Changes, page 57 of 60

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

All changes (11,844)

Changes 11,201 to 11,400 of 11,844, newest first. Page 57 of 60.

2026-10-29
PublishedBy usGary Marcus
User accounts must be more strongly validated and anonymous bot-resistant authentication should become mandatory.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Every social media company and search engine should support and extend Stack Overflow's ban, misleading automatically-generated content should not be welcome, and regularly posting it should be grounds for a user's removal.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Countries may need to make misinformation legally actionable like libel when created with sufficient malice and volume.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Propaganda-producing nation-states and bad actors are likely to use LLMs to attack social media and create fake websites at unprecedented volume.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
LLMs lack mechanisms for verifying truth.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
We need to integrate LLMs with classical AI tools such as databases, knowledge graphs, and reasoning to fight misinformation.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Generative AI systems like chatGPT pose a real and imminent threat to the fabric of society.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Executives at Meta and OpenAI are as enthusiastic about their tools as the proprietors of Jurassic Park were about theirs.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
The price of a custom-trained large language model (currently under $500,000) will soon drop further.
A piece of evidence didn’t measure what the claim was about.
2026-10-29
Held backBy usGary Marcus
For propagandists, LLM hallucinations and unreliability are a virtue rather than an obstacle.
The check used the wrong kind of test for this claim.
2026-10-29
Held backBy usGary Marcus
LLMs are great at generating misinformation but poor at fighting it.
A piece of evidence came from the wrong dates.
2026-10-29
PublishedBy usGary Marcus
Misinformation-capable LLM technology can no longer be contained or withdrawn.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
LLMs make greatly escalated propaganda volume possible, and propagandists might succeed in creating a world where people cannot know what to trust.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
The developments described (SpaceX leasing compute to Google and Anthropic, and firms buying hardware before there was demand for it) are not particularly great for Nvidia, CoreWeave, Oracle, Nebius, Micron, and a variety of other hardware and semiconductor providers.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
If the US government takes stakes in them, Europe and Asia will not trust companies like OpenAI, and maybe not Google either.
A reviewer said our test wasn’t fair to what was actually said.
2026-10-29
Held backBy usGary Marcus
So far, AI has had less impact on real-world productivity than many people expected.
A reviewer said our test wasn’t fair to what was actually said.
2026-10-29
Held backBy usGary Marcus
AI’s Black Friday
One of the AIs read the claim differently from our summary.
2026-10-29
PublishedBy usGary Marcus
Meta may end up burning the most money on AI for the least results.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
AI’s Black Friday
The final check failed: the write-up said more than the evidence showed.
2026-10-29
PublishedBy usGary Marcus
SpaceX's AI division is leasing out large numbers of GPUs in 2026 because it cannot figure out what to do with them.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
If an American AI company is partly owned by the US government, nobody is going to trust it.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
The market ought to keep falling on Monday, June 8, 2026, though Marcus says he does not know whether it will.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
AI’s Black Friday
One of the AIs read the claim differently from our summary.
2026-10-29
Held backBy usGary Marcus
On Friday, June 5, 2026, the tech industry lost roughly half a trillion dollars of market value.
A reviewer didn’t accept one of the pieces of evidence.
2026-10-29
PublishedBy usGary Marcus
Bailing out OpenAI is crony socialism and corruption that prevents capitalism from taking its natural course.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Marcus doubts that Friday, June 5, 2026 was as bad as the AI-related market downturn will get.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Google might be wise not to accept Trump's offered government investment, because so much of its income is global.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
OpenAI may be so financially desperate that it has no choice but to accept government investment.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
The new cold war will be about AI.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
As of the post, we are not yet at the point of building machines that help us reason critically, overcome bias, reason more reliably than humans in general (not just in limited circumstances), and synthesize information far better, because our early 21st-century knowledge of AI is primitive, and there is a lot of confusion about this.
A piece of evidence came from the wrong dates.
2026-10-29
PublishedBy usGary Marcus
AI should aim to supplement humans where they are cognitively weak and take on dull, dirty, and dangerous work, not replicate humans.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
LLMs are good at some kinds of pattern matching but poor at floating point arithmetic.
A reviewer didn’t accept one of the pieces of evidence.
2026-10-29
Held backBy usGary Marcus
A system that cannot maintain cognitive models of the world cannot reason reliably, regardless of memory or compute.
The final check failed: the write-up said more than the evidence showed.
2026-10-29
Held backBy usGary Marcus
LLMs pretend to be what the Star Trek computer was, a system with encyclopedic knowledge, broad mathematical, synthetic and inferential abilities, and a hallucination-free natural language interface, which implies they are not that.
One AI thought this could be checked, so it needs a full check first.
2026-10-29
PublishedBy usGary Marcus
AGI should be a reliable, hallucination-free system that solves human problems rather than creating new ones.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
We can probably make machines better if we first spend time thinking about human minds, both their strengths and their weaknesses.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
AI versus The Clumsy Human Mind
The final check failed: a cited source didn’t say what the write-up said.
2026-10-29
Held backBy usGary Marcus
Achieving AGI almost certainly depends on getting disparate algorithms, such as symbolic and data-driven learning methods, to work together well.
The claim is hedged, so it isn’t scored.
2026-10-29
PublishedBy usGary Marcus
As of November 2023, neither humans nor machines are more intelligent overall; they have different abilities.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
Machines are far ahead of humans at floating point arithmetic but far behind on cognitive flexibility and long-term planning in unusual situations.
A piece of evidence came from the wrong dates.
2026-10-29
PublishedBy usGary Marcus
Eventually, we will be able to build machines that help us reason critically, overcome human bias, and reason objectively, giving just the facts without confabulation, the way calculators do.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
People should recognize, as soon as possible, that humans and silicon machines are importantly different.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
As of November 2023, AI is still pretty poor at maintaining world models, inferring semantics from language, comprehending scenes, navigating the 3D world, and cognitive flexibility, where humans are good.
A piece of evidence came from the wrong dates.
2026-10-29
PublishedBy usGary Marcus
AI should not aim to replicate the human mind in every detail.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
General-purpose household robots like Rosie the Robot are not doing our housework, and this is in part because the physical reasoning such a robot would need cannot be done purely by running simulations.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
In recent Optimus demonstrations, all or most of the required commonsense was apparently supplied by human teleoperators.
A reviewer said our test wasn’t fair to what was actually said.
2026-10-29
PublishedBy usGary Marcus
The hope that reliable commonsense will emerge from large text and video datasets has repeatedly proven unrealistic.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
AI agents, meaning AI systems that act in the world on users' behalf, are already here to a limited degree, even though general-purpose domestic robots are a long way away.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
OpenAI's o1 often makes very basic spatial mistakes because it does not understand how objects are embedded and interact in space.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Recent LLMs no longer make GPT-3's specific mistakes but still lack a robust understanding of the real external world.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
Sora's outputs are unreliable pure image streams, not segmentations of the world into objects with properties.
A reviewer didn’t accept one of the pieces of evidence.
2026-10-29
PublishedBy usGary Marcus
General-purpose AI will begin to mature only once researchers directly tackle commonsense reasoning about entities interacting in space and persisting in time.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Much of the commonsense reasoning problem the authors raised in 2015 and 2019 remains unsolved in AI as of January 2025.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Current AI approaches have consistently fallen short on reasoning about entities and their properties.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Sora, now publicly available, struggles with physical reasoning.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Current LLM agents (Claude, Gemini, GPT-4 etc.) are not close to ready for prime time on real-world workplace tasks.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
AI still lacks “common” sense, 70 years later
One of the AIs read the claim differently from our summary.
2026-10-29
Held backBy usGary Marcus
AI agents will require common sense about objects, people, and entities to work reliably in the real world.
A reviewer said our test wasn’t fair to what was actually said.
2026-10-29
PublishedBy usGary Marcus
None of the three popular approaches to commonsense reasoning (simulation, LLM emergence, video generation) has proved fully satisfactory.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Physical simulation has proven much less effective for AI commonsense reasoning than for scientific, engineering, and entertainment applications.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
LLM commonsense failure examples keep appearing because no robust solution has been found, even though individual examples can be fixed.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
AI agents will be a major focus of attention in the AI field in 2025.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
AI agents will not work well until major progress is made in commonsense reasoning, which may require wholesale revisions to how AI is approached.
The final check failed: the write-up said more than the evidence showed.
2026-10-29
PublishedBy usGary Marcus
Sora's videos are fundamentally more like visual collages than game-engine renderings, which is why they sometimes have a surreal or magic-trick quality.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
There is a serious need to have independent scientists involved in decisions about managing AI risk.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
AI poses serious risk in both the short term and the long term, a view Marcus says he and Hinton now share.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
LLM-enhanced terrorism, perhaps coupled with mass AI-generated propaganda, could lead to nuclear war or to the deliberate spread of pathogens worse than covid-19, and many, many people could die and civilization could be utterly disrupted.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
Because LLMs are now widely known, interesting to criminals, and increasingly given external access, they can do more damage than when they were lab curiosities.
The claim is hedged, so it isn’t scored.
2026-10-29
Held backBy usGary Marcus
AI risk ≠ AGI risk
One of the AIs read the claim differently from our summary.
2026-10-29
Held backBy usGary Marcus
AI risk ≠ AGI risk
One of the AIs read the claim differently from our summary.
2026-10-29
PublishedBy usGary Marcus
AI does not need to be superintelligent to create serious problems.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
We do not have many controls over current AI, especially now that it can be connected to TaskRabbit and real-world software APIs.
A piece of evidence came from the wrong dates.
2026-10-29
PublishedBy usGary Marcus
Marcus does not see computers rapidly and radically improving themselves as an immediate possibility.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
AI risk ≠ AGI risk
One of the AIs read the claim differently from our summary.
2026-10-29
Held backBy usGary Marcus
AI risk ≠ AGI risk
One of the AIs read the claim differently from our summary.
2026-10-29
PublishedBy usGary Marcus
We should treat LLMs as a dress rehearsal for future AI, which may be even harder to control.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
We should think much more about criminal and terrorist misuse of LLMs and how to stop it, not just about Skynet scenarios.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Current AI already poses enormous risks we are ill-prepared for, and future AI could well make things worse.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
AI risk ≠ AGI risk
One of the AIs read the claim differently from our summary.
2026-10-29
PublishedBy usGary Marcus
We need to get serious about AI risk now, given rapid widespread deployment and unclear use cases and mitigations.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Adept.AI just raised $350 million to let large language models access pretty much everything, despite LLMs' clear tendencies toward hallucination and unreliability.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
We have let too much slide and done too little to face the risks of AI, which is Marcus's reason for signing the global call for AI red lines.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
The long-feared moment of generative AI influencing a democratic election may have arrived, through an AI-generated video in Argentina's May 2025 election.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
A lot of lawyers are using tools like ChatGPT to prepare briefs, and many of them still seem shocked when it makes up cases.
A reviewer said our test wasn’t fair to what was actually said.
2026-10-29
PublishedBy usGary Marcus
If educated people like lawyers still haven't gotten the message that LLMs hallucinate, and that they can be fined or publicly humiliated if they rely on LLMs, something is fundamentally wrong.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
LLM hallucinations persist and will not go away anytime soon.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Publicized cases of AI-fabricated legal citations are probably a tiny fraction of the total incidence.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Those in the media who have downplayed hallucinations and oversold current AI have done the public a disservice.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
The media should do more to alert people to the shortcomings of LLMs, including how regularly hallucinations happen.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Charlotin's database lists 112 cases, an average of a bit less than one publicly reported case a week since ChatGPT became popular, and if May is representative, the problem is getting worse.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
LLMs cannot sanity-check their own work, stick purely to known sources, or flag when they have invented something.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
LLM hallucinated case citations in legal filings are a routine occurrence, not an anomaly.
A reviewer didn’t accept one of the pieces of evidence.
2026-10-29
PublishedBy usGary Marcus
Hallucination is a deep, serious, unsolved problem with generative AI.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
AI literacy, hallucinations, and the law: A case study
The final check failed: a cited source didn’t say what the write-up said.
2026-10-29
PublishedBy usGary Marcus
AI hypesters are currently winning the public narrative.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
AI, layoffs, productivity and The Klarna Effect
A reviewer said our test wasn’t fair to what was actually said.
2026-10-29
PublishedBy usGary Marcus
Stopping hiring of junior employees in favor of AI risks destroying the pipeline to expertise.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
Marcus believes that, all too often, employers experimenting with replacing employees with AI are underestimating what humans can do and overestimating the technological readiness of current AI, which can often do large portions of a job but not the whole thing.
The final check failed: the write-up said more than the evidence showed.
2026-10-29
Held backBy usGary Marcus
AI can do some but often not all junior employee work and often lacks senior employees' judgement.
The final check failed: the write-up said more than the evidence showed.
2026-10-29
PublishedBy usGary Marcus
Employers replacing employees with AI may end up reversing course as Klarna did.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
Despite Hinton's 2016 prediction, to Marcus's knowledge no radiologist has been replaced by AI.
A piece of evidence came from the wrong dates.
2026-10-29
Held backBy usGary Marcus
AI, layoffs, productivity and The Klarna Effect
One of the AIs read the claim differently from our summary.
2026-10-29
Held backBy usGary Marcus
Klarna was among the first to announce major AI layoffs and among the first to realize that was a mistake.
The two reviewers reached different verdicts.
2026-10-29
Held backBy usGary Marcus
Only 25% of surveyed executives thought Microsoft Copilot was worth it.
The final check failed: the write-up said more than the evidence showed.
2026-10-29
Held backBy usGary Marcus
The 79% Copilot trial rate among surveyed executives refutes the narrative that people are ignoring AI.
A reviewer said our test wasn’t fair to what was actually said.
2026-10-29
Held backBy usGary Marcus
Practically everyone has tried AI, but users are not always satisfied with the results.
A reviewer didn’t accept one of the pieces of evidence.
2026-10-29
PublishedBy usGary Marcus
People are not ignoring generative AI; they are waiting to see whether it will work.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
He thinks it is fair to say that, slightly over a decade later, the 2014 Turing Test 'winner' Eugene Goostman contributed nothing of lasting value to AI.
One AI thought this could be checked, so it needs a full check first.
2026-10-29
PublishedBy usGary Marcus
Nobody has fully implemented Marcus's proposed comprehension challenge test.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Few people in the AI community take the Turing Test seriously, and few should.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
To do well in the new study, the system had to be coached to adopt the persona of a young, introverted, slang-using person knowledgeable about internet culture, so its success is still in no small part about tricks.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
If a careful comprehension challenge test were built, no current AI system would come close to human performance on it.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
AI has (sort of) passed the Turing Test; here’s why that hardly matters
The final check failed: a cited source didn’t say what the write-up said, and our summary didn’t match their words.
2026-10-29
Held backBy usGary Marcus
AI has (sort of) passed the Turing Test; here’s why that hardly matters
One of the AIs read the claim differently from our summary.
2026-10-29
Held backBy usGary Marcus
Those who could afford to build the comprehension challenge prefer to spend research money on scaling.
The two reviewers reached different verdicts.
2026-10-29
Held backBy usGary Marcus
Systems like Gemini can do the comprehension challenge a little but often hallucinate heavily, and he does not think they can yet take in a full Simpsons episode, let alone follow four seasons of Succession and explain in real time who is lying to whom and to what end.
A reviewer said our test wasn’t fair to what was actually said.
2026-10-29
Held backBy usGary Marcus
In the new study, ChatGPT without a persona prompt was beaten by ELIZA in the Turing Test.
A reviewer said our test wasn’t fair to what was actually said.
2026-10-29
PublishedBy usGary Marcus
None of the Turing Test chatbots, from ELIZA to ChatGPT 4.5, are anywhere near as intelligent as naive audiences might think.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Fooling people for a few minutes is not the measure of intelligence, and never was.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
AI has made great progress in human mimicry since 2014, but current systems still lack much in reasoning, planning, and common sense.
A reviewer didn’t accept one of the pieces of evidence.
2026-10-29
PublishedBy usGary Marcus
The Turing Test is a test of human gullibility, not a test of intelligence.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Today's Turing Test-taking AI systems are better than the earlier generation, but that does not mean anything is 'over'.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
AI has (sort of) passed the Turing Test; here’s why that hardly matters
One of the AIs read the claim differently from our summary.
2026-10-29
Held backBy usGary Marcus
AI has (sort of) passed the Turing Test; here’s why that hardly matters
One of the AIs read the claim differently from our summary.
2026-10-29
PublishedBy usGary Marcus
In a faceoff, Eugene Goostman might actually beat ChatGPT.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
AI has become Sauron’s Ring
One of the AIs read the claim differently from our summary.
2026-10-29
PublishedBy usGary Marcus
AI has gone down a path of enabling and accelerating surveillance capitalism, misinformation, nonconsensual deepfake porn and IP theft, which was not inevitable.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
The science of AI has been corrupted by money.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
AI development should go back to being genuinely in the public interest.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Dario Amodei's claim that nearly all code would be written by AI within the next year was made without evidence.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Coders and software architects are not on their way out due to AI.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
AI Coding Fantasy meets Pac-Man
One AI wasn’t sure who said it.
2026-10-29
PublishedBy usGary Marcus
Marcus thinks that most or all of the points he made last week, about the challenges of building and debugging novel systems with sufficient generality, will stand the test of time.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
AI coding tools can help a coder learn a new API, and maybe even a new programming language.
A reviewer said our test wasn’t fair to what was actually said.
2026-10-29
Held backBy usGary Marcus
AI Coding Fantasy meets Pac-Man
The final check failed: a cited source didn’t say what the write-up said.
2026-10-29
Held backBy usGary Marcus
In some instances, beginners' overdependence on AI coding tools will produce large amounts of hard-to-maintain spaghetti code and create security problems.
The check used the wrong kind of test for this claim.
2026-10-29
PublishedBy usGary Marcus
Current AI coding tools do not resolve the challenges of maintaining complex code bases over time.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
AI coding tools, acting as a fantastic, souped-up version of autocomplete, somewhat increase productivity.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
AI will serve as a tool to help people write code but will not soon replace people who know how to conceive of, write, and debug code.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Most or all trained coders will soon use AI as part of their workflow, if they are not already.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Marcus guesses that current AI coding tools would struggle to use 6502 assembly well, for multiple reasons including a lack of high-level libraries, fewer Stack Overflow queries to rely on, and a greater need to understand high-level architectural concerns that might be side-stepped in higher-level languages.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
In The Guardian's test, people using AI tools without coding knowledge did not succeed well at building a Pac-Man clone.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
If his analysis is correct, LLM dominance may soon begin to decline.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
If his analysis is correct, newer, sounder alternatives to LLMs may soon start to emerge.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
LLMs are not like humans and never will be.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Generalization is, and always has been, the vital question in AI.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
AI desperately needs world models.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Neurosymbolic AI will eventually displace pure LLMs.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
LLMs are unable to stop hallucinating.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
If verified-yet-anonymous proof-of-human is required for high-reach interactions, it will make it mathematically difficult and financially ruinous for one operator to secretly run ten thousand accounts.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Venture capital is already helping industrialize astroturfing; for example, Andreessen Horowitz-backed Doublespeed advertises a way to orchestrate actions on thousands of social accounts and to mimic natural user interaction on physical devices so the activity appears human.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Policymakers should incentivize cryptographic attestations and reputation standards to make fake personas expensive.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
AI bot swarms threaten to undermine democracy
The final check failed: the write-up said more than the evidence showed.
2026-10-29
Held backBy usGary Marcus
If swarms flood the web with duplicative, crawler-targeted content, they can execute 'LLM grooming', poisoning the training data that future AI models (and citizens) rely on.
The claim is hedged, so it isn’t scored.
2026-10-29
PublishedBy usGary Marcus
Legislation must guarantee vetted academic and civil society researchers free, privacy-preserving access to platform data.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
Because people rely on social evidence, fabricated bot swarms can make fringe views appear to be majority opinions.
The claim is hedged, so it isn’t scored.
2026-10-29
PublishedBy usGary Marcus
Even so-called 'thinking' (reasoning) AI models are vulnerable to training-data poisoning by swarms.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
AI bot swarms can infiltrate communities, build credibility, and adapt in real time to discover what persuades.
The claim is hedged, so it isn’t scored.
2026-10-29
Held backBy usGary Marcus
Because AI can generate unique text per post, detecting bots by copy-pasted content no longer works.
The two reviewers reached different verdicts.
2026-10-29
Held backBy usGary Marcus
Swarms of AI bots acting together in concert will emerge as a threat worse than earlier disinformation bots.
The final check failed: the write-up said more than the evidence showed, and a condition they stated was left out.
2026-10-29
PublishedBy usGary Marcus
An independent, non-governmental AI Influence Observatory of academic groups and NGOs should be created to detect coordinated swarms.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Purely reactive defenses will fail; defenders should proactively stress-test using agent-based simulations of AI swarms.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
AI bot swarms threaten to undermine democracy
The two reviewers read the claim differently.
2026-10-29
PublishedBy usGary Marcus
Social media platforms must abandon episodic whack-a-mole takedowns in favor of continuous monitoring for improbable coordination.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Emerging AI bot swarms can coordinate fleets of synthetic personas that are hard to distinguish from real communities.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
When it costs pennies to coordinate a fake mob and takes moments to counterfeit a human identity, the public square is left wide open to attack.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
As of January 2026, bad actors can launch massive bot swarms cheaply and with little risk.
The two reviewers reached different verdicts.
2026-10-29
Held backBy usGary Marcus
AI bot swarms threaten to undermine democracy
One of the AIs read the claim differently from our summary.
2026-10-29
Held backBy usGary Marcus
AI as Reality Television
One AI wasn’t sure who said it.
2026-10-29
PublishedBy usGary Marcus
Negative press coverage of Sam Altman and OpenAI, once seen as able to do no wrong, keeps coming.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Gary Marcus wishes that AI these days were more about research than about money, politics, and ego.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
AI as Reality Television
The two reviewers read the claim differently.
2026-10-29
Held backBy usGary Marcus
As of June 2024 the field has found no real solution to hallucinations, unreliability, poor reasoning or shaky alignment in LLMs.
A reviewer declined to make a call.
2026-10-29
PublishedBy usGary Marcus
OpenAI's ChatGPT task-scheduling agent will not work reliably.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
As with generative AI, with AI agents some prompts will work some of the time and some of the time they will give strange results, and it will be difficult to know in advance what to expect.
It didn’t pass every check.
2026-10-29
PublishedBy usGary Marcus
Eventually, though not soon, AI agents will do a huge fraction of human knowledge work, perhaps physical labor too, and be worth trillions.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
AI agents will not do a huge fraction of human knowledge work in 2025, 2026, or 2027, except in narrow use cases.
The final check of the write-up failed.
2026-10-29
PublishedBy usGary Marcus
AI agents probably will not do a huge fraction of human knowledge work this decade, except in narrow use cases.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
Throughout 2025 there will be many more reports of AI agents being brittle and failing.
The two reviewers reached different verdicts.
2026-10-29
Held backBy usGary Marcus
If what you want is agents with common sense, sound reasoning, tight grounding in reality, reliable instruction-following and enough theory of mind to infer intentions, generative AI alone is not the ticket, because those are exactly the sticking points that have stymied it over and over again.
A reviewer said our test wasn’t fair to what was actually said.
2026-10-29
Held backBy usGary Marcus
Reliable AI agents cannot be built unless the hallucination problem is solved.
A reviewer said our test wasn’t fair to what was actually said.
2026-10-29
PublishedBy usGary Marcus
As of January 2025 there is no principled, trustworthy solution to the hallucination problem.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
In 2025, AI agents will exist only as demos.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
You can't build agents unless you have common sense, sound reasoning, a tight grounding in reality, and systems that reliably do what you ask of them, with enough theory of mind to infer your intentions.
The final check failed: the write-up said more than the evidence showed.
2026-10-29
PublishedBy usGary Marcus
In the long term, everyone will have their own AI agents and companies will have armies of them.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
The AI agent demos of 2025 will not work very reliably.
The final check of the write-up failed.
2026-10-29
PublishedBy usGary Marcus
Someday, current agents' failure to produce much of lasting value will be seen as a poignant reflection of a massive intellectual and economic mistake: investing almost solely in LLMs as a kind of hopeful shortcut.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
AI agents in 2025 have largely not delivered the boundary-pushing capabilities that Kevin Roose promised.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
AI Agents have, so far, mostly been a dud
One AI wasn’t sure who said it.
2026-10-29
PublishedBy usGary Marcus
Yet more good money, in staggeringly large amounts, is being poured after bad into generative AI.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
He suspects it will be typical that AI agent errors compound over time, as in Penrose's account-balance benchmark.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
Marcus says he just doesn't see how AI agents can work out without neurosymbolic AI that is more deeply integrated in systems as a whole, with rich world models as a central component.
One AI thought this could be checked, so it needs a full check first.
2026-10-29
PublishedBy usGary Marcus
The gap between AI agent demos and real-world performance will probably persist for years to come.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
Reliable agents require robust, trustworthy AI, and LLMs are not getting us there.
A reviewer said our test wasn’t fair to what was actually said.
2026-10-29
Held backBy usGary Marcus
Pure scaling is not getting us to AGI, and returns to scaling are diminishing.
The two reviewers reached different verdicts.
2026-10-29
Held backBy usGary Marcus
AI Agents have, so far, mostly been a dud
The final check failed: the write-up said more than the evidence showed.
2026-10-29
PublishedBy usGary Marcus
Because LLMs are fundamentally unreliable, multi-step agentic tasks accumulate errors that sometimes lead to catastrophe.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
AI agents should not be trusted with your calendar, let alone your credit cards.
Published after both AI reviews and the final check passed.
2026-10-29
PublishedBy usGary Marcus
Perhaps, after enough failures of AI agents, reality may finally start to settle in.
Published after both AI reviews and the final check passed.
2026-10-29
Held backBy usGary Marcus
Neurosymbolic AI and other alternatives are wildly underfunded, receiving perhaps 1% or less of total AI investment.
A reviewer said our test wasn’t fair to what was actually said.