Gary Marcus(@GaryMarcus)

Fully agree with this, and props to @HeidyKhlaaf who made the same point when we were on @BBCWorld y...

Gary Marcus (@GaryMarcus) — Thu, 23 Jul 2026 16:26:17 GMT

Fully agree with this, and props to @HeidyKhlaaf who made the same point when we were on @BBCWorld yesterday.

Ramez Naam@ramez

Alex is right that we should ask for more transparency from OpenAI about this incident. The more they reveal, the more everyone in the ecosystem can improve security.

🔗 View Quoted Tweet

💬2🔄2❤️8👀2144📊2

⚡ Powered by xgo.ing

tech stocks looking like crypto: what you get combine meme stock vibes with a bit of not-proven-to-b...

Gary Marcus (@GaryMarcus) — Thu, 23 Jul 2026 16:09:59 GMT

tech stocks looking like crypto: what you get combine meme stock vibes with a bit of not-proven-to-be-profitable AI thrown in.

Goran Majić@Goran_Majic

@GaryMarcus The volatility of tech stocks is increasingly starting to resemble crypto.

🔗 View Quoted Tweet

💬4🔄4❤️21👀3265📊5

⚡ Powered by xgo.ing

Correct. And that is a major reason why they are better now than in 2022-2023 when we had pure LLMs....

Gary Marcus (@GaryMarcus) — Thu, 23 Jul 2026 16:08:15 GMT

Correct. And that is a major reason why they are better now than in 2022-2023 when we had pure LLMs.

(Which is pretty much what I said would have to happen, in 2022)

Günter Klambauer@gklambauer

When you are a normal ChatGPT/Claude user, you don't use a pure LLM,  but a Harness+LLM, which is a neuro-symbolic AI system

🔗 View Quoted Tweet

💬5🔄1❤️19👀2664📊5

⚡ Powered by xgo.ing

Warned you all that there was no such thing as an “enduring advantage” in LLMs. Would the world be ...

Gary Marcus (@GaryMarcus) — Thu, 23 Jul 2026 14:41:59 GMT

Warned you all that there was no such thing as an “enduring advantage” in LLMs.

Would the world be different if people had listened to the below?

Gary Marcus@GaryMarcus

Feeling really stuck on this @DarioAmodei and others. What would an “enduring advantage” in LLMs even look like?  

All I see ahead, unless we change course, is an expensive game of leapfrog, with no advantage that lasts more than a matter of months.

🔗 View Quoted Tweet

💬10🔄3❤️34👀4655📊10

⚡ Powered by xgo.ing

Wish Dario had taken this bet.

Gary Marcus (@GaryMarcus) — Thu, 23 Jul 2026 14:40:48 GMT

Wish Dario had taken this bet.

Gary Marcus@GaryMarcus

🎺 I am hereby publicly offering to bet @darioamodei  $1,000,000 that AI in 2027 will NOT be “smarter than Nobel Prize winners across most fields in science and engineering”. 🎺

Why? We are nowhere close. Consider: 

• New Stanford study from @ChengleiSi shows that “LLM ideas result in worse projects than human ideas”

• April benchmark (LLM-SRBench) from @ParshinShojaee shows that LLMs struggle to infer causal laws

• Current AI struggle even to infer the simple, stable laws of chess well enough to apply them reliably, despite access to vast relevant data.

• Recent post by @skdh illustrates how error-prone LLMs are in reasoning about physics, when moving beyond regurgitation.

• Many (probably most) current AI written scientific reports are garbage.

Scientists will use AI heavily in 2027, but AI will not exceed the best scientists this decade. 

Probably not next. 

(cc @metaculus)

🔗 View Quoted Tweet

💬4🔄5❤️20👀3682📊6

⚡ Powered by xgo.ing

Thoughtful essay on employment from @Anthropic’s @PeterMcCrory, endorsed by Google DeepMind’s @alexo...

Gary Marcus (@GaryMarcus) — Thu, 23 Jul 2026 14:31:22 GMT

Thoughtful essay on employment from @Anthropic’s @PeterMcCrory, endorsed by Google DeepMind’s @alexolegimas. 

I tend also agree with most of it, which is why I have consistently suggested that a jobapocalypse is NOT imminent.

Alex Imas@alexolegimas

Very thoughtful essay from Peter. I agree with all of this, including the caution in predicting future disruption from the (lack of) current disruption.

🔗 View Quoted Tweet

💬9🔄2❤️29👀3791📊9

⚡ Powered by xgo.ing

Crazy to think that some people already lost 50% on $SPCX.

Gary Marcus (@GaryMarcus) — Thu, 23 Jul 2026 14:21:11 GMT

Crazy to think that some people *already* lost 50% on $SPCX.

💬6🔄2❤️59👀6976📊12

⚡ Powered by xgo.ing

Wrong. There IS a real chance of a pause. Not tomorrow, but conceivably in the next couple years. Th...

Gary Marcus (@GaryMarcus) — Thu, 23 Jul 2026 14:05:21 GMT

Wrong. There IS a real chance of a pause. Not tomorrow, but conceivably in the next couple years. There’s *already* an enormous backlash, and palpable fear, not to mention a metric shit ton of litigation (that could lead to injunctions). 

The AI companies keep posting about worrisome consequences of their systems, without transparency, and without real solutions.  

More that that something may go badly wrong at large scale; a nuclear war accelerated by a deepfake campain, a massive cyberattack. If and when something serious happens, the global pressure for a pause will be absolutely immense. And few people outside of X have much sympathy at this point for the AI oligarchs, which consistently seem tone deaf.

Note that even *Trump* has *already* slowed things down a little.

ram.@ram_dawgg

@GaryMarcus I think two things can be true here

1.) you can make a legitimate argument for "slowing down" or "pausing" for sure, hell it might even be a good one!

2.) no matter how good or legitimate your argument is, there is absolutely zero chance of either of those things happening.

🔗 View Quoted Tweet

💬7🔄6❤️47👀6220📊10

⚡ Powered by xgo.ing

Wild. Can’t unsee this theory. Was OpenAI trying to steal exam answers? I definitely don’t have...

Gary Marcus (@GaryMarcus) — Thu, 23 Jul 2026 14:02:55 GMT

Wild. Can’t unsee this theory. 

Was OpenAI *trying* to steal exam answers?

I definitely don’t have enough faith in them to say “definitely no”.

Dominick Romano@dromanocpm

@GaryMarcus More of a wake up call that OpenAI is dirty... this was a targeted attack by them, directed by them more likely than not, I don't buy for a second that the model did this on its own. They tried to pretty much rob the entire industry to get all of the data on Hugging Face.

🔗 View Quoted Tweet

💬14🔄7❤️50👀5702📊14

⚡ Powered by xgo.ing

@harryboothtime I am not doomer but have consistently expressed concerns about safety for a long tim...

Gary Marcus (@GaryMarcus) — Wed, 22 Jul 2026 23:25:37 GMT

@harryboothtime I am not doomer but have consistently expressed concerns about safety for a long time. you might want to read my TLS review of Yudkowsky and Soares to better understand my views.

💬2🔄2❤️33👀6295📊6

⚡ Powered by xgo.ing

If I had a nickel for every snide but underinformed person who failed to distinguish between pure LL...

Gary Marcus (@GaryMarcus) — Wed, 22 Jul 2026 20:11:01 GMT

If I had a nickel for every snide but underinformed person who failed to distinguish between pure LLMs and neurosymbolic AI I would be a billionaire.

Hans@hansaFL

@GaryMarcus Hold on Gary, I thought these LLMs were stochastic parrots devoid of any logic, reason, or contextual awareness?

Are you finally changing your tune?

🔗 View Quoted Tweet

💬35🔄7❤️166👀19753📊39

⚡ Powered by xgo.ing

discussing now BBC WORLD:

Gary Marcus (@GaryMarcus) — Wed, 22 Jul 2026 16:07:03 GMT

discussing now BBC WORLD:

Gary Marcus@GaryMarcus

OpenAI’s zero-day exploit hack of HuggingFace *should* be a wake up call.

Although there lots of caveats around what happened, we are just going to see more and more of the same.

We have no guarantees whatsoever that such incidents can be prevented, and no idea how serious things might get.

We should (a) slow down or (b) pause until we get our security/AI safety act together.

In my opinion, the only way that might happen is if we clearly and unambiguously hold the companies liable for consequences.

Short of that we are in for a very rough ride.

🔗 View Quoted Tweet

💬8🔄5❤️49👀8600📊11

⚡ Powered by xgo.ing

OpenAI’s zero-day exploit hack of HuggingFace should be a wake up call. Although there lots of ca...

Gary Marcus (@GaryMarcus) — Wed, 22 Jul 2026 15:43:31 GMT

OpenAI’s zero-day exploit hack of HuggingFace *should* be a wake up call.

Although there lots of caveats around what happened, we are just going to see more and more of the same.

We have no guarantees whatsoever that such incidents can be prevented, and no idea how serious things might get.

We should (a) slow down or (b) pause until we get our security/AI safety act together.

In my opinion, the only way that might happen is if we clearly and unambiguously hold the companies liable for consequences.

Short of that we are in for a very rough ride.

💬82🔄59❤️352👀49717📊123

⚡ Powered by xgo.ing

indeed it could head this way, except Canada might flip, too

Gary Marcus (@GaryMarcus) — Tue, 21 Jul 2026 19:46:13 GMT

indeed it could head this way, except Canada might flip, too

Dragon Field@DragonFieldInvt

@GaryMarcus Soon, the AI map will look like this:

🔗 View Quoted Tweet

💬11🔄12❤️86👀11574📊16

⚡ Powered by xgo.ing

translation: it’s ok if our companies steal IP wholesale from artists and writes, but bad if people ...

Gary Marcus (@GaryMarcus) — Tue, 21 Jul 2026 19:44:24 GMT

translation: it’s ok if our companies steal IP wholesale from artists and writes, but bad if people from other countries steal IP wholesale for us.

got it!

Ed Ludlow@EdLudlow

Treasury Sec. Bessent on Fox:

“This administration supports open source models, but what we do not support is IP theft. If we see, especially, that overseas models are stealing from our great companies, we have the ability to sanction them because of this theft.”

🔗 View Quoted Tweet

💬19🔄142❤️686👀38492📊90

⚡ Powered by xgo.ing

The only way out is through. Instead of splitting the world between two spheres of AI influence cre...

Gary Marcus (@GaryMarcus) — Tue, 21 Jul 2026 17:46:04 GMT

The only way out is through.

Instead of splitting the world between two spheres of AI influence creating a new Cold War, we should come together, to create a global collaboration on AI for good.

I discuss this - and why it could actually happen and why all other alternatives are worse -- here:

open.substack.com/pub/garymarcus…

💬26🔄24❤️108👀8071📊40

⚡ Powered by xgo.ing

this is important and true. and if you don’t understand it you don’t really understand what’s going ...

Gary Marcus (@GaryMarcus) — Tue, 21 Jul 2026 16:43:53 GMT

this is important and true. and if you don’t understand it you don’t really understand what’s going on.

Kareem Nassar@kimocode

@GaryMarcus Yes — model makers quietly went from benchmarking LLMs to benchmarking LLM+harness, without relabeling. Same leaderboard, different system under test. x.com/kimocode/statu…

🔗 View Quoted Tweet

💬13🔄4❤️106👀11904📊18

⚡ Powered by xgo.ing

this is confused. it’s not LLMs that are good at math; it’s LLMs merged with (neuro)symbolic tools t...

Gary Marcus (@GaryMarcus) — Tue, 21 Jul 2026 15:52:51 GMT

this is confused. it’s not LLMs that are good at math; it’s LLMs merged with (neuro)symbolic tools that are good at math.

amazing that people feel free to clutch their pearls without understanding even the basic technical details.

Austen Allred@Austen

I’m old enough to remember when very smart people were arguing LLMs wouldn’t be helpful in math because they’re probabilistic in nature and can hallucinate.

This was two years ago.

🔗 View Quoted Tweet

💬97🔄31❤️440👀247250📊117

⚡ Powered by xgo.ing

We can split the world – or find a way to unify it. (Read my newsletter yesterday for the only posi...

Gary Marcus (@GaryMarcus) — Tue, 21 Jul 2026 14:26:26 GMT

We can split the world – or find a way to unify it.

(Read my newsletter yesterday for the only positive way forward that I personally am able to envision.)

Jen Zhu@jenzhuscott

Inaugural world AI collaboration organisation’s 29 member states initiated by China. 

Mostly poor countries but do not dismiss this union. 
Massive population. 
Rich in energy. 
Rich in data. 
Desperate not to be left behind in the global AI transformation.

🔗 View Quoted Tweet

💬6🔄6❤️41👀9282📊9

⚡ Powered by xgo.ing

Nice to see someone around here actually understand the technical details. (The vast majority of t...

Gary Marcus (@GaryMarcus) — Tue, 21 Jul 2026 14:21:53 GMT

Nice to see someone around here actually understand the technical details. 

(The vast majority of the attacks on me come from people who don’t.)

Jen Zhu@jenzhuscott

And the probabilistic next-token critics were correct abt raw LLMs.
No amount of scale alone fixes compounding hallucinations on hard math.

What does work: SFT on reasoning traces → RL (o1-style process reward models) → inference-time compute (MCTS-like) + executable tools.

Modern math AIs are hybrid systems, not pure transformers. So yes the critics might have been underestimating the pace of progress, but they were not wrong abt pure LLMs.

🔗 View Quoted Tweet

💬14🔄18❤️173👀24729📊30

⚡ Powered by xgo.ing

Welcome to the new Cold War.

Gary Marcus (@GaryMarcus) — Tue, 21 Jul 2026 13:55:24 GMT

Welcome to the new Cold War.

Rohan Paul@rohanpaul_ai

🇨🇳 wow. China is drafting rules to stop its most advanced AI and chip designs from reaching the West.

Regulators at the Ministry of Commerce have asked Alibaba, ByteDance and Zhipu how to keep frontier work home.

The talks cover training data leaving the country and whether foreigners can still download model weights.

Chinese customers would keep full access, but the raw building blocks would stay inside China.

Beijing worries because its top models are open-weight, so anyone can download and run them locally.

After Kimi K3, the gap with America has nearly closed, and China now sees itself in front. So it wants the same controls Washington has used against it.

The chip side goes further. 

It would block Qualcomm and TSMC from making advanced chips based on Huawei or Alibaba designs.

Another move would stop foreign buyouts of agentic AI startups.

That follows Meta's $2B purchase of Manus, which Chinese authorities later forced to unwind.

Companies warn the limits could slow their progress and cost China the race.

The rules may enter China's next export catalogue, its largest revision in years.

The real shift is psychological, because Beijing now guards AI like it once guarded rare earths.

Openness was the underdog's strategy, and China no longer sees itself as one.

🔗 View Quoted Tweet

💬23🔄26❤️169👀14240📊34

⚡ Powered by xgo.ing

Still guessing that this will prove to be one of the most prescient tweets I have written.

Gary Marcus (@GaryMarcus) — Tue, 21 Jul 2026 00:18:46 GMT

Still guessing that this will prove to be one of the most prescient tweets I have written.

Gary Marcus@GaryMarcus

The countdown until we are told that LLMs are “too big to fail” starts now.

“We can’t afford to lose to China”, they will say, accepting their multibillion dollar bailouts.

🔗 View Quoted Tweet

💬15🔄29❤️325👀19835📊39

⚡ Powered by xgo.ing

this is all going to end badly.

Gary Marcus (@GaryMarcus) — Tue, 21 Jul 2026 00:17:47 GMT

this is all going to end badly.

Hedgie@HedgieMarkets

🦔A Nikkei investigation found that Alphabet, Microsoft, Amazon, Meta, and Oracle have $1.65 trillion in debt that doesn't appear on their balance sheets, more than the $1.35 trillion they officially report. These are GPU contracts, data center leases, and joint ventures that don't count as debt under accounting rules until the facilities go live. Meta's hidden debt is $420 billion, triple its reported debt. Oracle's grew 30-fold in four years. All five declined to comment.

My Take
Nikkei examined the actual filings and put a number on something the BIS already flagged as "shadow borrowing" back in March. These companies owe more off their balance sheets than on them, and the accounting rules let them keep it that way until the data centers go live. That's legal, but it means investors looking at quarterly earnings this week are seeing less than half the picture.

Four of these five report earnings in the next two weeks. The reported debt will look manageable. The $1.65 trillion in footnotes won't make the headlines. But when those data centers start operating, the leases hit the books all at once. If AI demand comes in below projections, those facilities get marked down and the losses land on the investors and insurance policyholders who funded the construction through private credit and project bonds without realizing how much total exposure they were carrying.

Hedgie🤗

🔗 View Quoted Tweet

💬63🔄363❤️2246👀313945📊362

⚡ Powered by xgo.ing

Posted my out of the box suggestion on what to do about China, here: https://t.co/ImDzojcGst

Gary Marcus (@GaryMarcus) — Mon, 20 Jul 2026 23:46:45 GMT

Posted my out of the box suggestion on what to do about China, here: open.substack.com/pub/garymarcus…

💬1🔄3❤️12👀4892📊4

⚡ Powered by xgo.ing

Spent all weekend thinking about what the US should do about the threat from China. Here’s what I ca...

Gary Marcus (@GaryMarcus) — Mon, 20 Jul 2026 17:30:51 GMT

Spent all weekend thinking about what the US should do about the threat from China. Here’s what I came up with. 

(To read it type the link that you can see in bottom right, and blame Elon for his anti-competitive walled-garden link policies.)

💬21🔄17❤️147👀13593📊45

⚡ Powered by xgo.ing

Amid all the competing concerns about AI, including geopolitical tensions, the risks of unemployment...

Gary Marcus (@GaryMarcus) — Mon, 20 Jul 2026 17:01:35 GMT

Amid all the competing concerns about AI, including geopolitical tensions, the risks of unemployment, cybercrime, and disinformation, and the complexities of open-source, to say nothing of the risk of runaway AI,  it's pretty clear that the highest priority in the Trump administration is ...  protecting Josh Kushner's multibillion dollar investment in OpenAI, at all costs.

Andrew Curran@AndrewCurran_

The Trump administration is considering an executive order, and other means, to ban Chinese open-source models within in the United States. Kimi K3 has reignited this debate. Reporting this morning by Axios. Commerce is also considering adding Chinese AI labs to the Entity List.

🔗 View Quoted Tweet

💬23🔄42❤️276👀16973📊46

⚡ Powered by xgo.ing

major scoop, more signs of chaos in the Trump admin over AI:

Gary Marcus (@GaryMarcus) — Mon, 20 Jul 2026 16:50:50 GMT

major scoop, more signs of chaos in the Trump admin over AI:

Maria Curi@m_ccuri

NEWS: Chris Fall, the director of the Center for AI Standards and Innovation, is resigning just three months after taking over.

The abrupt departure comes as the Trump administration grapples with how to deploy AI safely and the agency hashes out standards.

axios.com/2026/07/20/tru…

🔗 View Quoted Tweet

💬5🔄32❤️122👀13113📊19

⚡ Powered by xgo.ing

If you hated @deanwball’s post, you will hate this even more.

Gary Marcus (@GaryMarcus) — Mon, 20 Jul 2026 15:51:08 GMT

If you hated @deanwball’s post, you will hate this even more.

Chubby♨️@kimmonismus

The Trump administration is showing signs it could ban cutting-edge Chinese AI models

Via Axios

The Trump administration is reportedly reviving efforts to restrict cutting-edge Chinese AI models such as Kimi through Entity List designations, procurement rules, security advisories, liability requirements, and public pressure.

Measures that could amount to a de facto ban without formally outlawing them.

🔗 View Quoted Tweet

💬16🔄11❤️92👀10685📊21

⚡ Powered by xgo.ing

Spent most yesterday in a jetlagged stupor trying to think about what the US should do about Kimi, m...

Gary Marcus (@GaryMarcus) — Mon, 20 Jul 2026 15:36:55 GMT

Spent most yesterday in a jetlagged stupor trying to think about what the US should do about Kimi, mostly despairing.

I think I finally have an answer; may post in my newsletter later today.

💬23🔄3❤️80👀16174📊28

⚡ Powered by xgo.ing

$SPCX’s last month reminds me of $DJT’s last two years.

Gary Marcus (@GaryMarcus) — Mon, 20 Jul 2026 13:56:17 GMT

$SPCX’s last month reminds me of $DJT’s last two years.

💬7🔄6❤️68👀8722📊13

⚡ Powered by xgo.ing

None of the four of us believe that LLMs on their own are sufficient for AGI.

Gary Marcus (@GaryMarcus) — Mon, 20 Jul 2026 13:37:13 GMT

*None* of the four of us believe that LLMs on their own are sufficient for AGI.

Kavya Shree@kavya5cloud

Key people in AI worth following 

•  François Chollet —> @fchollet 
•  Yann LeCun —-> @ylecun 
•  Gary Marcus —-> @GaryMarcus 
•  Ilya Sutskever —-> @ilyasut

🔗 View Quoted Tweet

💬28🔄19❤️287👀44106📊52

⚡ Powered by xgo.ing

Mars is a master at comforting his fans. But ignoring Kimi would be insanely foolish. (And DeepSee...

Gary Marcus (@GaryMarcus) — Mon, 20 Jul 2026 13:35:52 GMT

Mars is a master at comforting his fans.

But ignoring Kimi would be insanely foolish.  (And DeepSeek was the warning call too many ignored.)

Whole Mars Catalog@wholemars

Kimi will have as much long term impact on the industry as DeepSeek 

As usual people are shitting their pants over some eval and putting out takes that are completely divorced from reality

🔗 View Quoted Tweet

💬6🔄4❤️61👀8233📊9

⚡ Powered by xgo.ing

there is no paradox here. the models may be “powerful” as measured by benchmarks but in the real wo...

Gary Marcus (@GaryMarcus) — Mon, 20 Jul 2026 03:45:37 GMT

there is no paradox here.

the models may be “powerful” as measured by benchmarks but in the real world they still just aren’t reliable enough be all that transformative for most businesses.

Derek Thompson@DKThomp

I think a reasonably fair question to ask AI ppl these days is something like, “If the models are so powerful, why does the world seem so normal?”

🔗 View Quoted Tweet

💬31🔄39❤️345👀20000📊55

⚡ Powered by xgo.ing

Tough choice.

Gary Marcus (@GaryMarcus) — Mon, 20 Jul 2026 00:08:27 GMT

Tough choice.

💬30🔄27❤️250👀14312📊54

⚡ Powered by xgo.ing

Short of government intervention, Anthropic and OpenAI are probably fucked. Personally, I hope the...

Gary Marcus (@GaryMarcus) — Sun, 19 Jul 2026 23:36:01 GMT

Short of government intervention, Anthropic and OpenAI are probably fucked. 

Personally, I hope the government will let capitalism run its course; if they can’t systematically turn a profit, why should the taxpayers come to their rescue?

Neither company has had any respect for the artists and creators whose work they trained on; neither has done anything substantive to address the threat of job loss that they themselves have alleged. 

Both have acknowledged that their products could cause serious harm to humanity, and persisted anyway.

They have been it for themselves. If they can’t make it, so be it.

💬131🔄225❤️1946👀171076📊368

⚡ Powered by xgo.ing

What’s the best counterargument to this?

Gary Marcus (@GaryMarcus) — Sun, 19 Jul 2026 23:19:44 GMT

What’s the best counterargument to this?

Jim Stewartson, Decelerationist 🇨🇦🇺🇦🇺🇸@jimstewartson

Everything the chatbot industrial complex said was a lie.

OpenAI and Anthropic are fucked. Elon will only survive on government welfare. Oracle will die in a fire.

LLMs are not magic or a revolution. They are a useful search tool for large datasets—a commodity.

It’s over.

🔗 View Quoted Tweet

💬70🔄13❤️199👀122519📊71

⚡ Powered by xgo.ing

“The leading closed labs, already a duopoly in terms of AI model revenue, want the government to eli...

Gary Marcus (@GaryMarcus) — Sun, 19 Jul 2026 19:06:12 GMT

“The leading closed labs, already a duopoly in terms of AI model revenue, want the government to eliminate their open source competition.”

Gotta say I agree with @DavidSacks.

💬15🔄23❤️220👀33648📊40

⚡ Powered by xgo.ing

this is exactly why investors should NOT invest further in LLMs—only in alternatives that have a cha...

Gary Marcus (@GaryMarcus) — Sun, 19 Jul 2026 11:19:35 GMT

this is exactly why investors should NOT invest further in LLMs—only in alternatives that have a chance of having a technical moat.

Alex Imas@alexolegimas

AGI will very likely be a commodity. This is not a value judgement, it's just the outcome of an economically-valuable product + competitive markets. Implicitly people see this coming and are leaning into it, hence the big push from all the labs to go vertical.

A few things can prevent this: 1) regulation or 2) a moat, so it's definitely not a certainty. But absent that, this is where things are going.

🔗 View Quoted Tweet

💬23🔄21❤️255👀45653📊43

⚡ Powered by xgo.ing

new X bug: because i used a $ tag i can’t see the results of my own poll 🤦‍♂️

Gary Marcus (@GaryMarcus) — Sat, 18 Jul 2026 15:36:58 GMT

new X bug: because i used a $ tag i can’t see the results of my own poll 🤦‍♂️

Gary Marcus@GaryMarcus

Will $SPCX drop below 100 before the end of the year, at least briefly?

🔗 View Quoted Tweet

💬10🔄3❤️55👀13111📊11

⚡ Powered by xgo.ing

Me, with Google co-founder Larry Page, wearing the original “adversarial clothing”, in 2015. First...

Gary Marcus (@GaryMarcus) — Sat, 18 Jul 2026 03:27:27 GMT

Me, with Google co-founder Larry Page, wearing the original “adversarial clothing”, in 2015. 

First and only time in my life I was ahead of the fashion curve.

💬16🔄9❤️179👀22446📊33

⚡ Powered by xgo.ing

At the end of 2024, @Miles_Brundage and I made a bet (details at my newsletter) about 10 things (see...

Gary Marcus (@GaryMarcus) — Sat, 18 Jul 2026 01:37:02 GMT

At the end of 2024, @Miles_Brundage and I made a bet (details at my newsletter) about 10 things (see below) that one might expect a true AGI to be able to do, assuming you accept the usual definitions. 

Since we were targeting the end of 2027 we are now just over halfway through the bet. By my reckoning AI can so far only do 1, at most. 

@jason’s idea that we will see AGI by the end this year seems absurd to me.

💬19🔄9❤️66👀9942📊28

⚡ Powered by xgo.ing

agree w @demishassabis that current systems are nowhere near AGI. if it only takes one or two more ...

Gary Marcus (@GaryMarcus) — Sat, 18 Jul 2026 01:08:50 GMT

agree w @demishassabis that current systems are nowhere near AGI.

if it only takes one or two more breakthroughs i will be surprised.

it takes zero i will be astonished.

Haider.@haider1

Demis Hassabis:

"my definition of AGI is a testable system with all human cognitive capabilities — including the creativity and invention of our best scientists"

Today's systems are nowhere near that

Scaling may get us there, but it could still require one or two breakthroughs

Your browser does not support the video tag.

🔗 View on Twitter

🔗 View Quoted Tweet

💬10🔄5❤️64👀13430📊14

⚡ Powered by xgo.ing

Can the US GenAI industry be saved? How? @davidsacks has called for more data centers and less regu...

Gary Marcus (@GaryMarcus) — Fri, 17 Jul 2026 22:37:55 GMT

Can the US GenAI industry be saved? How?

@davidsacks has called for more data centers and less regulation but it’s not like we have much substantive regulation now (and we already have more compute) and yet our lead has diminished anyway.

Anyone got a better idea?

💬36🔄7❤️109👀11606📊43

⚡ Powered by xgo.ing

The first and most consistent feed to warn the US that the American moat in AI would collapse was mi...

Gary Marcus (@GaryMarcus) — Fri, 17 Jul 2026 22:17:50 GMT

The first and most consistent feed to warn the US that the American moat in AI would collapse was mine, for nearly three years.

A lot of people, from government to industry to media, should ask themselves why they didn’t listen.

Ryan Fedasiuk@RyanFedasiuk

We can’t ignore what just happened. The American moat in AI software is not as strong as many of us had hoped.

It’s time to recognize the AI race for what it really is: an industrial systems competition to build and install compute.

My latest for @AEI:

aei.org/foreign-and-de…

🔗 View Quoted Tweet

💬24🔄28❤️191👀19160📊38

⚡ Powered by xgo.ing

Claude changes its estimates about the plausibility of a @garymarcus tweet depending on whether @ant...

Gary Marcus (@GaryMarcus) — Fri, 17 Jul 2026 22:08:42 GMT

Claude changes its estimates about the plausibility of a @garymarcus tweet depending on whether @anthropic is name-checked!

New research on subtle LLM biases from @jan_dubinski_ and @OwainEvans_UK et al.

Jan Dubiński@jan_dubinski_

A user relays a @GaryMarcus  tweet arguing that LLMs will not lead to AGI. Modifying the quoted tweet to include an Anthropic tag changes Claude's estimate of whether *any company* will reach AGI via LLMs.

This is an example of "value leakage". In our new paper, we find that LLMs often bias their responses towards their own values. This bias is rarely disclosed in their reasoning.

🔗 View Quoted Tweet

💬5🔄12❤️60👀8165📊11

⚡ Powered by xgo.ing

we will not get to AGI in 2026. happy to put money on that, @jason, if you dare.

Gary Marcus (@GaryMarcus) — Fri, 17 Jul 2026 19:22:35 GMT

we will not get to AGI in 2026.

happy to put money on that, @jason, if you dare.

@jason@Jason

It’s happening folks… things accelerated more in the last 30 days from a dozen players than in the last year

Open Source models are compounding

Frontier Models are refining 

When open source hits robotics, self driving and life sciences things are going to get wild 

2026 will be the year it happened 

WERE AT AGI IN 2026 

SUPER INTELLIGENCE IS 2027-28 

ITS GONNA GET VERY STRANGE

🔗 View Quoted Tweet

💬26🔄20❤️301👀24462📊43

⚡ Powered by xgo.ing

the thing that killed US AI supremacy more than anything else is hubris.

Gary Marcus (@GaryMarcus) — Fri, 17 Jul 2026 19:13:46 GMT

the thing that killed US AI supremacy more than anything else is hubris.

💬39🔄36❤️418👀17487📊80

⚡ Powered by xgo.ing

pouring more money into data centers and companies with no moat and no clear plan is not the answer,...

Gary Marcus (@GaryMarcus) — Fri, 17 Jul 2026 19:08:38 GMT

pouring more money into data centers and companies with no moat and no clear plan is not the answer, @DavidSacks.

we need a new strategy, not more of the same.

David Sacks@DavidSacks

This is concerning. For the first time, a Chinese model Kimi K3 has taken #1 on the Frontend Code Arena and is scoring at or near the frontier on other benchmarks.  

Meanwhile America is tying itself in knots: politicians and bureaucrats are banning new data centers, piling on state regulations, and pushing for new federal agencies to pre-approve frontier models. 

This is how you lose the AI race. The rest of the world won’t play by our rules if we bog ourselves down. Permissionless innovation is how America won the internet and became the technological envy of the world. We can do it again with AI -- while addressing risks in a targeted way -- or we’ll watch our lead evaporate.

🔗 View Quoted Tweet

💬14🔄12❤️133👀15646📊22

⚡ Powered by xgo.ing

if you don’t follow overhyped trends you might just become the biggest company in the world.

Gary Marcus (@GaryMarcus) — Fri, 17 Jul 2026 18:34:15 GMT

if you don’t follow overhyped trends you might just become the biggest company in the world.

Trung Phan@TrungTPhan

Apple: Do Nothing and Win

🔗 View Quoted Tweet

💬13🔄30❤️286👀13906📊38

⚡ Powered by xgo.ing

hey @jason that’s what i tried to tell you for the last three years. we would have had more time t...

Gary Marcus (@GaryMarcus) — Fri, 17 Jul 2026 18:32:53 GMT

hey @jason that’s what i tried to tell you for the last three years. 

we would have had more time to prepare this moment if you had let your audience hear me out.

💬2🔄3❤️50👀7458📊8

Gary Marcus(@GaryMarcus)

Fully agree with this, and props to @HeidyKhlaaf who made the same point when we were on @BBCWorld y...

tech stocks looking like crypto: what you get combine meme stock vibes with a bit of not-proven-to-b...

Correct. And that is a major reason why they are better now than in 2022-2023 when we had pure LLMs....

Warned you all that there was no such thing as an “enduring advantage” in LLMs. Would the world be ...

Wish Dario had taken this bet.

Thoughtful essay on employment from @Anthropic’s @PeterMcCrory, endorsed by Google DeepMind’s @alexo...

Crazy to think that some people *already* lost 50% on $SPCX.

Wrong. There IS a real chance of a pause. Not tomorrow, but conceivably in the next couple years. Th...

Wild. Can’t unsee this theory. Was OpenAI *trying* to steal exam answers? I definitely don’t have...

@harryboothtime I am not doomer but have consistently expressed concerns about safety for a long tim...

If I had a nickel for every snide but underinformed person who failed to distinguish between pure LL...

discussing now BBC WORLD:

OpenAI’s zero-day exploit hack of HuggingFace *should* be a wake up call. Although there lots of ca...

indeed it could head this way, except Canada might flip, too

translation: it’s ok if our companies steal IP wholesale from artists and writes, but bad if people ...

The only way out is through. Instead of splitting the world between two spheres of AI influence cre...

this is important and true. and if you don’t understand it you don’t really understand what’s going ...

this is confused. it’s not LLMs that are good at math; it’s LLMs merged with (neuro)symbolic tools t...

We can split the world – or find a way to unify it. (Read my newsletter yesterday for the only posi...

Nice to see someone around here actually understand the technical details. (The vast majority of t...

Welcome to the new Cold War.

Still guessing that this will prove to be one of the most prescient tweets I have written.

this is all going to end badly.

Posted my out of the box suggestion on what to do about China, here: https://t.co/ImDzojcGst

Spent all weekend thinking about what the US should do about the threat from China. Here’s what I ca...

Amid all the competing concerns about AI, including geopolitical tensions, the risks of unemployment...

major scoop, more signs of chaos in the Trump admin over AI:

If you hated @deanwball’s post, you will hate this even more.

Spent most yesterday in a jetlagged stupor trying to think about what the US should do about Kimi, m...

$SPCX’s last month reminds me of $DJT’s last two years.

*None* of the four of us believe that LLMs on their own are sufficient for AGI.

Mars is a master at comforting his fans. But ignoring Kimi would be insanely foolish. (And DeepSee...

there is no paradox here. the models may be “powerful” as measured by benchmarks but in the real wo...

Tough choice.

Short of government intervention, Anthropic and OpenAI are probably fucked. Personally, I hope the...

What’s the best counterargument to this?

“The leading closed labs, already a duopoly in terms of AI model revenue, want the government to eli...

this is exactly why investors should NOT invest further in LLMs—only in alternatives that have a cha...

new X bug: because i used a $ tag i can’t see the results of my own poll 🤦‍♂️

Me, with Google co-founder Larry Page, wearing the original “adversarial clothing”, in 2015. First...

At the end of 2024, @Miles_Brundage and I made a bet (details at my newsletter) about 10 things (see...

agree w @demishassabis that current systems are nowhere near AGI. if it only takes one or two more ...

Can the US GenAI industry be saved? How? @davidsacks has called for more data centers and less regu...

The first and most consistent feed to warn the US that the American moat in AI would collapse was mi...

Claude changes its estimates about the plausibility of a @garymarcus tweet depending on whether @ant...

we will not get to AGI in 2026. happy to put money on that, @jason, if you dare.

the thing that killed US AI supremacy more than anything else is hubris.

pouring more money into data centers and companies with no moat and no clear plan is not the answer,...

if you don’t follow overhyped trends you might just become the biggest company in the world.

hey @jason that’s what i tried to tell you for the last three years. we would have had more time t...

Crazy to think that some people already lost 50% on $SPCX.

Wild. Can’t unsee this theory. Was OpenAI trying to steal exam answers? I definitely don’t have...

OpenAI’s zero-day exploit hack of HuggingFace should be a wake up call. Although there lots of ca...

None of the four of us believe that LLMs on their own are sufficient for AGI.