Changes, page 56 of 60

Everything we published, changed, or held back, newest first.

All changes (11,844)

Changes 11,001 to 11,200 of 11,844, newest first. Page 56 of 60.

2026-10-29
PublishedBy usGary Marcus
There should be an international agreement not to deploy demonstrably harmful AI architectures, and frontier labs should be held accountable for lax security.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
It may still be possible to get a net positive or even great outcome from AI, possibly transforming technology and medicine.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
More than 20 countries signed a call for action on AI that captures much of what Marcus has long advocated.
A reviewer didn’t accept one of the pieces of evidence.
2026-10-29
PublishedBy usGary Marcus
We need to wean ourselves from an addiction to large language models and to foster more research into outside-the-box alternatives that are more interpretable and more tractable.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
The choices made now about AI may have effects that last for generations.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Giving up monitoring is a mistake, and OpenAI should not have been allowed to decide on its own to release a less monitorable model.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
An international commission should continuously audit AI systems after release, and independent scientists on it should have real power to demand changes or recall dangerous systems.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
Belief in magic may be declining
The final check failed: a cited source didn’t say what the write-up said.
2026-10-29
Held backBy usGary Marcus
Gemini can't even remember who was on Apollo 11.
A piece of evidence came from the wrong dates.
2026-10-29
PublishedBy usGary Marcus
ChatGPT had an unexplained meltdown.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Gary Marcus has said from day 1 that LLMs are oversold.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
Belief in magic may be declining
One of the AIs read the claim differently from our summary.
2026-10-29
PublishedBy usGary Marcus
Social media platforms will be polluted by automated misinformation much as the internet as a whole will be.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Automatic misinformation generation tools will accelerate the degradation of the internet.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Businesses that do not use automatic misinformation generation systems may be at a disadvantage.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Wholesale generation of misinformation should be punishable and enforced against.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
The killer app of AGI may be killing truth.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Because the White House AI agreements with 7 tech companies are voluntary, they are inadequate and do nothing to stop bad actors building their own automated misinformation software.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
We need AI smart enough to know when it is being used to spread falsehoods.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
The temptation for all political sides to use automatic misinformation tools is likely to be overwhelming.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
We ought to demand that any automatically generated content like this (the CounterCloud-style misinformation discussed) be labeled and watermarked, and no existing law requires either.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Elections may hinge on automatic misinformation generation systems.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Existing laws are not sufficient to protect against the full scope of harms generative AI is beginning to enable.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
We have no AI smart enough to know when it is being used to spew falsehood, and building it would require a major reshaping of research directions toward AI based in facts and competent in reasoning, rather than just the powerful text-prediction tools we have now.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
As far as the author knows, wholesale misinformation is not illegal in the US except arguably in narrow domains like securities fraud.
A reviewer said our test wasn’t fair to what was actually said.
2026-10-29
PublishedBy usGary Marcus
OpenAI tempered expectations for its next event and said we would not see GPT-5 at that event.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Nvidia's stock drop could be due to production delays and also to growing market concern about whether its big GenAI customers can make money deploying its chips.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Nvidia stock fell 6% on August 5, 2024, and 20% over the previous month.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
LLMs on their own are not and will never be AGI.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Scaling alone was never going to be enough to reach AGI.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
August 5, 2024, a big day in tech?
One of the AIs read the claim differently from our summary.
2026-10-29
PublishedBy usGary Marcus
Elon Musk's new lawsuit against OpenAI could force a court to discuss what AGI means.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
August 5, 2024, a big day in tech?
One of the AIs read the claim differently from our summary.
2026-10-29
PublishedBy usGary Marcus
It is not out of the question that August 5, 2024 could someday end up being seen as a turning point.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Adding AI into the mix of social media could easily make things a lot worse.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
OpenAI's investors may well succeed in getting the board to resign and bringing Sam Altman back.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
The public has no say in how OpenAI's governance crisis is resolved, including its implications for safety and humanity.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
People should check whether boasted ARR means Annual Recurring Revenue or Annualized Run Rate.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Anthropic's Q2 2026 level of revenue may well not recur at the same scale in 2027, as companies shift to cheaper open source models.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
Following the collapse of tokenmaxxing, companies like AT&T are increasingly turning to open source models from other vendors to save costs.
A piece of evidence came from the wrong dates.
2026-10-29
Held backBy usGary Marcus
ARR vs ARR. Watch out for this one sly trick.
The two reviewers read the claim differently.
2026-10-29
Held backBy usGary Marcus
Are social media and coordinated misinformation campaigns fomenting extremism?
One AI wasn’t sure who said it.
2026-10-29
PublishedBy usGary Marcus
TikTok has no interest in showing users a balanced or statistically representative view of the world.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
TikTok's recommendation algorithm, when applied to daily news, is likely doing serious damage to people's ability to co-exist.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
It is not inconceivable that AI-generated propaganda could be significantly intensifying the scale and impact of a possible foreign (e.g., Russian) active-measures campaign to destabilize the United States through social media posts.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Many AI researchers have started regularly receiving emails from people who believe they have seen signs of consciousness in LLMs.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
Current AI only absorbs and mimics data and never reflects critically on it.
A reviewer said our test wasn’t fair to what was actually said.
2026-10-29
PublishedBy usGary Marcus
Current AI systems cannot tell jokes and fiction apart from reality, while well-motivated humans usually can.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
An LLM just spits back what it has read, sometimes in transmogrified form.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Everyone should stand with Anthropic in its lawsuit against the US government.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Gary Marcus fully supports Anthropic's new lawsuit against the US government, although he says Dario Amodei is no saint.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
Anthropic has broad support against Hegseth's actions, which are unprecedented.
A piece of evidence came from the wrong dates.
2026-10-29
Held backBy usGary Marcus
LLMs will never be secure, trustworthy AI.
The final check failed: a cited source didn’t say what the write-up said.
2026-10-29
PublishedBy usGary Marcus
The attack surface of LLMs appears to be infinite, violating the cybersecurity principle of keeping attack surfaces small.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
There appears to be an endless number of ways to attack LLMs.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
With hundreds of millions of people using LLMs daily and companies like OpenAI promising to put them in all-purpose agents with essentially full device-level access, the lack of any kind of guarantee about LLMs gets more worrisome by the day, though it would be fine if LLMs were kept in the lab, where they arguably still belong.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Even as some jailbreaks are patched, there will very likely always be more jailbreak attacks on LLMs.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
LLMs arguably still belong in the lab rather than in widespread deployment.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
Because nobody knows exactly how LLMs work, nobody can ever issue strong guarantees about them.
A reviewer said our test wasn’t fair to what was actually said.
2026-10-29
Held backBy usGary Marcus
An Open Letter to Sam Altman
One of the AIs read the claim differently from our summary.
2026-10-29
Held backBy usGary Marcus
Much of SB-1047's requirements are limited to models with training runs of $100 million or more, and companies that can afford such runs, presumably valued in the billions, are not exactly little tech.
The final check failed: a cited source didn’t say what the write-up said.
2026-10-29
PublishedBy usGary Marcus
Much of what SB-1047 would require is something AI companies already voluntarily committed to at the White House and in Seoul.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Asking for standards is not unique to AI: in many industries, such as pharmaceuticals, aviation and automobiles, it is common to ask companies to evaluate the safety of their products against set standards.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
As Marcus understands the latest version of SB-1047, its kill switch requirement does not apply to open-source models once they are out of the original developer's control.
A reviewer said our test wasn’t fair to what was actually said.
2026-10-29
PublishedBy usGary Marcus
Marcus finds it reasonable for SB-1047 to focus on specific critical harms, and he does not understand what would justify an exemption from those provisions.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
SB-1047 does not require AI developers to predict every possible use of their models.
A reviewer didn’t accept one of the pieces of evidence.
2026-10-29
PublishedBy usGary Marcus
With no other serious proposal on offer, Marcus personally favours SB-1047, though he would welcome discussion of alternatives.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
SB-1047 requires developers only to implement reasonable safeguards against critical harms, not to fully mitigate them.
The final check failed: a cited source didn’t say what the write-up said.
2026-10-29
PublishedBy usGary Marcus
The US should at least consider pivoting toward working with China on AI for global good instead of starting a new AI cold war.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
If the US starts a new AI cold war with China, it will probably end uncomfortably for everyone, with no clear winners.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
An epistemic clusterfuck in the making
One of the AIs read the claim differently from our summary.
2026-10-29
PublishedBy usGary Marcus
Automatically amplifying the already inherent tendency towards extremism without fact-checking or reasoned analysis is a recipe for chaos.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
The epistemic situation may actually get worse than social media, which Marcus says might itself be an example of an epistemic clusterfuck.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
LLMs whose synthesis is confined to X tweets will be even less grounded in reality than LLMs in general.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
In Marcus's opinion, the investor audience at the forum was being lied to: they were being assured that everything in AI is under control and well plotted out, when in fact science is hard and the field is in its infancy.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Many people will not be able to afford or want humanoid robots, and there will not be six humanoid robots per car anytime soon.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Quantitative vibes, meaning gut feelings expressed as numbers, are still just vibes.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Humanoid robots will not sell 200 times Roomba's roughly 50 million units in the near term.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
As of October 2024, nobody knows how to build even one safe, reliable, generally useful humanoid robot at any price.
The final check failed: a cited source didn’t say what the write-up said.
2026-10-29
PublishedBy usGary Marcus
Even if one believed scaling was infinite and that useful data would not run out, there is just no way to predict all this, including how efficient future models would be in energy and chips, out to 2040 with any certainty.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Because investors working with other people's money get paid large fees even when investments do not pan out, so many of them love plausible stories with big numbers.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
AI will not be able to do anything any human can do within the next one to two years, and Marcus would bet a million dollars on it.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Masayoshi Son's figure that artificial superintelligence will be 10,000 times smarter than humans is a made-up round number with no real meaning; pure hype.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
By OpenAI's own data, o1 is certainly not 100 times better than GPT-4.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
People should not take Musk's or Son's AI predictions from the Riyadh forum seriously.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Generative AI uses so much power that it keeps fossil fuels in demand, which benefits oil-rich countries whether or not GenAI ever becomes reliable.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Those making these outlandish claims get free press for them, and that drives up the whole thing, at least for a while.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
Elon Musk has a track record of overoptimistic AI predictions, and his humanoid robot prediction is another one.
The check used the wrong kind of test for this claim.
2026-10-29
PublishedBy usGary Marcus
When Musk extrapolated that AI would be 10,000 times better in 4 years, he did not consider potential bottlenecks around data or compute, or inherent problems with LLMs.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Nobody can predict AI's future power, chip, and capital needs with the precision Son claimed.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
AI, even in its current form, is about to have a huge impact on society.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
AI should pursue hybrid approaches combining neural networks with reasoning and structured representations.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Current approaches to getting machines to reason about causality are not adequate, at least not yet.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
The remaining years of the 2020s will be wild for AI.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Recent advances in AI-generated content may lead to an avalanche of misinformation.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Contemporary approaches to AI may never solve reasoning, abstraction, compositionality, and factuality, which any intelligent machine should have.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
AI research should place greater emphasis on understanding human intelligence.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
SpaceX will not return 2,538x its IPO investment as Amazon did, which would imply a market cap of about $4,442 trillion.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
If AI boosts productivity, the boost may appear first in coding.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
A 2.7x increase in productivity relative to the entire economy is an enormous ask, though if there is a productivity boost it will perhaps show up first in coding.
One AI thought this could be checked, so it needs a full check first.
2026-10-29
Held backBy usGary Marcus
The AI productivity boom shows up in investment data but not yet in productivity data.
One AI thought this could be checked, so it needs a full check first.
2026-10-29
PublishedBy usGary Marcus
OpenAI is in trouble.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
OpenAI may not survive.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
An entire industry is being propped up by math that is insane.
The two reviewers read the claim differently.
2026-10-29
Held backBy usGary Marcus
SpaceX stock will never deliver returns comparable to Amazon's since its IPO.
The final check failed: the write-up said more than the evidence showed.
2026-10-29
PublishedBy usGary Marcus
Buying the SpaceX IPO reflects poor understanding of math or history.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
The costs of developing AI data centers, including OpenAI's Stargate, are rising rapidly.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Generative AI is a risky bet that could pay off or could well tank the economy.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
An entire industry is being propped up by math that is insane.
One of the AIs read the claim differently from our summary.
2026-10-29
Held backBy usGary Marcus
AI is not yet profitable.
A reviewer didn’t accept one of the pieces of evidence.
2026-10-29
PublishedBy usGary Marcus
AI may never become profitable.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Anthropic has a chance of becoming profitable.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
An entire industry is being propped up by math that is insane.
One of the AIs read the claim differently from our summary.
2026-10-29
Held backBy usGary Marcus
An entire industry is being propped up by math that is insane.
The two reviewers read the claim differently.
2026-10-29
PublishedBy usGary Marcus
To Marcus, Midjourney's output closely resembling copyrighted film stills sure looks like automated, digital plagiarism.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
An artist fights back, and Midjourney has embarrassed themselves
The final check failed: a cited source didn’t say what the write-up said.
2026-10-29
PublishedBy usGary Marcus
The evidence suggests Midjourney was trained on high-resolution copyrighted images, which it may or may not have a license for.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
Southen's experiment shows that a Midjourney user could easily infringe copyright without meaning to.
The claim is hedged, so it isn’t scored.
2026-10-29
Held backBy usGary Marcus
An artist fights back, and Midjourney has embarrassed themselves
One of the AIs read the claim differently from our summary.
2026-10-29
Held backBy usGary Marcus
Southen's experiment getting Midjourney to produce near-copies of copyrighted film stills took him almost no effort.
The two reviewers reached different verdicts.
2026-10-29
PublishedBy usGary Marcus
Midjourney's output was not labelled as being directly based on copyrighted work.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
Midjourney retaliated against Southen by revoking his access and wiping his history on the same day.
A reviewer said our test wasn’t fair to what was actually said.
2026-10-29
PublishedBy usGary Marcus
Scaling laws are merely empirical generalizations that held for a limited period while enough fresh data and compute were available.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
The era when simply adding data and compute produced large gains may well be over.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
If the rumor that a second major AI company is seeing diminishing returns from scaling is verified, the AI industry could see the equivalent of a bank run.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
There has never been a principled argument that scaling would solve hallucinations, reasoning, or open-world edge cases, or that synthetic data would suffice indefinitely.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Marcus strongly suspects that pure scaling of AI models has reached a point of diminishing returns, as of November 2024.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Scaling (improving AI by adding more data and compute) was always a hypothesis.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
The American people should have a say in AI policy, especially AI policy of this magnitude, and Congress should deliberate.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Hegseth's demand would in principle allow Anthropic's software to be applied to nuclear weapons without humans in the loop.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Hegseth's deadline to Anthropic is a power grab meant to circumvent Congress and cut others out of the decision.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Nothing Marcus has read convinces him that Secretary Hegseth has a nuanced understanding of the strengths and limits of current AI, or that Hegseth will show restraint in how he applies it.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Marcus endorses the reader's view that whether AI may be used for surveillance of Americans and lethal strikes without humans in the loop must be decided publicly by Congress, not forced by the Pentagon.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
We are headed exactly toward a situation where one individual, a Cabinet official, decides at gunpoint on mass surveillance and AI-fueled weapons, possibly nuclear, without humans in the loop.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Mass surveillance and AI-fueled weapons, possibly nuclear, without humans in the loop are categorically not things that one individual, even one in the Cabinet, should be allowed to decide at gunpoint.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
America, and probably the world, stands on a precipice.
The final check failed: a cited source didn’t say what the write-up said, and a condition they stated was left out.
2026-10-29
Held backBy usGary Marcus
Hegseth is demanding, under heavy threats, full unrestricted US military access to Anthropic's AI for uses including surveillance and autonomous weapons without humans in the loop.
A reviewer didn’t accept one of the pieces of evidence.
2026-10-29
PublishedBy usGary Marcus
Because the Block Nuclear Launch by Autonomous AI Act failed, there might be no human like Stanislav Petrov to stop a mistaken nuclear launch next time.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
America, and probably the world, stands on a precipice.
One of the AIs read the claim differently from our summary.
2026-10-29
PublishedBy usGary Marcus
Unrestricted military use of Anthropic's AI could well extend to nuclear weapons.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
If Hegseth forces Anthropic's Dario Amodei to fold, two monstrous precedents will be set.
The two reviewers reached different verdicts.
2026-10-29
PublishedBy usGary Marcus
The real reason for Sam Altman's move to cut into Elon Musk's turf is to get more data and to pivot OpenAI into surveillance.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
OpenAI's move into Musk's turf is part of an effort to pivot OpenAI into surveillance.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
Generative AI systems still hallucinate.
A piece of evidence came from the wrong dates.
2026-10-29
Held backBy usGary Marcus
Generative AI may never again be as popular as in the past year, but new, better techniques addressing its failings will come.
It didn’t pass every check.
2026-10-29
Held backBy usGary Marcus
Marcus says one finds over and over that LLMs are better at things they have encountered before than at things they haven't, and are especially poor when new items differ in important and subtle ways from old ones.
A piece of evidence came from the wrong dates.
2026-10-29
PublishedBy usGary Marcus
LLMs should either be abandoned or used only as components within larger systems that reason and plan better.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Generative AI's reliability failings can never be solved, given its inherent nature.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Generative AI has no killer app.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Marcus says that, for the most part, all the major tech companies are building the same thing (bigger LLMs on more data) and getting almost the same results (GPT-4-level models that barely differ and all struggle with hallucinations and boneheaded errors), with hardly any investment in anything else.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
AlphaProof, AlphaGeometry, ChatGPT, and why the future of AI is neurosymbolic
The final check of the write-up failed.
2026-10-29
PublishedBy usGary Marcus
The current generative AI approach has reached a plateau.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Marcus says that, as far as he can tell, hallucinations, boneheaded reasoning errors and 'discomprehensions' are an inevitable byproduct of LLMs, and that they certainly have not yet disappeared despite many promises to the contrary.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Generative AI, especially ChatGPT, will soon flame out.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Google DeepMind's work shows neural networks and symbolic systems can be profitably combined.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
No generative AI company has a competitive moat.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
AGI cannot be achieved without neurosymbolic AI.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
The generative AI bubble will begin to burst by July 28, 2025.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
Marcus says OpenAI has for the most part stuck to Hinton and LeCun's dogma, putting the far greater part of its efforts into scaling LLMs and trying hard to eschew symbols and symbolic rules, hiding them behind the curtain when they are used.
One AI thought this could be checked, so it needs a full check first.
2026-10-29
PublishedBy usGary Marcus
The current AI investment climate is hugely biased against genuine novelty.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
GenAI is vastly overrated, and its collapse may well lead to an AI winter of sorts.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
Marcus says LLMs literally do not reason through problems, and that in this riddle case ChatGPT instead retrieved an answer to a slightly different problem, which was wrong.
A reviewer said our test wasn’t fair to what was actually said.
2026-10-29
PublishedBy usGary Marcus
The impending generative AI collapse will not lead to the absolute death of AI.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
AlphaProof, AlphaGeometry, ChatGPT, and why the future of AI is neurosymbolic
A reviewer said our test wasn’t fair to what was actually said.
2026-10-29
PublishedBy usGary Marcus
AlphaProof and AlphaGeometry 2 rely on human coders translating Olympiad problems into formal form, a serious shortcoming.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
Generative AI systems still make boneheaded errors.
A piece of evidence came from the wrong dates.
2026-10-29
PublishedBy usGary Marcus
People are starting to realize generative AI has plateaued, lacks a killer app, still hallucinates and errs, and has no moat.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
AlphaProof, AlphaGeometry, ChatGPT, and why the future of AI is neurosymbolic
One of the AIs read the claim differently from our summary.
2026-10-29
PublishedBy usGary Marcus
Marcus says any reasoning or planning in LLMs is hit or miss: viable if a situation's particulars are close enough to the training set, but highly vulnerable if they are not.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
LLMs cannot sanity-check their own work.
The two reviewers reached different verdicts.
2026-10-29
Held backBy usGary Marcus
AlphaProof, AlphaGeometry, ChatGPT, and why the future of AI is neurosymbolic
The final check failed: our summary didn’t match their words.
2026-10-29
Held backBy usGary Marcus
Chatbots like ChatGPT have huge gaps in geometric understanding and reasoning.
A reviewer said our test wasn’t fair to what was actually said.
2026-10-29
PublishedBy usGary Marcus
Greater modularity will be a necessary part of future AI.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
It is disappointing that a top research team confused success on a prestigious competition with mastery of a foundational subject, and tiresome that the media overhype things as ever.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
AlphaGeometry2: Impressive accomplishment, but still a long path ahead
One of the AIs read the claim differently from our summary.
2026-10-29
PublishedBy usGary Marcus
DeepMind's 'gold-medalist performance' claim seems too strong to the authors, because preconverting each problem into formal notation means the system sidesteps understanding the English problem statements, which in a sense gives it an unfair start.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Combining the best general AIs with AlphaGeometry2 would still fall far short of a talented high school math student's geometric understanding and problem-solving.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
AlphaGeometry2 is a specialized system, nowhere close to a single general-purpose learning architecture.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
AlphaGeometry2: Impressive accomplishment, but still a long path ahead
One of the AIs read the claim differently from our summary.
2026-10-29
PublishedBy usGary Marcus
Benchmarks are often narrow, and strong benchmark results should not be taken as full conceptual understanding.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
AlphaGeometry does very well on IMO problems but a whole world of geometry lies outside its grasp.
A piece of evidence came from the wrong dates.
2026-10-29
PublishedBy usGary Marcus
The TechCrunch headline's claim about AlphaGeometry2 was not made by DeepMind and is not true.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Of the 50 IMO geometry problems from 2000 to 2024, a composite AI system starting from the English problem statements could solve at most 60%.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
The authors see no reason at all to call AlphaGeometry2's solutions 'superhuman'.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
The claim that AlphaGeometry2 'just figured out geometry' is false because it does not cover many basic aspects of geometry.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
AlphaGeometry 1 and 2 and AlphaProof are flagship examples of successful neuro-symbolic AI.
A piece of evidence came from the wrong dates.
2026-10-29
Held backBy usGary Marcus
AlphaGeometry2: Impressive accomplishment, but still a long path ahead
The final check failed: a cited source didn’t say what the write-up said.
2026-10-29
PublishedBy usGary Marcus
AlphaGeometry2 solving 42 of 50 IMO geometry problems is, without question, a very impressive accomplishment.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
AlphaGeometry2: Impressive accomplishment, but still a long path ahead
The two reviewers read the claim differently.
2026-10-29
PublishedBy usGary Marcus
OpenAI's reported effort to launch a social media network is presumably motivated in part by a wish to collect more private data.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
OpenAI's only financially viable path may be to move into surveillance, and it has already taken steps in that direction such as adding Paul Nakasone to its board and investing in a webcam company.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
People should speak up now against the growing AI-driven surveillance.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Two reports make clear that the era of Big Brother is about to enter the United States in massive new ways, not just via OpenAI but through multiple vectors.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
LLMs are here to stay and society is almost certain to be overrun by a tidal wave of misinformation.
A reviewer said our test wasn’t fair to what was actually said.
2026-10-29
PublishedBy usGary Marcus
If Stack Overflow is flooded with worthless generated code, programmers will leave, its database will become untrustworthy, and the site will die.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
If Stack Overflow is flooded with worthless ChatGPT-generated code examples and dies, then, because it is one of the most central resources the world's programmers rely on, the consequences for software quality and developer productivity could be immense.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Generative AI systems like ChatGPT, Dall-E 2, and Lensa can easily be automated to generate misinformation at unprecedented scale.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Because generative AI systems like ChatGPT, Dall-E 2, and Lensa cost almost nothing to operate, they are on a path to reducing the cost of generating disinformation to zero.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
ChatGPT is about as capable as Galactica of producing nonsense misinformation.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
Generative AI systems like ChatGPT, Dall-E 2, and Lensa are inherently unreliable, frequently making errors of both reasoning and fact, and are prone to hallucination.
A reviewer said our test wasn’t fair to what was actually said.