Claims that can’t be checked by Gary Marcus, page 4 of 24
Deadlines are judged as of 2026-10-29. About Gary Marcus and the sources covered.
Claims that can’t be checked by Gary Marcus (2353)
Public evidence cannot settle these claims, so they are recorded and attributed, not scored. Most were screened out: both AI providers read the claim on its own and found it a value judgment, rhetoric, hedged, too vague to test, or about something no public evidence could show. The rest were found to be recommendations or too open to test in a full review.
Page 4 of 24: claims 301 to 400 of 2353, ordered by the date first said, newest first. To find a claim by its words, use search.
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“You can’t just look at the outputs, without investigating the underlying mechanisms, and conclude that two entities with similar outputs reach those similar outputs by similar means.”
Our reading: One must not look only at the outputs of two entities with similar outputs, without investigating the underlying mechanisms, and conclude that they reach those outputs by similar means.
No horizon
Scope: two entities with similar outputs
Source: The Pope appears to understand AI better than Geoffrey Hinton does.
Assessment: Can’t be checkedSaid
Observation
“a. That’s the very Eliza illusion I am pointing out. It is possible to fool people into seeing things that aren’t real (e.g., pareidolia, and”
Our reading: Grimes's subjective impression that there is something there in AI is the Eliza illusion Marcus is pointing out, and it is possible to fool people into seeing things that aren't real.
No horizon
Scope: Grimes's subjective impression
Source: The Pope appears to understand AI better than Geoffrey Hinton does.
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“And Hinton should know better.”
Our reading: Geoffrey Hinton should know better than to treat what LLM researchers create as beings.
No horizon
Source: The Pope appears to understand AI better than Geoffrey Hinton does.
Assessment: Can’t be checkedSaid
Forecast
“A month later, it already looks like the moment has passed; tokenmaxxing may have jacked up Anthropic’s second quarter, but it looks like it won’t last.”
Our reading: The tokenmaxxing trend that may have boosted Anthropic's second quarter already seems to have passed and probably will not last.
No horizon
Source: Breaking: bad news for three of the biggest IPOs in history
Assessment: Can’t be checkedSaid
Forecast
“The second sentence— “Tokens got burned for millions of dollars without any real significant ROI to show for it” – might well turn out to be the epitaph for an era.”
Our reading: The current era of AI spending may well be remembered as one in which millions of dollars of tokens were burned without significant ROI.
No horizon
Source: Breaking: bad news for three of the biggest IPOs in history
Assessment: Can’t be checkedSaid
Conditional forecast
“And it could get worse, if tokenmaxxing is indeed a fad whose time has already begun to pass.”
Our reading: If tokenmaxxing is indeed a fad whose time has already begun to pass, the poor AI-return picture in the projections he cites, and its possible fallout for the economy, could get worse.
No horizon
Only if: if tokenmaxxing is indeed a fad whose time has already begun to pass
Source: Breaking: bad news for three of the biggest IPOs in history
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“Tokenmaxxing is stupid.”
Our reading: Tokenmaxxing is stupid.
No horizon
Source: Breaking: bad news for three of the biggest IPOs in history
Assessment: Can’t be checkedSaid
Forecast
“A lot of banks may take a hint as well.”
Our reading: A lot of banks may take a hint as well.
No horizon
Qualifier: as well, following the scenario described just before
Source: If enough other companies report the same, the bubble pops. 🫧
Assessment: Can’t be checkedSaid
Conditional forecast
“Same logic may well apply to the upcoming IPOs of OpenAI and Anthropic.”
Our reading: If the S&P 500 rule changes are adopted, OpenAI's and Anthropic's IPOs may likewise be added quickly to index funds at fantasy-level early valuations, leaving retirement investors exposed to rapid price drops.
No horizon
Only if: if S&P 500 adopts proposed rule changes
Source: This one weird trick might cost your retirement fund billions
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“If you have a mutual fund and don’t want a significant part of your holdings to flip essentially automatically into speculative, not yet proven big tech plays like SpaceX, OpenAI and Anthropic, you might want to call your congressperson ASAP to complain about this rule, and the risks it could pose to your retirement.”
Our reading: Mutual fund holders who don't want a significant part of their holdings to flip essentially automatically into speculative, not yet proven big tech plays like SpaceX, OpenAI and Anthropic might want to call their congressperson ASAP to complain about the proposed S&P 500 rule change and the risks it could pose to their retirement.
No horizon
Scope: for mutual fund holders who don't want a significant part of their holdings to flip essentially automatically into speculative big tech plays
Source: This one weird trick might cost your retirement fund billions
Assessment: Can’t be checkedSaid
Observation
“The exact number is not given but that discount may well be bigger than the projected $559M profit.”
Our reading: Although the exact figure is not given, SpaceX's one-time compute discount to Anthropic may well be bigger than Anthropic's projected $559M quarterly profit.
No horizon
Qualifier: the exact discount figure is not given
Source: Checking the math behind OpenAI and Anthropic’s latest headlines
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“Caveat emptor, IPO buyers.”
Our reading: IPO buyers should beware.
No horizon
Source: Checking the math behind OpenAI and Anthropic’s latest headlines
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“Clearly impressive. But as with so much else, it should be viewed with skepticism.”
Our reading: OpenAI's claimed result on the Erdős planar unit distance problem, though impressive, should be viewed with skepticism.
No horizon
Source: Checking the math behind OpenAI and Anthropic’s latest headlines
Assessment: Can’t be checkedSaid
Observation
“Context matters, and it is unclear whether subsequent quarters will be profitable.”
Our reading: It is unclear whether Anthropic's quarters after the projected profitable one will be profitable.
No horizon
Source: Checking the math behind OpenAI and Anthropic’s latest headlines
Assessment: Can’t be checkedSaid
Observation
“The great Trump AI repositioning that I forecast may have begun!”
Our reading: Trump's forecast repositioning away from pro-AI-industry policy may have begun by May 2026.
No horizon
Assessment: Can’t be checkedSaid
Conditional forecast
“And if the backlash has enough force behind it, Trump may move further away from last year’s anti-regulatory policies.”
Our reading: If public backlash against AI is strong enough, Trump may move further away from his 2025 anti-regulatory AI policies.
No horizon
Only if: if the backlash has enough force behind it
Assessment: Can’t be checkedSaid
Forecast
“Every time I think about the numbers and lack of RoI for most pilot studies and the reliability problems I think so.”
Our reading: Marcus thinks the multitrillion-dollar investment in AI could turn out to be another epic, arrogance-fueled mistake like the Vietnam war.
No horizon
Qualifier: 'I think so' in answer to a 'could' question
Assessment: Can’t be checkedSaid
Conditional forecast
“As I argued at the end of Taming Silicon Valley, which I hope you will read or reread now that is so directly relevant, if the people unite, we might actually be able to get AI to a better place.”
Our reading: If the public unites, it might be possible to steer AI to a better place.
No horizon
Only if: if the people unite
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“More technical; also excellent: why we so badly need neurosymbolic AI and world models, and why software verification is more important than ever in the LLM era.”
Our reading: The AI field badly needs neurosymbolic AI and world models.
No horizon
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“More technical; also excellent: why we so badly need neurosymbolic AI and world models, and why software verification is more important than ever in the LLM era.”
Our reading: In the LLM era, software verification is more important than ever.
No horizon
Qualifier: in the LLM era
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“That’s not good for anyone, businesses or consumers.”
Our reading: The situation of many US AI bills without anything that feels like a coherent AI policy is not good for anyone, businesses or consumers.
No horizon
Scope: anyone, businesses or consumers
Source: US AI policy is a clumsy mess. Here’s what to do about it.
Assessment: Can’t be checkedSaid
Forecast
“Might sound nerdy, but the alternative, to which we are headed, is poorly considered chaos.”
Our reading: Without a coherent framework, US AI policy is headed toward poorly considered chaos.
No horizon
Qualifier: without a coherent framework for AI legislation
Source: US AI policy is a clumsy mess. Here’s what to do about it.
Assessment: Can’t be checkedSaid
Observation
“Instead, its techniques likely work best with things like coding and math, where formal verification (good old symbolic AI for the win!) can straightforwardly apply.”
Our reading: Mythos's techniques likely work best with things like coding and math, where formal verification can straightforwardly apply.
No horizon
Scope: things like coding and math
Source: Misplaced panic over AI progress
Assessment: Can’t be checkedSaid
Observation
“(Incidentally, this is yet another vindication of neurosymbolic AI – not proof that LLMs themselves can be perpetually scaled. Nor a proof that another trillion dollars will indefinitely continue the trends shown in the graph.)”
Our reading: Recent progress vindicates neurosymbolic AI and does not prove LLMs can be perpetually scaled or that another trillion dollars will continue the trend indefinitely.
No horizon
Source: Misplaced panic over AI progress
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“In short, there is no need (yet?) to panic.”
Our reading: In short, there is no need (yet?) to panic over the AI progress shown by the METR graph.
No horizon
Qualifier: (yet?) — tentative, questioning time qualifier
Source: Misplaced panic over AI progress
Assessment: Can’t be checkedSaid
Observation
“And most importantly, solving (some aspects of) software design is not open-ended intelligence. AI is definitely getting better at some things, but there is no reason to think that it is close to fully general yet.”
Our reading: AI is improving at some things but there is no reason to think it is close to fully general intelligence; software design progress is not open-ended intelligence.
No horizon
Source: Misplaced panic over AI progress
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“To anticipate that Anthropic will have $2T revenue in 2030 is a perfect example of what I have often called the trillion pound baby fallacy”
Our reading: Anticipating that Anthropic will have $2T revenue in 2030 is a perfect example of the trillion pound baby fallacy.
No horizon
Source: Misplaced panic over AI progress
Assessment: Can’t be checkedSaid
Causal claim
“As noted in my newsletter, a lot of the advance in recent months is likely from the incorporation of symbolic tools (like code interpreters, verification, and harnesses) rather than from model scaling per se.”
Our reading: Much of the recent-months AI advance is likely due to symbolic tools like code interpreters, verification and harnesses rather than model scaling itself.
No horizon
Qualifier: recent months
Source: Misplaced panic over AI progress
Assessment: Can’t be checkedSaid
Observation
“More broadly, the key problem with GenAI has been reliability; a graph that demands only 50% success does not address reliable performance. At all.”
Our reading: Reliability has been the key problem with GenAI, and the METR graph at 50% success does not address reliable performance at all.
No horizon
Source: Misplaced panic over AI progress
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“For exactly that reason, Wildeford’s concern about Mythos “breaking the graph” is a red herring.”
Our reading: Wildeford's concern that Mythos has 'broken' the METR time-horizon graph is a red herring, because the graph only measures 50% success.
No horizon
Qualifier: because the graph measures only 50% success
Source: Misplaced panic over AI progress
Assessment: Can’t be checkedSaid
Observation
“Claude Code is a real advance; Mythos probably builds on some of what is learned there.”
Our reading: Claude Code is a real advance, and Mythos probably builds on some of what was learned from it.
No horizon
Source: Misplaced panic over AI progress
Assessment: Can’t be checkedSaid
Forecast
“some types of challanges (e.g., reasoning accurately with respect to world models, reducing hallucinations etc) may simply not be amenable to current approaches”
Our reading: AI progress may hit limits from resource constraints, benchmark-focused tooling, formal verification limits in less formal problems, and challenges like world-model reasoning and hallucinations not amenable to current approaches.
No horizon
Source: Misplaced panic over AI progress
Assessment: Can’t be checkedSaid
Forecast
““they hadn’t figured out how OpenAI would pay for it” may turn out to be the epitaph for an entire era.”
Our reading: The line "they hadn't figured out how OpenAI would pay for it" may turn out to be the epitaph for an entire era.
No horizon
Qualifier: may turn out to be
Source: Breaking news: “they hadn’t figured out how OpenAI would pay for it”
Assessment: Can’t be checkedSaid
Rhetoric
“We were not wrong.”
Our reading: Marcus and Hamiel's earlier warning that LLM coding agents are a security nightmare has been borne out.
No horizon
Qualifier: refers to August essay 'LLM + Coding Agents = Security Nightmare'
Assessment: Can’t be checkedSaid
Causal claim
“Greed, overhype, and recklessness have screwed it up.”
Our reading: AI could have been great but greed, overhype, and recklessness have ruined it.
No horizon
Source: The growing AI backlash
Assessment: Can’t be checkedSaid
Forecast
“over-leveraged environment-damaging data centers that risk causing a recession”
Our reading: Over-leveraged AI data centers risk causing a recession.
No horizon
Source: The growing AI backlash
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“But Generative AI has hurt more than it has helped, and been managed irresponsibly.”
Our reading: Generative AI has hurt more than it has helped, and it has been managed irresponsibly.
No horizon
Source: The growing AI backlash
Assessment: Can’t be checkedSaid
Forecast
“Public backlash against AI will so be strong by 2028 that anti-AI sentiment will probably be a major factor in the 2028 US Presidential election.”
Our reading: Anti-AI public backlash will probably be strong enough by 2028 to be a major factor in the 2028 US Presidential election.
No horizon
Source: The growing AI backlash
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“Outside of coding (where there is clear value), and a handful of other domains (e.g. brainstorming), Generative AI has been a net negative for society.”
Our reading: Outside of coding, where there is clear value, and a handful of other domains such as brainstorming, generative AI has been a net negative for society.
No horizon
Scope: outside of coding and a handful of other domains such as brainstorming
Source: The growing AI backlash
Assessment: Can’t be checkedSaid
Forecast
“Frustration with AI may even factor into the midterms.”
Our reading: Frustration with AI may be a factor in the 2026 US midterm elections.
No horizon
Source: The growing AI backlash
Assessment: Can’t be checkedSaid
Observation
“That prediction very seems to be on track.”
Our reading: His earlier prediction, that anti-AI sentiment will probably be a major factor in the 2028 US Presidential election, seems to be on track.
No horizon
Qualifier: seems to be
Source: The growing AI backlash
Assessment: Can’t be checkedSaid
Forecast
“All that said, I honestly believe some future form of AI might be great.”
Our reading: Some future form of AI, unlike current generative AI, might be great.
No horizon
Source: The growing AI backlash
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“I don’t think we are better off than we were four years ago.”
Our reading: He does not think we are better off than we were four years ago.
No horizon
Qualifier: I don't think
Source: The growing AI backlash
Assessment: Can’t be checkedSaid
Observation
“AI will surely someday be a major boon for medicine, but current tools such as domain-general chatbots may not be up to the job.”
Our reading: Current AI tools such as domain-general chatbots may not be adequate for the job of improving medicine.
No horizon
Qualifier: current tools
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“Inferring sentience directly from outputs is naive.”
Our reading: Inferring sentience in AI systems directly from their outputs is naive.
No horizon
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“Dawkins should never have glorified such a thing.”
Our reading: Dawkins should never have glorified Claude, which Marcus describes as akin to a counterfeit person.
No horizon
Assessment: Can’t be checkedSaid
Observation
“Claude’s outputs are the product of a form of mimicry, rather than as a report of genuine internal states.”
Our reading: Claude's outputs come from a form of mimicry and are not reports of genuine internal states.
No horizon
Assessment: Can’t be checkedSaid
Observation
“Consciousness is about internal states; the mimicry, no matter how rich, proves very little.”
Our reading: LLM mimicry of human speech, however rich, proves very little about consciousness.
No horizon
Qualifier: no matter how rich the mimicry
Assessment: Can’t be checkedSaid
Observation
“Dawkins seems to imagine that since LLMs say things people do, they must be like people, and that simply does not follow.”
Our reading: It does not follow from LLMs saying things people say that LLMs are like people.
No horizon
Assessment: Can’t be checkedSaid
Observation
“And there is no reason to think that Claude feels anything at all.”
Our reading: There is no reason to think Claude feels anything at all.
No horizon
Assessment: Can’t be checkedSaid
Observation
“A chess computer is by some definitions intelligent, but that doesn’t make it conscious.”
Our reading: Machine intelligence, as in a chess computer, does not entail consciousness.
No horizon
Assessment: Can’t be checkedSaid
Rhetoric
“Claude is akin to a counterfeit person.”
Our reading: Claude is akin to a counterfeit person.
No horizon
Assessment: Can’t be checkedSaid
Observation
“Replace LaMDA with Claude, and every word still applies.”
Our reading: His 2022 critique of LaMDA applies fully to Claude: it matches statistical patterns, its words lack meaning, it has no coherent understanding of the world, and it is not sentient.
No horizon
Assessment: Can’t be checkedSaid
Observation
“I am sure Claude can draw on its training data to wax poetic about orgasm, but that doesn’t mean it has ever felt one.”
Our reading: Claude being able to draw on its training data to wax poetic about orgasm, which Marcus is sure it can do, does not mean Claude has ever felt one.
No horizon
Qualifier: I am sure Claude can draw on its training data to wax poetic about orgasm
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“Coders (especially vibe coders with little experience) beware.”
Our reading: Coders, especially inexperienced vibe coders, should be wary of the robustness of AI-generated code.
No horizon
Qualifier: especially inexperienced vibe coders
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“And all you OpenClaw devotees, that goes 10x, if not 100x, for you.”
Our reading: OpenClaw devotees should beware the robustness of AI-generated code even more than other coders, by what Marcus puts as '10x, if not 100x'.
No horizon
Scope: OpenClaw devotees
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“As it should.”
Our reading: Skepticism about AI's technical and economic prospects, including the heavy spending on AI infrastructure, should be rising.
No horizon
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“The world would be better off if the transition to for-profit was unwound forcing the company to benefit humanity, as Musk has asked.”
Our reading: OpenAI's transition to for-profit should be unwound so the company is forced to benefit humanity.
No horizon
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“I hope his lawyers will keep focus on OpenAI’s pattern of behavior, rather than on his own contributions to the company.”
Our reading: Musk's lawyers should focus the trial on OpenAI's pattern of behavior rather than Musk's contributions.
No horizon
Assessment: Can’t be checkedSaid
Observation
“Tactically, from what I can glean from the reporting, Musk may be making the trial too much about himself.”
Our reading: Musk may be making the OpenAI trial too much about himself.
No horizon
Qualifier: based on reporting
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“AI agents are wildly premature technology that is being rolled out way too fast.”
Our reading: AI agents are wildly premature technology that is being rolled out way too fast.
No horizon
Source: Dario Amodei, hype, AI safety, and the explosion of vibe-coded AI disasters
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“In the hands of very skilled practitioners who pay a lot of attention, and treat the outputs with considerable scrutiny, coding agents can be astonishing. But that kind of expert knowledge, which coding agents can’t be counted on for, is exactly why we need to keep software engineers in the loop.”
Our reading: Because coding agents can't be counted on for expert knowledge, software engineers need to be kept in the loop.
No horizon
Qualifier: coding agents can't be counted on for expert knowledge
Source: Dario Amodei, hype, AI safety, and the explosion of vibe-coded AI disasters
Assessment: Can’t be checkedSaid
Causal claim
“Such a total failure was, of course, totally predictable (as I noted on X in a tweet with around 2 million views), and unsurprising in a system that mimics data without truly understanding what was asked of it.”
Our reading: The viral vibe-coding data-loss failure was predictable because the system mimics data without truly understanding what was asked.
No horizon
Source: Dario Amodei, hype, AI safety, and the explosion of vibe-coded AI disasters
Assessment: Can’t be checkedSaid
Forecast
“Maintainability is yet another serious issue beyond the recurring issues of data loss, privacy leaks, and security breaches. In the long run, it may be the most serious.”
Our reading: Maintainability of AI-generated code is a serious issue and may in the long run be the most serious one.
No horizon
Qualifier: in the long run
Source: Dario Amodei, hype, AI safety, and the explosion of vibe-coded AI disasters
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“AI coding tools are genuinely revolutionizing the software industry, but more than almost anything I have ever seen, they fall into the category of things you maybe shouldn’t try at home without significant prior experience:”
Our reading: People maybe shouldn't use AI coding tools without significant prior experience.
No horizon
Qualifier: without significant prior experience
Source: Dario Amodei, hype, AI safety, and the explosion of vibe-coded AI disasters
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“𝗪𝗵𝗶𝗰𝗵 𝗶𝘀 *𝗲𝘅𝗮𝗰𝘁𝗹𝘆* 𝘄𝗵𝘆 𝘄𝗲 𝘀𝗵𝗼𝘂𝗹𝗱 𝗻𝗼𝘁 𝘁𝗿𝘂𝘀𝘁 𝗰𝗼𝗱𝗶𝗻𝗴 𝗮𝗴𝗲𝗻𝘁𝘀.”
Our reading: We should not trust coding agents.
No horizon
Source: Dario Amodei, hype, AI safety, and the explosion of vibe-coded AI disasters
Assessment: Can’t be checkedSaid
Observation
“There seems to be some excitement around “ChatGPT’s powerful new image engine”, but as ever, its functional understanding of the world seems limited.”
Our reading: ChatGPT's new image engine appears to have limited functional understanding of the world.
No horizon
Assessment: Can’t be checkedSaid
Observation
“The system doesn’t actually understand how the various parts function.”
Our reading: ChatGPT's new image engine does not actually understand how the parts of a bike function.
No horizon
Assessment: Can’t be checkedSaid
Observation
“As in the first example, the lack of functional understanding is manifest.”
Our reading: ChatGPT's new image engine's tall tandem bike image clearly shows a lack of functional understanding.
No horizon
Qualifier: tall tandem bike with rack and panniers prompt
Assessment: Can’t be checkedSaid
Causal claim
“(An expert doctor might use the LLM to better effect, by asking the right questions.)”
Our reading: An expert doctor might use an LLM to better effect than lay users by asking the right questions.
No horizon
Qualifier: by asking the right questions
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“There will always be better models, but for now, and until proven otherwise, we should not take the apparent “confidence” of large language models — itself an illusion of how they are trained — to mean that we should trust large language with our lives.”
Our reading: For now and until proven otherwise, people should not take LLMs' apparent confidence as reason to trust them with their lives.
No horizon
Qualifier: for now
Assessment: Can’t be checkedSaid
Rhetoric
“Maybe I spoke too soon.”
Our reading: Marcus may have spoken too soon in saying that the Oracle stock spike would be seen as the peak of absurdity, because a new example has now appeared.
No horizon
Source: Peak absurdity, Part II
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“Even if Mythos was somewhat oversold in the media, the time to get our cybersecurity house in order is now (or better yet last year) — especially given the sudden profusion of agent-written code that may in fact be both weakly defended and vulnerable.”
Our reading: Cybersecurity should be shored up immediately in response to AI attack capabilities, especially given the surge in agent-written code.
No horizon
Qualifier: even if Mythos was oversold
Source: Claude Mythos, evaluated
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“In reality, the replies to the Apple reasoning paper were weak sauce (see also this).”
Our reading: The replies to the Apple reasoning paper were weak.
No horizon
Source: Even more good news for the future of neurosymbolic AI
Assessment: Can’t be checkedSaid
Observation
“But there is more and more reason to think that neurosymbolic AI is a stronger starting point, with potential for better generalization and more efficient performance.”
Our reading: Evidence increasingly suggests neurosymbolic AI is a stronger starting point than pure LLMs, with potential for better generalization and efficiency.
No horizon
Qualifier: compared with LLMs
Source: Even more good news for the future of neurosymbolic AI
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“As a society we could spend another trillion dollars pursue LLMs, or a trillion dollars seeking new, hybrid approaches. I for one know which I would pick.”
Our reading: Given a choice between spending another trillion dollars on LLMs or a trillion dollars on new hybrid approaches, Marcus says he knows which he would pick, implicitly the hybrid approaches.
No horizon
Qualifier: stated as personal preference between hypothetical investments
Source: Even more good news for the future of neurosymbolic AI
Assessment: Can’t be checkedSaid
Assessment: Can’t be checkedSaid
Causal claim
“Meanwhile, the implications for the allocation of capital are pretty massive: smartly adding in bits of symbolic AI can do a lot more than scaling alone, and even Anthropic has now discovered (though they haven’t acknowledged publicly) scaling is no longer the essence of innovation.”
Our reading: Smartly adding in bits of symbolic AI can do a lot more than scaling alone, Anthropic has discovered (without acknowledging it publicly) that scaling is no longer the essence of innovation, and the implications for capital allocation are pretty massive.
No horizon
Qualifier: smartly adding in bits of symbolic AI
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“What we really need to do to get trustworthy AI rather than the current unpredictable “jagged” mess, is to go in the knowledge-, reasoning-, and world-model driven direction I laid out in 2020, in an article called The Next Decade in AI, in which neurosymbolic AI is just the *starting point* in a longer journey.”
Our reading: To get trustworthy AI, the field should pursue knowledge-, reasoning- and world-model driven approaches beyond neurosymbolic AI, as current AI is an unpredictable jagged mess.
No horizon
Qualifier: to get trustworthy AI
Assessment: Can’t be checkedSaid
Observation
“But Anthropic figured out that if you really need to get your patterns right, you can’t trust a pure LLM. They are too probabilistic. And too erratic.”
Our reading: Pure LLMs are too probabilistic and erratic to be trusted when patterns must be gotten right, as Anthropic recognized.
No horizon
Qualifier: if you really need to get your patterns right
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“The demo was definitely proof of concept that we need to get our regulatory and technical house in order, but not the immediate threat the media and public was lead to believe.”
Our reading: The Mythos demo shows we need to get our regulatory and technical house in order.
No horizon
Source: Three reasons to think that the Claude Mythos announcement from Anthropic was overblown
Assessment: Can’t be checkedSaid
Observation
“Mythos is more sophisticated but perhaps not head-and-shoulders the way it was portrayed.”
Our reading: Mythos is more sophisticated than open-weight models but perhaps not as far ahead as it was portrayed.
No horizon
Source: Three reasons to think that the Claude Mythos announcement from Anthropic was overblown
Assessment: Can’t be checkedSaid
Rhetoric
“To a certain degree, I feel that we were played.”
Our reading: Marcus feels that, to a certain degree, we were played by the Mythos announcement.
No horizon
Qualifier: I feel
Source: Three reasons to think that the Claude Mythos announcement from Anthropic was overblown
Assessment: Can’t be checkedSaid
Forecast
“But some of their competitors (such as OpenAI and xAI) might well not.”
Our reading: Competitors such as OpenAI and xAI might well not show restraint and could publicly release potentially dangerous technology.
No horizon
Source: What should we take from Anthropic’s (possibly) terrifying new report on Mythos?
Assessment: Can’t be checkedSaid
Observation
“It’s probably not as bad as they say, as AI and cybersecurity expert Heidy Khlaaf explains in this thread, which I highly recommend:”
Our reading: Mythos's cybersecurity danger is probably not as bad as claimed.
No horizon
Source: What should we take from Anthropic’s (possibly) terrifying new report on Mythos?
Assessment: Can’t be checkedSaid
Observation
“It probably isn’t; it’s telling that the report says very little about overall capabilities.”
Our reading: Mythos is probably not AGI, as suggested by the report saying little about overall capabilities.
No horizon
Source: What should we take from Anthropic’s (possibly) terrifying new report on Mythos?
Assessment: Can’t be checkedSaid
Observation
“Whether Mythos is as scary as it sounds or not, the reality is that without any government oversight on what can be released, we are entirely at the mercy of individual CEOs, some of whom have decidedly not earned our trust.”
Our reading: Without government oversight of AI releases, the public is entirely at the mercy of individual AI CEOs, some of whom have not earned trust.
No horizon
Qualifier: without government oversight on releases
Source: What should we take from Anthropic’s (possibly) terrifying new report on Mythos?
Assessment: Can’t be checkedSaid
Forecast
“Mythos may wreak havoc even if it hallucinates and lacks reliability outside of domains like coding and math.”
Our reading: Mythos may wreak havoc even if it hallucinates and is unreliable outside domains like coding and math.
No horizon
Qualifier: even if it hallucinates and lacks reliability outside coding and math
Source: What should we take from Anthropic’s (possibly) terrifying new report on Mythos?
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“We need an international agency and treaty.”
Our reading: An international AI agency and treaty are needed.
No horizon
Source: What should we take from Anthropic’s (possibly) terrifying new report on Mythos?
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“Self-regulation is too little, too late.”
Our reading: Self-regulation (of AI companies) is too little, too late.
No horizon
Source: What should we take from Anthropic’s (possibly) terrifying new report on Mythos?
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“Anthropic showed some admirable restraint in not publicly releasing a potentially dangerous technology.”
Our reading: Anthropic showed some admirable restraint in not publicly releasing a potentially dangerous technology (Mythos).
No horizon
Qualifier: some admirable restraint
Source: What should we take from Anthropic’s (possibly) terrifying new report on Mythos?
Assessment: Can’t be checkedSaid
Rhetoric
“I can’t tell exactly how far Mythos (or similar competitive systems that will no doubt follow) leave us up that creek, but I do know that the time for crafting paddles is running out.”
Our reading: Time to build safeguards and regulation against AI risks is running out.
No horizon
Source: What should we take from Anthropic’s (possibly) terrifying new report on Mythos?
Assessment: Can’t be checkedSaid
Causal claim
“The situation is this: three years of self-serving and misleading arguments about how regulation would allegedly preclude innovation has left us up shit’s creek, without a paddle.”
Our reading: Three years of self-serving and misleading arguments that regulation would allegedly preclude innovation have left us in a bad predicament ('up shit's creek, without a paddle').
No horizon
Scope: three years of arguments
Source: What should we take from Anthropic’s (possibly) terrifying new report on Mythos?
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“1Anthropic also announced a big consortium, partly supported by their financial support, to work on cybersecurity, which is likely a good thing, though one that should work in conjunction with law enforcement and regulation.”
Our reading: Anthropic's cybersecurity consortium is likely a good thing but should work in conjunction with law enforcement and regulation.
No horizon
Source: What should we take from Anthropic’s (possibly) terrifying new report on Mythos?
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“A central subtheme of this newletter, brought together in this 2024 Guardian article, has been that Sam Altman should not be trusted:”
Our reading: Sam Altman should not be trusted.
No horizon
Assessment: Can’t be checkedSaid
Observation
“As a postscript, another thing I have noted about Altman is that when he is in trouble, he often tries to change the narrative with hype,”
Our reading: When Sam Altman is in trouble, he often tries to change the narrative with hype.
No horizon
Qualifier: when he is in trouble
Assessment: Can’t be checkedSaid
Observation
“the economics of his company make no sense.”
Our reading: OpenAI's economics make no sense.
No horizon
Assessment: Can’t be checkedSaid
Observation
“Today’s new OpenAI report on Superintelligence (allegedly a New Deal for the AI era) isn’t about something real or imminent (as far as I can tell),”
Our reading: As far as Marcus can tell, OpenAI's April 2026 Superintelligence report is not about anything real or imminent.
No horizon Evidence as of
Limitation: This does not endorse OpenAI's expectations about superintelligence.
Assessment: Can’t be checkedSaid
Causal claim
“instead I suspect that it’s just another Altmanian attempt to distract, in this case from the fact that the economics of his company make no sense.”
Our reading: Marcus suspects that OpenAI's new Superintelligence report is just another Altman attempt to distract, in this case from the fact that the economics of OpenAI make no sense.
No horizon
Qualifier: I suspect
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“All in all, glorifying Medvi was not The New York Times’ finest hour.”
Our reading: Glorifying Medvi was not The New York Times' finest hour.
No horizon
Qualifier: all in all
Source: The back story behind the first “$1.8 Billion” dollar “AI Company”