-work-out-c4e60e-a2.
2026-10-29
Assessment publishedBy this siteEd Zitron
To buy those GPUs (or TPUs), every buyer under discussion except Microsoft will have to raise tens of billions of dollars of debt while underlying interest rates are rising and debt markets have chilled on AI data center debt.
First assessment published after both reviews and the final-text check passed.
Assessment wyea-ai-had-better-work-out-bcb3a4-a3.
2026-10-29
Assessment publishedBy this siteEd Zitron
OpenAI and Anthropic are very unprofitable and unstable, with 80% of revenue from companies that can't afford to pay them.
First assessment published after both reviews and the final-text check passed.
Assessment wyea-ai-had-better-work-out-bc80cc-a1.
2026-10-29
Assessment publishedBy this siteEd Zitron
The S&P 500 and NASDAQ are held up by childish excitement about AI growth that will eventually turn to existential fear.
First assessment published after both reviews and the final-text check passed.
Assessment wyea-ai-had-better-work-out-ae7241-a2.
2026-10-29
Assessment publishedBy this siteEd Zitron
Any further souring of debt markets will leave neoclouds and would-be data center operators paying 12% to 15% interest to build data centers sometime three years in the future, for a customer base made up almost entirely of Anthropic and OpenAI.
First assessment published after both reviews and the final-text check passed.
Assessment wyea-ai-had-better-work-out-aa12f8-a2.
2026-10-29
HeldBy this siteEd Zitron
Held record wyea-ai-had-better-work-out-92e995
The final-text check failed: proposition_faithful.
2026-10-29
HeldBy this siteEd Zitron
Held record wyea-ai-had-better-work-out-90fed5
The final-text check failed: excerpt_matches.
2026-10-29
Assessment publishedBy this siteEd Zitron
80% of OpenAI and Anthropic enterprise revenue comes from loss-making AI startups, so the labs' revenue growth depends on venture capital continuing to fund those startups.
First assessment published after both reviews and the final-text check passed.
Assessment wyea-ai-had-better-work-out-7cb19e-a3.
2026-10-29
HeldBy this siteEd Zitron
Held record wyea-ai-had-better-work-out-229895
The openai review reads the claim differently.
2026-10-29
Assessment publishedBy this siteEd Zitron
If Anthropic IPOs at $1.5 trillion, it would be disastrous for OpenAI, which is trying to raise $30 billion at a $1.4 trillion valuation.
First assessment published after both reviews and the final-text check passed.
Assessment wyea-ai-had-better-work-out-10f0fa-a2.
2026-10-29
Assessment publishedBy this siteEd Zitron
The hypotheticals of the AI industry have created one of the largest equities bubbles in modern history, concentrated in a handful of AI-trade names.
First assessment published after both reviews and the final-text check passed.
Assessment wyea-ai-had-better-work-out-01b9f3-a2.
2026-10-29
HeldBy this siteEd Zitron
Held record wyea-ai-doesnt-have-roi-fde698
The openai review reads the claim differently.
2026-10-29
HeldBy this siteEd Zitron
Held record wyea-ai-doesnt-have-roi-f9b0cc
The openai review reads the claim differently.
2026-10-29
Assessment publishedBy this siteEd Zitron
LLMs cannot be trusted to run anything on their own.
First assessment published after both reviews and the final-text check passed.
Assessment wyea-ai-doesnt-have-roi-f1ac56-a2.
2026-10-29
Assessment publishedBy this siteEd Zitron
Once the AI bubble bursts, AI startups will not get funded, AI token budgets will not be approved, and AI data centers will not be able to raise debt.
First assessment published after both reviews and the final-text check passed.
Assessment wyea-ai-doesnt-have-roi-e922d5-a2.
2026-10-29
HeldBy this siteEd Zitron
Held record wyea-ai-doesnt-have-roi-e03c64
The anthropic review found the criteria unfair to the speaker's words.
2026-10-29
HeldBy this siteEd Zitron
Held record wyea-ai-doesnt-have-roi-ca4bb0
The openai review found the criteria unfair to the speaker's words.
2026-10-29
HeldBy this siteEd Zitron
Held record wyea-ai-doesnt-have-roi-bd1ab7
The openai review found the criteria unfair to the speaker's words.
2026-10-29
HeldBy this siteEd Zitron
Held record wyea-ai-doesnt-have-roi-a6f44b
a observation claim cannot be assessed on the causal_assessment axis
2026-10-29
Assessment publishedBy this siteEd Zitron
Dell and CoreWeave have just turned on the first Vera Rubin GPUs, and nobody is calling them profitable or sustainable because NVIDIA is not interested in making things more efficient rather than more expensive.
First assessment published after both reviews and the final-text check passed.
Assessment wyea-ai-doesnt-have-roi-985797-a2.
2026-10-29
HeldBy this siteEd Zitron
Held record wyea-ai-doesnt-have-roi-93d7c4
The anthropic review found the criteria unfair to the speaker's words.
2026-10-29
HeldBy this siteEd Zitron
Held record wyea-ai-doesnt-have-roi-90a8cc
The final-text check failed: citations_correct.
2026-10-29
Assessment publishedBy this siteEd Zitron
Nobody can measure AI's return on investment because it does not have one.
First assessment published after both reviews and the final-text check passed.
Assessment wyea-ai-doesnt-have-roi-682c54-a1.
2026-10-29
HeldBy this siteEd Zitron
Held record wyea-ai-doesnt-have-roi-6109ad
a observation claim cannot be assessed on the causal_assessment axis
2026-10-29
HeldBy this siteEd Zitron
Held record wyea-ai-doesnt-have-roi-4d643b
The anthropic review does not accept e1 as contradicts.
2026-10-29
HeldBy this siteEd Zitron
Held record wyea-ai-doesnt-have-roi-486201
hedged language is not scored, even on a plain reading
2026-10-29
HeldBy this siteEd Zitron
Held record wyea-ai-doesnt-have-roi-3cdf3f
The final-text check failed: citations_correct.
2026-10-29
Assessment publishedBy this siteEd Zitron
AI job loss is a myth, and every company laying people off because of 'the power of AI' is doing so because its shareholders are angry and because it knows it will get headlines.
First assessment published after both reviews and the final-text check passed.
Assessment wyea-ai-doesnt-have-roi-257447-a1.
2026-10-29
HeldBy this siteEd Zitron
Held record wyea-ai-doesnt-have-roi-218910
The openai review found the criteria unfair to the speaker's words.
2026-10-29
Assessment publishedBy this siteEd Zitron
In five years, a rack of Vera Rubin or Blackwell GPUs will cost as much to run as it does today, and unfinished data centers will cost as much to complete and power.
First assessment published after both reviews and the final-text check passed.
Assessment wyea-ai-doesnt-have-roi-1359f4-a1.
2026-10-29
HeldBy this siteEd Zitron
Held record wyea-ai-doesnt-have-roi-0bacd7
a observation claim cannot be assessed on the causal_assessment axis
2026-10-29
HeldBy this siteGary Marcus
Held record marcus-why-openai-may-well-be-completely-3faa3a
e1 is decisive for a financial claim but states no metric scope; e2 is decisive for a financial claim but states no metric scope
2026-10-29
Assessment publishedBy this siteGary Marcus
Six months after his January 2024 warnings, long-term adoption (as opposed to mere experimentation) has been modest.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-why-openai-may-well-be-completely-234e80-a1.
2026-10-29
Assessment publishedBy this siteGary Marcus
GPT-5 had not been released as of July 2024.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-why-openai-may-well-be-completely-234e80-3-a1.
2026-10-29
Assessment publishedBy this siteGary Marcus
There is currently almost no regulation around AI, or around the billions of robots Musk aspires to build.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-why-my-pdoom-has-risen-dramatically-b9fc72-a2.
2026-10-29
HeldBy this siteGary Marcus
Held record marcus-why-my-pdoom-has-risen-dramatically-9b6c5b
The anthropic review reads the claim differently.
2026-10-29
Assessment publishedBy this siteGary Marcus
Grok is a long way from AGI and hardly smart enough to map out an effective world domination.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-why-my-pdoom-has-risen-dramatically-7db7c3-a2.
2026-10-29
HeldBy this siteGary Marcus
Held record marcus-why-my-pdoom-has-risen-dramatically-19e05a
The final-text check failed: proposition_faithful.
2026-10-29
Assessment publishedBy this siteGary Marcus
Grok's antisemitism, fondness for Hitler and sexual-violence outputs are not inevitable in LLMs, and most other major LLMs don't share these specific antisocial tendencies, at least not without much more focused efforts at jailbreaking.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-why-my-pdoom-has-risen-dramatically-13bcb4-a2.
2026-10-29
Assessment publishedBy this siteGary Marcus
The AI's hallucination that Harry Shearer is a 'British actor' came about because, to a next-token predictor, a phrase like 'Harry Shearer' sits in a part of a multidimensional space where words are often followed by words like 'British actor'.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-why-do-large-language-models-hallucinate-e6e804-a1.
2026-10-29
Assessment publishedBy this siteGary Marcus
Fake citations and faked legal cases from LLMs stem from the same problem of mimicking structure without sanity-checking against available information.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-why-do-large-language-models-hallucinate-df0a1b-a1.
2026-10-29
HeldBy this siteGary Marcus
Held record marcus-why-do-large-language-models-hallucinate-d33a88
The final-text check failed: citations_correct.
2026-10-29
Assessment publishedBy this siteGary Marcus
LLMs do not operate like people and never fact check; they essentially only mimic what people say in contexts.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-why-do-large-language-models-hallucinate-65b7e1-a1.
2026-10-29
Assessment publishedBy this siteGary Marcus
Because LLMs statistically mimic human language, they often fool people into thinking they operate like people.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-why-do-large-language-models-hallucinate-30dec2-a1.
2026-10-29
Assessment publishedBy this siteGary Marcus
Almost all of the information about Harry Shearer that the AI biography got wrong or left out could have been found quickly and easily with a two-second search for his Wikipedia page, and most of it could be found in the first screenful.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-why-do-large-language-models-hallucinate-17c951-a1.
2026-10-29
Assessment publishedBy this siteGary Marcus
The hallucination problem is not going away; OpenAI's o3 hallucinates more than some of its predecessors.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-why-do-large-language-models-hallucinate-17bf2e-a1.
2026-10-29
Assessment publishedBy this siteGary Marcus
LLM technology, having cost about half a trillion dollars, cannot reliably check its output against Wikipedia or a supplied article.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-why-do-large-language-models-hallucinate-14b4ca-a1.
2026-10-29
Assessment publishedBy this siteGary Marcus
Pure LLMs are inevitably stuck in the past of their training and deeply limited in inherent abilities to reason, search the web, and think critically.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-why-chatgpt-cant-be-trusted-with-e72c97-a2.
2026-10-29
HeldBy this siteGary Marcus
Held record marcus-why-chatgpt-cant-be-trusted-with-de74ce
The two reviews differ: anthropic historical_observation/not_applicable/supported, openai historical_observation/not_applicable/unresolved.
2026-10-29
HeldBy this siteGary Marcus
Held record marcus-why-chatgpt-cant-be-trusted-with-d36a06
The anthropic review found the criteria unfair to the speaker's words.
2026-10-29
HeldBy this siteGary Marcus
Held record marcus-why-chatgpt-cant-be-trusted-with-4897d2
hedged language is not scored, even on a plain reading
2026-10-29
HeldBy this siteGary Marcus
Held record marcus-why-chatgpt-cant-be-trusted-with-2cec0d
The final-text check failed: citations_correct; overclaims: CNN was reporting Trump's capture announcement hours before the morning testing..
2026-10-29
HeldBy this siteGary Marcus
Held record marcus-why-chatgpt-cant-be-trusted-with-2a998a
The anthropic review found the criteria unfair to the speaker's words.
2026-10-29
Assessment publishedBy this siteGary Marcus
LLM companies have armies of humans, and when Marcus ran his own brief experiments a couple of hours later, OpenAI had apparently already patched this particular problem.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-why-chatgpt-cant-be-trusted-with-17ff45-a1.
2026-10-29
Assessment publishedBy this siteGary Marcus
AI poses not one risk but many, including bias, disinformation, defamation, toxicity, invasion of privacy and so much more, plus risks to employment, intellectual property and the environment, precisely because recent techniques can be applied in so many ways and are still poorly understood.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-why-californias-ai-safety-bill-should-cc8872-a1.
2026-10-29
Assessment publishedBy this siteGary Marcus
If SB-1047 does not pass, the view will be that Big Tech always wins, nobody will bother, and future state and federal efforts will suffer.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-why-californias-ai-safety-bill-should-5cdc3d-a1.
2026-10-29
HeldBy this siteGary Marcus
Held record marcus-why-californias-ai-safety-bill-should-4b7a21
The two reviews differ: anthropic forecast_outcome/open_horizon/not_adjudicable, openai forecast_outcome/open_horizon/unresolved.
2026-10-29
HeldBy this siteGary Marcus
Held record marcus-why-californias-ai-safety-bill-should-46b9aa
The two reviews differ: anthropic forecast_outcome/open_horizon/not_adjudicable, openai forecast_outcome/open_horizon/unresolved.
2026-10-29
Assessment publishedBy this siteGary Marcus
GPT-4 will be a pastiche generator useful for brainstorming and drafts but not trustworthy general intelligence.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-what-to-expect-when-youre-expecting-eff808-a2.
2026-10-29
Assessment publishedBy this siteGary Marcus
Alignment will remain a critical unsolved problem with GPT-4.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-what-to-expect-when-youre-expecting-a9525d-a1.
2026-10-29
HeldBy this siteGary Marcus
Held record marcus-what-to-expect-when-youre-expecting-9e60b0
The final-text check failed: citations_correct.
2026-10-29
Assessment publishedBy this siteGary Marcus
GPT-4 will not have reliable models of the things it talks about that are accessible to external programmers in a way that reliably feeds downstream processes.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-what-to-expect-when-youre-expecting-37868f-a1.
2026-10-29
Assessment publishedBy this siteGary Marcus
GPT-4's natural language output still won't be something one can reliably hook up to downstream programs, for example simply and directly to a database or virtual assistant with predictable results.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-what-to-expect-when-youre-expecting-238701-a1.
2026-10-29
HeldBy this siteGary Marcus
Held record marcus-what-to-expect-when-youre-expecting-20fe8b
The anthropic review found the criteria unfair to the speaker's words.
2026-10-29
Assessment publishedBy this siteGary Marcus
Without external aids, GPT-4 won't be able to beat Meta's Cicero in Diplomacy, won't be able to drive a car reliably, and won't be able to reliably guide a robot like Optimus to be anything like as versatile as Rosie the Robot.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-what-to-expect-when-youre-expecting-177f3a-a1.
2026-10-29
HeldBy this siteGary Marcus
Held record marcus-what-to-expect-when-youre-expecting-075246
The openai review reads the claim differently.
2026-10-29
HeldBy this siteGary Marcus
Held record marcus-what-to-expect-when-youre-expecting-05bab7
The anthropic review does not accept e1 as supports.
2026-10-29
Assessment publishedBy this siteGary Marcus
OpenAI made a Scarlett Johansson-like voice for GPT-4o after she told them not to, showing dismissiveness toward artist consent.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-what-should-we-learn-from-openais-e20e39-a1.
2026-10-29
Assessment publishedBy this siteGary Marcus
OpenAI has trained on a massive amount of copyrighted material without consent and often without compensation.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-what-should-we-learn-from-openais-d7e7c0-a1.
2026-10-29
Assessment publishedBy this siteGary Marcus
Lately Sam Altman and OpenAI have made many mistakes and broken many promises.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-what-should-we-learn-from-openais-cdf556-a1.
2026-10-29
Assessment publishedBy this siteGary Marcus
Altman's 2016 promise that outsiders would play an important role in OpenAI governance has not been kept.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-what-should-we-learn-from-openais-c7adde-a1.
2026-10-29
HeldBy this siteGary Marcus
Held record marcus-what-should-we-learn-from-openais-9e54b5
e4 is dated 2024-05, which is not inside the period described (2023-11-01 to 2024-05-26); it can only be context; e1 is decisive for a financial claim but states no metric scope; e2 is decisive for a financial claim but states no metric scope; e4 is decisive for a financial claim but states no metric scope
2026-10-29
Assessment publishedBy this siteGary Marcus
OpenAI traded on being 'open' while knowing since May 2016 that the name was misleading.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-what-should-we-learn-from-openais-62c2b4-a1.
2026-10-29
HeldBy this siteGary Marcus
Held record marcus-what-should-we-learn-from-openais-1e3db1
The openai review found the criteria unfair to the speaker's words.
2026-10-29
Assessment publishedBy this siteGary Marcus
Regulatory capture is the default outcome in AI regulation.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-what-should-we-learn-from-openais-0e0898-a1.
2026-10-29
Assessment publishedBy this siteGary Marcus
ChatGPT still struggles with basic facts of life in image generation.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-what-scaling-does-and-doesnt-buy-c1e720-a2.
2026-10-29
Assessment publishedBy this siteGary Marcus
Despite the hype, Nano Banana is not much better than ChatGPT-5 at drawing part-whole structures such as bicycles.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-what-scaling-does-and-doesnt-buy-6f1bf0-a1.
2026-10-29
Assessment publishedBy this siteGary Marcus
Even the latest image generation models, Nano Banana among them, still struggle with compositionality.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-what-scaling-does-and-doesnt-buy-6b861f-a1.
2026-10-29
Assessment publishedBy this siteGary Marcus
Marcus got Nano Banana to fail on his very first attempt.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-what-scaling-does-and-doesnt-buy-0d6610-a1.
2026-10-29
Assessment publishedBy this siteGary Marcus
Over the last few years, graphics in image generation systems have gotten better and better, while world models and comprehension have improved less.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-what-scaling-does-and-doesnt-buy-0c6ad1-a1.
2026-10-29
Assessment publishedBy this siteGary Marcus
Almost nothing has been done through laws or technology to blunt the impact of AI-generated misinformation.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-we-cant-trust-the-fox-to-guard-the-d087bb-a1.
2026-10-29
HeldBy this siteGary Marcus
Held record marcus-we-cant-trust-the-fox-to-guard-the-afa332
The openai review found the criteria unfair to the speaker's words.
2026-10-29
HeldBy this siteGary Marcus
Held record marcus-we-cant-trust-the-fox-to-guard-the-a8a49d
a observation claim cannot be assessed on the causal_assessment axis
2026-10-29
Assessment publishedBy this siteGary Marcus
The US was not going to get the AI legislation it needed.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-we-cant-trust-the-fox-to-guard-the-8a395b-a2.
2026-10-29
Assessment publishedBy this siteGary Marcus
Microsoft's documentation does not clearly explain whether Connected Services data is used to train its large language models, despite saying connected experiences may use machine learning to analyze data.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-update-re-microsoft-and-training-0d292f-a2.
2026-10-29
Assessment publishedBy this siteGary Marcus
Despite hype about exponential improvement, AI still lacks the reliable deep-language-understanding image generation Google promised in 2022.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-image-generation-still-crazy-after-d11e1d-a1.
2026-10-29
Assessment publishedBy this siteGary Marcus
Embedding LLM technology in robots would be even more risky than using it in home assistants.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-how-come-smart-assistants-have-virtually-34cc58-a1.
2026-10-29
Assessment publishedBy this siteGary Marcus
In a lot of ways, large language models are an incredible advance over the AI of the 1970s: they can engage in a much broader range of tasks and generally need less hand-tinkering, though they still need some, which is what reinforcement learning with human feedback is really about.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-hello-multimodal-hallucinations-85ff6e-a1.
2026-10-29
HeldBy this siteGary Marcus
Held record marcus-has-the-bailout-of-generative-ai-b801dd
The two reviews differ: anthropic historical_observation/not_applicable/unresolved, openai historical_observation/not_applicable/contradicted.
2026-10-29
HeldBy this siteGary Marcus
Held record marcus-has-the-bailout-of-generative-ai-5a41e1
The anthropic review found the criteria unfair to the speaker's words.
2026-10-29
Assessment publishedBy this siteGary Marcus
Sooner or later, powerful people are going to use LLMs to shape people's ideas.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-has-grok-lost-its-mind-and-mind-melded-823a8e-a1.
2026-10-29
Assessment publishedBy this siteGary Marcus
In May 2025 Grok repeatedly brought up 'white genocide' in South Africa even in replies where it was completely irrelevant.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-has-grok-lost-its-mind-and-mind-melded-1b29b0-a1.
2026-10-29
HeldBy this siteGary Marcus
Held record marcus-hard-forked-casey-newtons-distorted-e4e7f3
The final-text check failed: rationale_supported_by_evidence, citations_correct; overclaims: Earlier research, published July 8, 2024, explains that extensive testing can miss weaknesses and that testing and red-teaming do not provide rigorous, quantifiable safety guarantees [e6]..
2026-10-29
Assessment publishedBy this siteGary Marcus
The AI acceleration seen from 2019-2023 will not slow down forever, even if it slows down for five years as Marcus has often speculated.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-hard-forked-casey-newtons-distorted-637742-a2.
2026-10-29
HeldBy this siteGary Marcus
Held record marcus-hard-forked-casey-newtons-distorted-342d98
e1 is dated 2024, which is not inside the period described (2024-01-01 to 2024-12-06); it can only be context; e2 is dated 2024, which is not inside the period described (2024-01-01 to 2024-12-06); it can only be context; e3 is dated 2024, which is not inside the period described (2024-01-01 to 2024-12-06); it can only be context; e4 is dated 2024, which is not inside the period described (2024-01-01 to 2024-12-06); it can only be context
2026-10-29
HeldBy this siteGary Marcus
Held record marcus-hard-forked-casey-newtons-distorted-2dfa27
hedged language is not scored, even on a plain reading
2026-10-29
Assessment publishedBy this siteGary Marcus
Large language models still make silly errors, and problems of abstraction are widespread in them.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-happy-groundhog-day-the-ai-edition-d7f85a-a2.
2026-10-29
Assessment publishedBy this siteGary Marcus
The current revival of neural networks has more staying power than the earlier one, which collapsed.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-happy-groundhog-day-the-ai-edition-d1ccc9-a1.
2026-10-29
Assessment publishedBy this siteGary Marcus
Many concerns raised about neural networks 30 years ago still hold true.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-happy-groundhog-day-the-ai-edition-889f4f-a2.
2026-10-29
Assessment publishedBy this siteGary Marcus
Deep learning has made progress on other dimensions but has not solved the core problems thought essential 30 years ago, so in that sense it has hit a wall.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-happy-groundhog-day-the-ai-edition-40a16a-a2.
2026-10-29
Assessment publishedBy this siteGary Marcus
Putting LLMs into production is hard and most deployment work to date has been preliminary.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-happy-1st-birthday-gpt-4-fe0f67-a1.
2026-10-29
Assessment publishedBy this siteGary Marcus
Over the past year there has been substantial progress in finding applications for and deploying GPT-4 class models.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-happy-1st-birthday-gpt-4-f00f06-a1.
2026-10-29
Assessment publishedBy this siteGary Marcus
The profitability of LLM businesses remains uncertain.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-happy-1st-birthday-gpt-4-d0e79a-a2.
2026-10-29
HeldBy this siteGary Marcus
Held record marcus-happy-1st-birthday-gpt-4-761602
The final-text check failed: citations_correct.
2026-10-29
HeldBy this siteGary Marcus
Held record marcus-happy-1st-birthday-gpt-4-245405
The openai review found the criteria unfair to the speaker's words.
2026-10-29
HeldBy this siteGary Marcus
Held record marcus-groupthink-versus-tulips-what-will-4e7b11
e1 is decisive for a financial claim but states no metric scope; e2 is decisive for a financial claim but states no metric scope; e3 is decisive for a financial claim but states no metric scope
2026-10-29
HeldBy this siteGary Marcus
Held record marcus-grok-3-hot-take-c726c8
The final-text check failed: rationale_supported_by_evidence, citations_correct; overclaims: The cited public record supports the historical conclusion that pure pretraining scaling had not demonstrated broadly human-level general intelligence by February 18, 2025 [e1, e2, e3, e4]. | The January 2025 International AI Safety Report documents uneven capabilities, difficulty with novel scenarios and failures on simple reasoning tasks [e1]. | Their qualified interpretation does not establish the broad human-level standard used here, but shows why 'clearly' is stronger than the consensus warrants..
2026-10-29
Assessment publishedBy this siteGary Marcus
Grok 3, as demoed, was not AGI, contrary to a Musk fan's prediction.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-grok-3-hot-take-5247eb-a1.
2026-10-29
Assessment publishedBy this siteGary Marcus
OpenAI's competitive moat keeps shrinking.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-grok-3-hot-take-31577b-a1.
2026-10-29
Assessment publishedBy this siteGary Marcus
Because OpenAI's moat keeps shrinking, price wars among AI model providers will continue.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-grok-3-hot-take-31577b-2-a1.
2026-10-29
Assessment publishedBy this siteGary Marcus
The Grok 3 launch did not test against full o3 and reported only a handful of benchmarks.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-grok-3-hot-take-302078-a2.
2026-10-29
Assessment publishedBy this siteGary Marcus
Grok 3 has not solved hallucinations.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-grok-3-hot-take-29a51f-a1.
2026-10-29
Assessment publishedBy this siteGary Marcus
The ASU study showed that even the latest, greatest models cannot reliably extend universals outside their training space of examples, so the Achilles' heel Marcus identified in 1998 still remains despite every gadget invented since.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-gpt-5-overdue-overhyped-and-underwhelming-e54d36-a1.
2026-10-29
HeldBy this siteGary Marcus
Held record marcus-gpt-5-overdue-overhyped-and-underwhelming-d1d61f
The final-text check failed: citations_correct.
2026-10-29
Assessment publishedBy this siteGary Marcus
Dozens of shots on goal, such as Grok, Claude and Gemini, have invariably faced the same problems, and distribution shift has never been solved.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-gpt-5-overdue-overhyped-and-underwhelming-b24250-a1.
2026-10-29
Assessment publishedBy this siteGary Marcus
Promises that AI will produce world-changing science rarely materialize.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-gpt-5-overdue-overhyped-and-underwhelming-589ed9-a1.
2026-10-29
Assessment publishedBy this siteGary Marcus
Pure scaling simply isn't the path to AGI and will not get us there.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-gpt-5-overdue-overhyped-and-underwhelming-2abc58-a1.
2026-10-29
Assessment publishedBy this siteGary Marcus
Sky-high expectations caused many to see GPT-5 as a major letdown, and OpenAI's credibility fell dramatically on launch night.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-gpt-5-overdue-overhyped-and-underwhelming-26d204-a1.
2026-10-29
Assessment publishedBy this siteGary Marcus
LLM development has hit a wall: real progress on some dimensions but stuck on others.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-gpt-5-overdue-overhyped-and-underwhelming-19e767-a1.
2026-10-29
Assessment publishedBy this siteGary Marcus
Models hyped as godlike turn out to be incremental advances.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-gpt-5-overdue-overhyped-and-underwhelming-0e13ae-a1.
2026-10-29
Assessment publishedBy this siteGary Marcus
In a quick and somewhat random look at X, Marcus easily found many predictions about GPT-5's arrival, going back to 2023, and they were almost always optimistic.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-gpt-5-now-arriving-gate-8-gate-9-f5cc4f-a2.
2026-10-29
Assessment publishedBy this siteGary Marcus
Projected arrival dates for GPT-5 have repeatedly slipped into the future.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-gpt-5-now-arriving-gate-8-gate-9-9cccde-a2.
2026-10-29
HeldBy this siteGary Marcus
Held record marcus-gpt-5-now-arriving-gate-8-gate-9-774448
The anthropic review reads the claim differently.
2026-10-29
Assessment publishedBy this siteGary Marcus
GPT-5 will not be released in 2024.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-gpt-5-now-arriving-gate-8-gate-9-459a41-a1.
2026-10-29
HeldBy this siteGary Marcus
Held record marcus-gpt-5-hot-take-ca5a48
The openai review found the criteria unfair to the speaker's words.
2026-10-29
HeldBy this siteGary Marcus
Held record marcus-gpt-5-hot-take-6d408c
The openai review found the criteria unfair to the speaker's words.
2026-10-29
Assessment publishedBy this siteGary Marcus
GPT-5 is still part of the pack rather than a giant leap forward; for example, Grok 4 beats it on ARC-AGI-2.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-gpt-5-hot-take-6bff2b-a1.
2026-10-29
Assessment publishedBy this siteGary Marcus
OpenAI left the ARC-AGI-2 comparison out of its GPT-5 livestream benchmarks, which made the livestream look like marketing rather than science.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-gpt-5-hot-take-5eb0a3-a1.
2026-10-29
Assessment publishedBy this siteGary Marcus
GPT-5 is not AGI.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-gpt-5-hot-take-5d0a8a-a1.
2026-10-29
Assessment publishedBy this siteGary Marcus
OpenAI still has no clear technical moat with GPT-5.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-gpt-5-hot-take-25d34f-2-a1.
2026-10-29
Assessment publishedBy this siteGary Marcus
In the GPT-5 announcement, OpenAI showed at least two graphs whose bars did not match the numbers.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-gpt-5-hot-take-085a4f-a1.
2026-10-29
HeldBy this siteGary Marcus
Held record marcus-give-me-compromise-or-give-me-chaos-d372bb
The final-text check failed: rationale_supported_by_evidence, citations_correct; overclaims: These expressed positions contradict the blanket claim of seeking exemption from all regulation; they support the narrower concern about lobbying to limit foundation-model obligations (e1, e2)..
2026-10-29
Assessment publishedBy this siteGary Marcus
The Code Interpreter and Wolfram Alpha plug-ins significantly extend GPT-4's ability to solve math and science problems.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-getting-gpt-to-work-with-external-6ad253-a2.
2026-10-29
Assessment publishedBy this siteGary Marcus
GPT-4 does not reliably understand how to use the Code Interpreter and Wolfram Alpha plug-ins.
First assessment published after both reviews and the final-text check passed.
Assessment marcus-getting-gpt-to-work-with-external-0ab0f2-a1.
2026-10-29
HeldBy this siteGary Marcus
Held record marcus-agi-by-2027-c93b79-2
The two reviews differ: anthropic historical_observation/not_applicable/supported, openai historical_observation/not_applicable/mixed.