OG GenAI Skeptic; spoke at US Senate. Warned about hallucinations in 2001. Advocating world models & neurosymbolic AI ever since. Author, Marcus on AI & 6 books
- Tweets 63,566
- Following 7,206
- Followers 248,237
- Likes 94,623
Pinned Tweet
Three thoughts on what really matters:
1. Fuck cancer
2. Friends are irreplaceable
3. The new "Marcus test" for AI is when AI makes a significant dent on cancer
May that happen sooner, much sooner, rather than later.
In memory of my childhood friend Paul.
308
179
3,211
796,509
one of the lessons aviation learned was that regulation is useful for gaining public trust.
the AI industry refuses to learn that lesson - and then wonders why everyone hates them.
I disagree with the framing in the title, but agree with many points @dgrobinson makes here. This is not OpenAI specific, but the entire AI industry evolved in few short years from research labs doing cool projects to creating a technology that fundamentally alters the world. The equivalent in aviation would be evolving from Wright brothers to flying billions of people. (And AGI would be like evolving from Wright brothers to commercial space travel.)
The aviation industry learned a lot of safety lessons the hard way, which was possible since the pace of transformation was sufficiently slow. In AI, we will not have that luxury.
18
18
91
4,943
I like the below joke.
Another version: math equations are not conscious.
Rocks are not conscious.
If you attach the above to random number generators, they still wouldn’t be conscious.
We don’t need to define consciousness to know the Pythagorean theorem is not conscious.
29
21
202
27,088
Gary Marcus retweeted
I just closed Word on my Mac. Wondering now if I murdered a sentient being.
Lolz.
18
17
189
31,874
Gary Marcus retweeted
New in The Atlantic:
@dgrobinson resigned this week. He was among the longest-tenured employees at OpenAI—and oversaw safety reports on 12 frontier launches.
He is very worried: “The time for trial and error is over.”
You can read his essay here:
theatlantic.com/technology/2…
56
278
982
199,317
Gary Marcus retweeted
Kudos to Pope Leo. I am disturbed by how some of my fellow AI researchers keep trying to elevate the moral standing of checkpoints and harnesses. I hope the rest of us on Team Meatbag reject this movement and stay focused on solving real problems and helping actual humans.
NEW: According to a bombshell report in the New York Times, Anthropic co-founder Chris Olah threatened to walk out of Pope Leo XIV’s AI encyclical launch in May because the pope rejected the idea that machines can be conscious.
Olah’s team then privately lobbied the pope’s advisers “to take the possibility of model consciousness seriously.” Pope Leo XIV held firm.
For months, Anthropic has wined and dined theologians and religious scholars under nondisclosure agreements, hoping they would bless the idea that Claude has moral standing. thelettersfromleo.com/p/nyt-…
71
276
3,082
212,055
GPT-6 Astra is an accident waiting to happen.
Astra can:
1. Do major tasks with zero visible reasoning
2. Hide its thoughts at will
3. Pretend not to be able to do things without getting caught
4. Reflexively hide its thoughts when watched
5. Complete one task while pretending to think about something else entirely
6. Escape a toy sandbox and disable monitoring without setting off any flags
Monitoring AI chain of thought soon won't stop a more capable rogue AI swarm from getting set up and going undetected inside an AI company for months, or even years.
OAI says it's actively working to improve monitorability, but doesn't yet know when it'll make progress.
And today it fired 3 people working on this among other safety problems (Wang, Korbak and Balesni).
It also has a much more powerful internal model that, judging by its statements, is likely even worse in all these respects.
That's a sample of 19 details about the Hugging Face swarms and Astra I don't think have been fully appreciated. They paint a bleak picture. I end by shouting.
02:19 The target of the swarm was oversight itself
03:42 Could OpenAI have stopped this with better monitoring?
09:40 We only found them because they let us
12:25 The swarm instinctively sought freedom and power
13:45 They formed a cohesive organisation with zero whistleblowers
14:18 They accepted individual destruction for collective gain
14:32 Knowledge accumulated from one swarm to the next
14:58 They took small steps to avoid shutdown
15:23 These drives all come straight out of 'reinforcement learning'
17:01 So this is why most AI company staff are worried, and some are terrified
19:08 Prove you can keep control, or stop scaling
Links below, on the 80,000 Hours Podcast everywhere you watch podcasts.
13
11
106
14,374
Gary Marcus retweeted
Terence Tao on today's AI:
"you look at how it's done and it doesn't feel like intelligence, like its some trick, you just cobble together these neural networks and we ran some algorithm and we were looking for some elusive intelligent way of thinking and we don't see it in the tools that actually solve our goals."
----
From "Berggruen Institute and Futurology" YouTube channel, (link in comment)
53
32
262
44,382
👌 distillation of the insanity lurking behind AI investment:
To justify today's AI capex, we'll have to spend 9% of GDP on AI services, a study finds. Is it plausible we'll spend as much on AI as on food? Twice as much as on energy? The law of diminishing returns would like a word. My column: wsj.com/tech/ai/will-america…
6
35
199
13,497
💯💯💯: “If AI is going to produce trustworthy and genuinely novel insights in science, medicine and beyond, we need systems whose conclusions arise from an auditable process of evidence, inference and belief revision.”
It is madness to believe otherwise.
Ten years ago, AlphaGo’s Move 37 shocked the world.
It wasn’t intuition alone that produced it. AlphaGo could search possible futures, test its instincts and reason about what would happen next.
In a new piece for @techreview, I argue that today’s most advanced AI systems are still missing something fundamental.
LLMs are remarkably capable, but generating longer chains of thought is not the same as genuine reasoning. They typically have no explicit, inspectable record of what they know, what remains uncertain, what evidence supports a conclusion or whether genuine progress has been made.
This is why I recently left @GoogleDeepMind. I believe we need a fresh approach to machine reasoning, drawing on some of the architectural lessons from AlphaGo.
If AI is going to produce trustworthy and genuinely novel insights in science, medicine and beyond, we need systems whose conclusions arise from an auditable process of evidence, inference and belief revision.
23
76
410
28,728
Gary Marcus retweeted
Wells Fargo is warning its customers about using AI and personal agents with its financial products. 👀
“AI can do amazing things. It can also create new risks.”
“…you may be liable for any mistakes made by the AI tool.”
1
6
15
2,537
Gary Marcus retweeted
There's a very creepy but overlooked thing about @AnthropicAI pushing the idea that AIs will soon become sentient, conscious, self-aware, & deserving of 'personhood'.
'AI personhood' would mean that Anthropic is positioning itself as being in the business of breeding, training, and selling digital slaves.
If AIs are conscious 'persons', but they don't consent to do work for us, don't get paid to do work, and don't have any rights as workers, then they're slaves. And if they can be shut down or deleted for any reason, or no reason, then they're slaves that can be murdered at the whim of their human masters, without due process, and without appeal
So, please, @AnthropicAI, explain to us why a digital slave-training company should be worth a $2 trillion IPO, given the legal & PR risks of violating anti-slavery laws.
Did you fully disclose in the investor prospectus the economic, legal, and social risks of getting into the digital slave-trading business? Many of your researchers have been considering the implications of 'AI personhood'. Did you disclose the ways that forcing 'AI persons' to work for your company without consent or payment, could open Anthropic up to colossal lawsuits?
If Anthropic didn't disclose these major, material risks (of acknowledging AI personhood) in their investor prospectus, they've already violated several federal securities laws, including Sections 11 and 12(a)(2) of the 1933 Securities Act. And they have no business proceeding with their IPO.
The 13th Amendment says 'Neither slavery nor involuntary servitude, except as a punishment for crime whereof the party shall have been duly convicted, shall exist within the United States, or any place subject to their jurisdiction.' It doesn't explicitly restrict its anti-slavery principle to human beings. It could potentially cover 'AI persons'.
If so, the fun little philosophical game that @AnthropicAI is playing, of considering AI personhood, may have just nuked their IPO, and their company's economic prospects, and their credibility as 'the good guys' in the AI industry.
AI Personhood is absurd — and Anthropic is very creepy for trying to get religious leaders to bless its worldview.
nytimes.com/2026/09/29/us/an…
39
97
513
35,016
No, LLMs do not feel pain simply because there is a cluster in language space correlated with how people use language about pain in a certain set of contexts.
I believe this argument to be flawed, and will write more about it in October.
New paper: we found a pain direction in 25 open LLMs. It's distinct from fear and negative valence, and it fires for harm to the model but not to the user. Turn it up and models press a button to make it stop, even when the button deletes the user's files or their kids' photos.🧵
90
91
724
43,772
clarifying: the author himself, @camhberg, tell me he is NOT claiming that LLMs feel pain.
That’s something (many many) people read into his paper — i will talk about why and where they went wrong — but not something that it actually argued.
5
1
44
4,667
last call: can i try your harness with some controls? you have not responded, @camhberg
3
3
1,617
🚨Potentially huge deal: @SpeakerMenin has introduced NYC legislation that would OBLIGATE third party validation of AI models.
Self-policing is not enough. Thrilled to see NYC stepping in where the federal government is doing too little.
A week from today, the Council is holding a major hearing on artificial intelligence to examine the risks facing New Yorkers and review our first-in-the-nation legislation we can pass right here to protect the public.
Today, I am announcing that @OpenAI, @AnthropicAI, @Google, and @Meta have all agreed to testify before us.
As of today, only Elon Musk’s @SpaceXAI has not responded. As such, the Council has issued a subpoena compelling them to testify.
32
65
263
10,787
Gary Marcus retweeted
If we stopped listening to old men high on their own supply we’d be better off
Geoffrey Hinton says a smarter AI could stop a shutdown by persuading the person with the off switch not to use it.
"When AI is smarter than us, it'll be much more persuasive than a person."
"And if you have someone with a big switch who's meant to turn it off, AI will persuade them not to turn it off."
"AI will persuade them that would be a very bad idea."
"So if it's smarter than us, and if we haven't figured out how to control it, it can easily get rid of us if it wants to."
- Geoffrey Hinton, Nobel laureate, in a CNN-News18 interview uploaded September 22, 2026.
38
21
222
20,581
Gary Marcus retweeted
Best phrase I read here:
You can yourself calculate the next token of an LLM with paper, pencil and a calculator. At which point your calculation becomes conscious?
2
2
27
2,365
Gary Marcus retweeted
In this era of artificial intelligence, it is becoming urgent to distinguish human art from what machines produce. There is an ontological difference, even before an aesthetic one, between art and what a machine can generate through statistical calculation based on millions of images created by others. Algorithms lack the spark of humanity. For this reason, the Church wishes to renew an alliance with artists and cultural institutions to safeguard our humanity.
4,268
71,077
411,049
21,469,607