[GUEST ACCESS MODE: Data is scrambled or limited to provide examples. Make requests using your API key to unlock full data. Check https://lunarcrush.ai/auth for authentication information.]

@GaryMarcus Gary Marcus

Gary Marcus posts on X about open ai, agi, elon musk, prediction the most. They currently have XXXXXXX followers and 3600 posts still getting attention that total XXXXXXX engagements in the last XX hours.

Engagements: XXXXXXX #

X Week XXXXXXXXX +57%
X Month XXXXXXXXX +16%
X Months XXXXXXXXXX +28%
X Year XXXXXXXXXX -XX%

Mentions: XXX #

X Month XX -XX%
X Months XXX +6.50%
X Year XXXXX -XX%

Followers: XXXXXXX #

X Week XXXXXXX +0.65%
X Month XXXXXXX +2.10%
X Months XXXXXXX +11%
X Year XXXXXXX +22%

CreatorRank: XXXXXX #

Social Influence #

Social category influence technology brands celebrities finance travel destinations stocks countries social networks

Social topic influence open ai #127, agi #75, elon musk #601, prediction, told #1707, llm, vibe #1413, money #3344, 6969, woke

Top assets mentioned Alphabet Inc Class A (GOOGL) Artificial Intelligence (AI4)

Top Social Posts #

Top posts by engagements in the last XX hours

"This is vastly oversimplified. But still a good reminder of how much more complex the brain is than so-called neural networks"
@GaryMarcus on X 2025-07-19 00:27:22 UTC 184.5K followers, 21.4K engagements

"Desperately need a name for the fallacy of thinking that a single impressive result entails AGI given how many people here committed that error in the last XX hours"
@GaryMarcus on X 2025-07-20 15:19:13 UTC 184.6K followers, 27.1K engagements

"@elonmusk My prescription: stop accelerating recklessly and endorse measures for AI safety more vigorously"
@GaryMarcus on X 2025-07-20 01:26:30 UTC 184.6K followers, 81.3K engagements

"woke up early on a saturday to hype my latest product anecdotally without giving any data at all. i am good at hype very very good"
@GaryMarcus on X 2025-07-19 21:08:08 UTC 184.5K followers, 15.5K engagements

"@chatgpt21 i thought dm would get there this year or next. but with a different approach. the surprise was how OpenAI did it but details are scant"
@GaryMarcus on X 2025-07-20 03:56:00 UTC 184.4K followers, XXX engagements

"if I said this I would be stoned alive. but it is true. (where LLMs refers to base model excluding additional gadgets)"
@GaryMarcus on X 2025-07-08 22:54:27 UTC 184.5K followers, 13.8K engagements

"@OwainEvans_UK Such an amazing result. Like I get the theory in the abstract. But wow"
@GaryMarcus on X 2025-07-22 17:15:08 UTC 184.6K followers, XXX engagements

"psychiatrists / therapists / social workers and other mental health workers only: Are you seeing significantly more psychosis among your patients than you did a few years ago If yes feel free to respond in comments about whether this appears to be tied to AI"
@GaryMarcus on X 2025-07-18 20:01:27 UTC 184.6K followers, 11.7K engagements

"@DHuskytron I think o3 has some genuine uses but all this AGI talk is not realistic"
@GaryMarcus on X 2025-07-22 03:38:25 UTC 184.6K followers, XXX engagements

"OpenAI vs DeepMind smackdown Except taken from forthcoming essay with Ernest Davis (6th in US Math Olympiad as a high school junior top X in Putnam as a college sophomore)"
@GaryMarcus on X 2025-07-22 02:45:51 UTC 184.5K followers, 1521 engagements

"When we get to 2028 and AGI hasnt yet been achieved and LLMs are causing psychiatric problems and undermining democracy and cybercrime is an epidemic insidious microtargeted ads are everywhere nobody trusts anything colleges no longer work and climate change is worse will people look back at the choices we made in 2025 fondly"
@GaryMarcus on X 2025-06-13 14:51:30 UTC 184.6K followers, 78.6K engagements

"Apparent LLM-induced delusions in one of OpenAIs early backers. Sad case but I do hope it draws more attention to a growing problem"
@GaryMarcus on X 2025-07-17 22:10:21 UTC 184.6K followers, 12.6K engagements

"@zjasper666 @OpenAI smart question conjecturing that the deepmind approach might be cleaner"
@GaryMarcus on X 2025-07-20 03:01:04 UTC 184.5K followers, 4339 engagements

"About two people a minute are responding to this poll. I did not at all expect that. If people are answering honestly that seems like a sign that something serious is going on. I will post final results in the end; please dont reply unless you or a love one have experienced delusions connected with AI"
@GaryMarcus on X 2025-07-18 20:42:40 UTC 184.5K followers, 7910 engagements

"I thought only New York Times columnists could be this gullible"
@GaryMarcus on X 2025-07-20 22:24:24 UTC 184.6K followers, 11.1K engagements

"So (as I speculated yesterday) @GoogleDeepMind DeepMind got EXACTLY the same score as OpenAI did on the IMO. Neither model is yet available for outside review"
@GaryMarcus on X 2025-07-21 18:17:57 UTC 184.6K followers, 17.8K engagements

"expect a comment from me on the OpenAI IMO Gold - more questions than answers since so little info is available about the system that won - sometime this weekend"
@GaryMarcus on X 2025-07-19 13:07:48 UTC 184.6K followers, 20.6K engagements

"A computer scientists perspective on vibe coding:"
@GaryMarcus on X 2025-05-17 00:25:52 UTC 184.6K followers, 842.4K engagements

"Incredible how little people are talking about this when it is essentially the same result as OpenAI (literally the same score) just reported a couple days later presumably out of deference to the IMO and its participants. OpenAI burned a lot of capital in the math community to get a small short-lived PR victory. Lets see now which result is more robustly usable in the real world"
@GaryMarcus on X 2025-07-21 19:28:38 UTC 184.6K followers, 40.9K engagements

"Even in my meme below I didnt plot zero progress. I have often talked about diminishing returns. I have said publicly e.g. that o1 and o3 and the new model are impressive. Never once did I say that the field has made no progress post GPT-4. (I did say that certain problems like hallucinations would stymie pure LLMs; that was and remains true.) No matter. The best OpenAI can do to respond to me is with insults (one dude literally gave me the finger earlier today) or with strawmen (as below) pretending I said zero progress and then refuting that strawman rather than addressing my actual"
@GaryMarcus on X 2025-07-22 03:15:24 UTC 184.6K followers, 16.8K engagements

"When the the AI turns us all into paper clips it may well be because someone at the Pentagon or in Beijing forgot to add the right special token at the end of a prompt. 🤷♂"
@GaryMarcus on X 2025-07-22 02:14:34 UTC 184.6K followers, 16.2K engagements

"@zjasper666 @YiMaTweets @ylecun thats sensible. if he were more open to including symbolic AI into the mix he would have a better way of specifying those goals"
@GaryMarcus on X 2025-07-20 04:27:00 UTC 184.4K followers, XXX engagements

"Will GPT-5 be decisively better than o3 and Grok X across most measures and use cases"
@GaryMarcus on X 2025-07-13 00:15:12 UTC 184.5K followers, 29.2K engagements

"please help me with a title: a. Welcome to GenAI Dystopia b. Enshittification goes mainstream c. The downsides of Generative AI are spreading like wildfire or d. other (specify in the comments)"
@GaryMarcus on X 2025-07-04 20:23:22 UTC 184.5K followers, 8246 engagements

"@PixelCanuck @CerciDuman77635 @Michael_ReedSEA LOL. The context for those who missed it was that I (known GenAI skeptic) said a new AI model was impressive and people leapt to the utterly unlicensed conclusion that AGI was imminent. In all likelihood we have a long long journey left"
@GaryMarcus on X 2025-07-20 23:33:33 UTC 184.5K followers, XX engagements

"Personal Update: My p(doom) has gone up. I dont foresee machines with malice any time soon. But I am starting to see the harm that powerful yet reckless humans who are indifferent to humanity could cause"
@GaryMarcus on X 2025-07-14 01:02:56 UTC 184.6K followers, 69.9K engagements

"Radical changes in Elon Musks behavior and outlook have elevated my p(doom). To really screw up the planet you might need something like the following. A really powerful person with tentacles across the entire planet Substantial influence over the worlds information ecosphere A large number of devoted followers willing to justify almost any choice Leverage over world governments and their leaders Physical boots on the ground in a wide part of the world A desire for military contracts Some form of massively empowered (not necessarily very smart) AI Incomplete or poor control over that AI A"
@GaryMarcus on X 2025-07-15 14:08:18 UTC 184.4K followers, 33.4K engagements

"a common assumption that should seriously be questioned: As artificial intelligence (AI) rapidly advances many AI experts predict that the first state to develop artificial general intelligence (AGI) that can perform a wide range of tasks better than humans will gain huge advantages in military and economic power. more likely that advantage will like many historical technical advantages be short-lived"
@GaryMarcus on X 2025-07-17 17:58:58 UTC 184.4K followers, 11.3K engagements

"OpenAIs high level of maturity and intellectual integrity simply cant be beat I love the carefully-reasoned argument. cc @_KarenHao"
@GaryMarcus on X 2025-07-21 18:21:00 UTC 184.6K followers, 12.8K engagements

"this is hilarious. like they just woke up and discovered that black boxes are uninterpretable. in reality this was always going to happen and something many of us warned about for years. arrogance is what has gotten us to this point"
@GaryMarcus on X 2025-07-20 16:16:40 UTC 184.6K followers, 23.2K engagements

"reaction to the OAI gold with a new undisclosed model coming soon"
@GaryMarcus on X 2025-07-19 14:11:41 UTC 184.5K followers, 5914 engagements

"@lifeafterAi_ i fully stand by my prediction that that (AGI end of 2025) aint happening"
@GaryMarcus on X 2025-07-19 13:31:15 UTC 184.3K followers, XXX engagements

"here is the fuller context impressive yes. agi no"
@GaryMarcus on X 2025-07-19 19:28:29 UTC 184.4K followers, 6581 engagements

"near-human since when that is certainly NOT what people like @bengoertzel @peterevoss had in mind when they coined the term. or when DeepMind said first solve intelligence then solve everything else more broadly AGI generally defined as doing the things people can do with some room to argue around average human or better than experts that I captured here:"
@GaryMarcus on X 2025-07-20 23:20:42 UTC 184.5K followers, 1350 engagements

"@littmath @heseynus @AIMelGibson @polynoamial @alexwei_ thats impressive"
@GaryMarcus on X 2025-07-19 14:19:33 UTC 184.6K followers, 92.9K engagements

"Black box AI has been a massive possibly fatal mistake wrt alignment. This man speaks truth"
@GaryMarcus on X 2025-07-20 03:57:05 UTC 184.6K followers, 26.5K engagements

"@chatgpt21 @ylecun my guess: alex wang wouldnt be running the show if that were case"
@GaryMarcus on X 2025-07-21 03:40:47 UTC 184.6K followers, 5506 engagements

"Personal observation: The level of intellectual discussion with @GoogleDeepMind vs @OpenAI that I am able to have is literally night and day. DeepMind knows my work can raise serious objections propose and develop alternatives etc. OpenAI speaks to me with insulting memes and misrepresents my arguments. One place seems intellectually serious the other doesnt"
@GaryMarcus on X 2025-07-22 00:51:18 UTC 184.6K followers, 43.2K engagements

"@NLeseul @gandamu_ml @FarmerVaderMD it is not actually clear what they did. the thing at the top of trace could be the AIs summary i just cant tell"
@GaryMarcus on X 2025-07-20 00:01:19 UTC 184.4K followers, XXX engagements

"but. Gary nobody is actually overhyping this result"
@GaryMarcus on X 2025-07-19 22:54:55 UTC 184.6K followers, 16.9K engagements

"last year people were claiming GPT-5 was going to be some monstrously large whale. now they are hoping for a 10-20% improvement"
@GaryMarcus on X 2025-07-21 14:52:45 UTC 184.6K followers, 31.5K engagements

"DeepMind and OpenAI at The IMO compared and discussed:"
@GaryMarcus on X 2025-07-22 20:24:42 UTC 184.6K followers, 3909 engagements

"Sure but the idea that I said zero progress is still a strawman. (I only mentioned benchmarks because you did.) Please go back and read what I actually said deep learning is hitting a wall; the problems I raised there remain problems and the primary technique i pointed to make progress bringing in symbol-manipulation has accounted for some of that progress. I dont think OpenAI has ever public acknowledged this. My experience in getting these systems to draw images from data have been mixed. I still cant for example get a decent labeling of a bicycle. And I dont think you really addressed my"
@GaryMarcus on X 2025-07-22 03:25:45 UTC 184.6K followers, XXX engagements

"Andrew fascinating thread. Two questions X. Who has actually argued Even now (neuro)symbolic advocates often argue that pure symbol manipulation is the only "real" intelligence and "messy" deep learning should just be an input to symbol manipulation I consider myself to be one of the leading advocates of neurosymbolic integration but dont recall ever having said that or seen it. X. What do you see the role of symbols in the current system I cant quite follow from your thread or the post what was done here (as opposed to the AlphaProof work)"
@GaryMarcus on X 2025-07-21 22:24:08 UTC 184.6K followers, XXX engagements

"GPT-5: Last Years Dreams vs This Years Reality"
@GaryMarcus on X 2025-07-22 02:01:01 UTC 184.6K followers, 23.7K engagements

"This is looking like a solid prediction from January as made (separately) by @fchollet @elonmusk and me"
@GaryMarcus on X 2025-07-22 05:06:45 UTC 184.6K followers, 9513 engagements

"@Swarm_caste XXX% certainty not this year 95%+ not this decade. i am a scientist not a bullshitter so uncertainty has to expand with distance in the future"
@GaryMarcus on X 2025-07-20 23:23:33 UTC 184.5K followers, XXX engagements

"Every time I talk about AGI some people run this bait and switch:"
@GaryMarcus on X 2025-07-20 23:24:41 UTC 184.6K followers, 8721 engagements

"How o3 and Grok X Accidentally Vindicated Neurosymbolic AI is now live Worked very hard on this one because the issues are absolutely essential for a clear understanding of where we are in AI now and how we got here. Link below. Featuring a lot of history a critique of @geoffreyhintons less generous side and some sharply-observed bits from @_KarenHaos new book"
@GaryMarcus on X 2025-07-13 17:20:34 UTC 184.6K followers, 165.4K engagements

"@chatgpt21 ironic since project orion was supposed to be X and got relabeled 4.5"
@GaryMarcus on X 2025-07-21 14:55:11 UTC 184.6K followers, 1009 engagements

"What happens to OpenAI if GPT-5 is NOT decisively better than o3 and Grok X WIll be people be disappointed Will the company lose its shine"
@GaryMarcus on X 2025-07-13 00:33:07 UTC 184.5K followers, 22.5K engagements

"Neurosymbolic AI and how it was accidentally vindicated"
@GaryMarcus on X 2025-07-13 17:20:36 UTC 184.5K followers, 9008 engagements

"Quote of the day: I certainly don't agree that machines which can solve IMO problems will be useful for mathematicians doing research in the same way that when I arrived in Cambridge UK as an undergraduate clutching my IMO gold medal I was in no position to help any of the research mathematicians there it is still entirely unclear whether things will scale from machines being able to do mathematics which can be solved using high school techniques to machines being able to help with mathematics which can only be solved by having a deep understanding of modern research ideas this is a big open"
@GaryMarcus on X 2025-07-19 15:33:16 UTC 184.6K followers, 150K engagements

"but some people did in fact basically do that taking last nights announcement as evidence that i was wrong to claim last night that Schmidts conjecture (top mathematicians within a year) was implausible. others took the new result to indicate that AGI was imminent. its those people whose logic i challenged above. my fuller take is here"
@GaryMarcus on X 2025-07-19 21:17:11 UTC 184.4K followers, 1335 engagements

"@karadeku @elonmusk china is actually more actively regulating AI than the US"
@GaryMarcus on X 2025-07-20 01:35:27 UTC 184.4K followers, XXX engagements

"The paradox of the OpenAI IMO discussion is that the new model scored only slightly better than DeepMinds system from last year (as @NeelNanda5 notes); but that we assume that the new model is far more general. Yet we have not yet seen any direct evidence of that"
@GaryMarcus on X 2025-07-19 21:49:59 UTC 184.6K followers, 10.2K engagements

"To a certain extent this is confused. Frontier models usually means really big LLMs but the new system from @GoogleDeepMind is a neurosymbolic hybrid. No pure LLM is anywhere near getting a silver medal in a math olympiad. The most important lesson from the new system (AlphaProof) is that we need to broaden our search of architectures and embrace symbolic AI as a component contra recent remarks from @ylecun. Pure deep learning had a good run but it is time to move on"
@GaryMarcus on X 2024-07-25 17:39:13 UTC 184.3K followers, 15K engagements

"Ok then @elonmusk stop accelerating recklessly and endorse measures for AI safety more vigorously"
@GaryMarcus on X 2025-07-20 01:29:29 UTC 184.6K followers, 31.4K engagements

"a good set of challenges for AGI from 2022 none have been solved community accepted them as legit at the time goalposts havent moved translation and lessons: X. people are making shit up when they say I am moving my goalposts; they havent changed. X. those with money or ego at stake have been trying to redefining the terms relative to what was once basically consensus. X. people like @kevinroose like to ignore this part of the past in order to paint a narrative; if they do that dont trust them"
@GaryMarcus on X 2025-07-20 18:51:24 UTC 184.6K followers, 9670 engagements

"@Bighubba98 there havent been that many serious breakthroughs. i will be shocked if AGI is achieved this decade and offered to put a lot of money where my mouth is"
@GaryMarcus on X 2025-07-20 15:26:35 UTC 184.5K followers, XXX engagements

"No sir you missed the point. Aside from objecting your tone I find your reasoning to be inaccurate and sloppy: X. Complex reasoning is erratic. There are all kinds of illustrations of failures in complex reasoning such as Apples illustration of failure of Tower of Hanoi on X discs and several other recent papers showing similar collapse in light of moderate complexity. And then you have agents committing tax fraud etc. X. Many people leapt to inferences about AGI based on IMO results. Also last week Eric Schmidt said that AI would be at level of top mathematician next year etc. X. I am hardly"
@GaryMarcus on X 2025-07-21 14:30:11 UTC 184.6K followers, 9689 engagements

"Grok X Hot Take Good progress on public benchmarks But only XX% on AGI-ARC-2 Still struggling on visual understanding and image understanding Vindication for neurosymbolic AI - most of the boost comes from integrating symbolic tools not pure scaling see upcoming Marcus on AI Substack for discussion No major new techniques or other innovations revealed No mention of real progress on hallucinations (as far I saw) Hard to know how much benefit people will see on individual real-world use cases. Not a lot of confidence expressed around Xai being able to align Grok to be morally good. Lots of"
@GaryMarcus on X 2025-07-10 05:07:06 UTC 184.4K followers, 61.4K engagements

"@Xmachina9607492 not yet. a lot of what i said in my hot take around OpenAIs result still applies. the community will need to take it for a spin before firm conclusions are reached"
@GaryMarcus on X 2025-07-21 19:32:43 UTC 184.6K followers, 2132 engagements

"Hot take on OpenAIs IMO gold What does it mean I dont know (yet). The fact that no tools coding or internet was used is genuinely impressive. That said my overall impression is that OpenAI has told us the result but not how it was achieved. That leaves me with many questions assuming that the result which has not been independently validated by the IMO is legitimate. Architecture and training. The model that got the Gold did much better than any publicly available model (Though DeepMind is rumors to also have a new similar result). Why is the new system better How does it differ from previous"
@GaryMarcus on X 2025-07-19 16:58:28 UTC 184.6K followers, 150.8K engagements

"dear @grok i thought your promised me you would do better. this still seems like very poorly considered advice"
@GaryMarcus on X 2025-07-17 22:05:19 UTC 184.4K followers, 12.8K engagements

"@boazbaraktcs I explained the justification for that assertion (including some of the history) here:"
@GaryMarcus on X 2025-07-22 03:35:46 UTC 184.6K followers, XXX engagements

"Detailed comparison of the new DeepMind and OpenAI gold performances (to the extent that is even possible) co-written with @ErnestSDavis only at Marcus on AI"
@GaryMarcus on X 2025-07-22 20:21:55 UTC 184.6K followers, 24.8K engagements

"@BasedBeffJezos @Drachs1978 @elonmusk bullshit but i will answer the question above if your care"
@GaryMarcus on X 2025-07-20 03:36:05 UTC 184.4K followers, XXX engagements

"@ojoshe godspeed to making that avenue work. i certainly think that education has a lot of headroom to get better even if I dont immediately see the path you see"
@GaryMarcus on X 2025-07-22 17:08:04 UTC 184.5K followers, XX engagements

"@flowersslop we stopped biological weapons. one could have made similar arguments. sanity prevailed"
@GaryMarcus on X 2025-07-20 03:14:17 UTC 184.5K followers, 1186 engagements

"@chatgpt21 you realize deepmind got a silver last year right with a score just a tiny bit lower"
@GaryMarcus on X 2025-07-19 22:08:30 UTC 184.3K followers, 2058 engagements

"@sjgadler i literally say i am critiquing those who inferred that AGI is imminent from this particular result"
@GaryMarcus on X 2025-07-19 21:18:39 UTC 184.4K followers, XXX engagements

"1. I plotted some progress not none. So youre post is strawman that doesnt correctly capture what I plotted. (Of course what I plotted was not to be taken literally. Sorry if that was not clear.) X. We all know there has been a LOT of focus on benchmarks and virtually no disclosure of what augmentation has been used etc which makes 1:1 comparisons on many benchmarks suspect in a Goodharts law sort of way. It is extremely difficult from the outside to know how much of the variance to attribute to improvements in base models vs other contributors. X. o3 hallucinated more than o1. In some places"
@GaryMarcus on X 2025-07-22 03:05:15 UTC 184.6K followers, 1212 engagements

"I love this. Whats your list of XX things GPT-5 must absolutely be able to do so you dont end up coping later and calling it good when its actually mid If there were no memeably stupid errors #10 I will be shocked"
@GaryMarcus on X 2025-07-21 03:50:18 UTC 184.6K followers, 15.7K engagements

"@FeltSteam i think it is hard to adjudicate the extent to which it is a pure LLM without more detail. there is likely a symbolic outer loop aggregating things for example and maybe calls to external tools etc"
@GaryMarcus on X 2025-07-19 13:27:59 UTC 184.5K followers, 5160 engagements

"@FeltSteam @realKingRoland9 @VeredShwartz had some examples i mentioned in my recent substack on world models where performance was poor on stuff humans would find obvious"
@GaryMarcus on X 2025-07-21 20:49:16 UTC 184.5K followers, XXX engagements

"Priceless line from Ernest Davis (6th in US Math Olympiad as a high school junior top X in Putnam as a college sophomore) in our forthcoming discussion of DeepMind and OpenAIs IMO golds:"
@GaryMarcus on X 2025-07-22 02:57:28 UTC 184.6K followers, 16K engagements

"BREAKING: Explosive new paper from MIT/Harvard/UChicago. Things just got worse a lot worse for LLMs and the myth that they can understand and reason. The paper documents a pattern they called Potemkins a kind of reasoning inconsistency (see figure below). They show that LLMs - even models like o3 make these errors frequently. You cant possibly create AGI based on machines that cannot keep consistent with their own assertions. You just cant. success on benchmarks only demonstrates potemkin understanding: the illusion of understanding driven by answers irreconcilable with how any human would"
@GaryMarcus on X 2025-06-27 16:05:56 UTC 184.6K followers, 418.1K engagements

"Not only is this a fallacious argument (perhaps said in jest) but in reality the last time I said an OpenAI result was impressive wasnt even that long ago. it was December. And it was from OpenAIand it (ARC-AGI w o3) didnt even replicate Reality is pretty much the opposite of your joke"
@GaryMarcus on X 2025-07-20 15:34:56 UTC 184.5K followers, 2627 engagements

"@boazbaraktcs X. the returns that i think are diminishing are on base models with pure pretraining scaling which is what I warned about in 2022 X. i think the gains are increasingly coming from the use of (neuro)symbolic tools. X. i agree that o3 is better that GPT-4"
@GaryMarcus on X 2025-07-22 03:34:03 UTC 184.6K followers, XXX engagements

"FTFY: most aggressive hype. We will not have AGI by the end of 2025. Willing to publicly bet @sama or @elonmusk. (My criteria generally accepted at the time were laid out in my 2022 essay Dear Elon Musk)"
@GaryMarcus on X 2025-07-20 21:32:48 UTC 184.6K followers, 17.7K engagements

"@Leegaul @GoogleDeepMind @OpenAI i find the insecurity remarkable given how many times i have personally and publicly called their work impressive and it is never enough for them or their fans"
@GaryMarcus on X 2025-07-22 01:27:30 UTC 184.6K followers, 3559 engagements

"@GestaltU i literally called it impressive in my longer hot take. but it is not AGI or even close. there is nothing disingenuous about saying that. learn to make your arguments without insult or be blocked"
@GaryMarcus on X 2025-07-20 15:11:06 UTC 184.5K followers, XXX engagements

"I seriously dont see how alignment could ever possibly be achieved in (pure) LLMs in light of results like these. Game. Set. Match. We need to move on"
@GaryMarcus on X 2025-07-22 17:30:52 UTC 184.6K followers, 8159 engagements

"Judge decides case based on hallucinated cases. It just gets worse and worse"
@GaryMarcus on X 2025-07-04 02:36:30 UTC 184.3K followers, 35.4K engagements

"Terence Tao quietly seething at OpenAI is a sight to behold"
@GaryMarcus on X 2025-07-20 04:09:25 UTC 184.6K followers, 50.1K engagements

"this poll is ONLY for people who have either (a) experienced delusions in connection with AI or (b) have loved ones who have. which system do you most associate that experience with"
@GaryMarcus on X 2025-07-18 20:15:00 UTC 184.5K followers, 14.3K engagements

"All the tech bros this morning thinking that AGI has been achieved because some (insanely expensive) new form of LLMs can now match top high school students on one specific task its almost cute ☺"
@GaryMarcus on X 2025-07-19 15:52:18 UTC 184.6K followers, 234.6K engagements

"Custom t-shirts and lawn-chairs for while you wait for AGI"
@GaryMarcus on X 2025-07-20 23:45:40 UTC 184.6K followers, 8407 engagements

"The Marcus Test for AGI (I didnt name it that but it means no AGI til Marcus says its AGI) is a pretty good test for AGI.* But dont bullshit yourself about what it is. Its not about achieving a single resultI have used the word impressive (or similar) to describe aspects of many systems from AlphaGo to AlphaFold to AlphaGeometry to Cicero to o3 to OpenAIs new IMO system. None of these are close to AGI. Getting to AGI is going to require many advances not just one. Not every impressive result will have staying power or lead to AGI. Cicero was impressive but nobody these days uses it at all."
@GaryMarcus on X 2025-07-20 16:07:51 UTC 184.6K followers, 36.1K engagements

"@zjasper666 @OpenAI my guess is your prediction is right. we shall see and yes i do hope we will there albeit on safer ground: more lean and such with verifiably less hope-for-the-best approximation"
@GaryMarcus on X 2025-07-20 03:51:31 UTC 184.4K followers, 2963 engagements

"@BasedBeffJezos @Drachs1978 @elonmusk tagged you above if you want a real conversation"
@GaryMarcus on X 2025-07-20 03:38:40 UTC 184.4K followers, XXX engagements

"when did it might kill us but I need to build it faster become fashionable"
@GaryMarcus on X 2025-07-20 02:27:54 UTC 184.6K followers, 34.7K engagements

"what almost every amateur who has been sold bullshit about vibe coding is eventually going to encounter"
@GaryMarcus on X 2025-07-08 06:58:32 UTC 184.6K followers, 238.7K engagements

"@majesticcoder yes - but are they compatible we dont know yet also DMs approach is likely vastly more efficient"
@GaryMarcus on X 2025-07-20 15:16:05 UTC 184.5K followers, XXX engagements

"Stargate struggles to get off the ground; builds a small data center instead"
@GaryMarcus on X 2025-07-22 04:58:34 UTC 184.6K followers, 5853 engagements

"A world-famous media outlet just asked me what exactly can we conclude from the new OpenAI and DeepMind IMO results And I gave my honest answer. We really dont know. Because there has been zero scientific scrutiny so far of the results. Might be a big deal. Or might be like chess computers that led nowhere. Only when we can do science will we know"
@GaryMarcus on X 2025-07-21 22:29:12 UTC 184.6K followers, 18.9K engagements

"Same guy less than ten minutes apart doesnt see the irony"
@GaryMarcus on X 2025-07-20 01:42:13 UTC 184.5K followers, 19.7K engagements

"memorization versus reasoning is a false dichotomy. almost all machine learning systems can go beyond their training data. the question is how far they can go (only within distribution out of distribution accessibly reliably out of distribution etc). since OpenAI hasnt disclosed anything about their training data or techniques we cant yet reach any sound conclusions about how general the solution is. surprised to see something so premature and combative from a Princeton professor. (further reply to him below)"
@GaryMarcus on X 2025-07-21 15:00:14 UTC 184.6K followers, 11.7K engagements

"🤷♂ all that nonsense about AI imminently replacing programmers and mathematicians was nonsense. even LLM fans are starting to see it"
@GaryMarcus on X 2025-07-19 06:44:03 UTC 184.6K followers, 34.9K engagements

"@Drachs1978 @elonmusk here is the most recent of the hundreds of times i have answered this question (cc @BasedBeffJezos in case he wants a serious rather than name-calling conversation):"
@GaryMarcus on X 2025-07-20 03:38:09 UTC 184.4K followers, XXX engagements

"@liron teleoperate a robot to tidy up random kitchens. if a human did it nobody would be remotely impressed. AI controlling robots wont be able to handle it two years hence except as limited demos"
@GaryMarcus on X 2025-07-20 03:25:51 UTC 184.5K followers, 5727 engagements

"Why did everyone get excited when I called (one aspect of) the new OpenAI model impressive I also said that each of these models were (in some respects) impressive. r1 o1 o3 AlphaFold AlphaGeometry2 Cicero Genesis and thats just a partial list. None of these were AGI. The new one isnt either. Nobody even remembers Cicero. Its a long journey and we have a long way to ago. Lets celebrate progress but not get overexcited either"
@GaryMarcus on X 2025-07-21 00:43:24 UTC 184.6K followers, 15.8K engagements

"15 years ago: robot folding laundry sped up 50x"
@GaryMarcus on X 2025-07-20 01:18:17 UTC 184.6K followers, 7494 engagements

"@CarolineRommer @GoogleDeepMind @OpenAI i am referring to the researchers (and management)"
@GaryMarcus on X 2025-07-22 01:28:13 UTC 184.6K followers, 4239 engagements

"👇Breaking coda to the IMO Gold story: In some ways especially transparency and replicability this dances circles around the OpenAI and DeepMind IMO reports. Not only does @lyang36 get comparable performance out of an already publicly available model supplemented with some additional mechanisms he actually explains what he did"
@GaryMarcus on X 2025-07-22 17:01:15 UTC 184.6K followers, 8720 engagements

"Someday the whole field will understand this. See link below read by 100000 people for an explanation of what it means"
@GaryMarcus on X 2025-07-18 17:01:58 UTC 184.6K followers, 20.3K engagements

"and it might cause significant harm to humanity. but please dont hold us liable"
@GaryMarcus on X 2025-07-21 13:20:17 UTC 184.6K followers, 6604 engagements

"Helicopters are hella impressive. They dont get us to the moon. Why is that hard to understand"
@GaryMarcus on X 2025-07-20 23:28:16 UTC 184.6K followers, 10.5K engagements

"@l_hey_l so many people said this because they wanted it to be true. dozens maybe hundreds plus many likes on many similar statements"
@GaryMarcus on X 2025-07-20 15:25:42 UTC 184.5K followers, XXX engagements

"@truth_zeeker what definitive take(s) are you referring to i am aware of the above"
@GaryMarcus on X 2025-07-19 13:39:20 UTC 184.3K followers, 1063 engagements

"You think AGI is coming this year GPT-5 (which will) didnt win IMO Gold. A different experimental model did. That model probably wont come out this year. And it didnt even win IMO (25+ high school students did better; most professional mathematicians probably could if they put the time in). And IMO isnt professional math. Or professional anything else. AGI means doing everything top professionals can do. That aint happening this year. It aint happening next year. And most likely not this decade. Dont be suckered by the hype"
@GaryMarcus on X 2025-07-20 22:36:06 UTC 184.6K followers, 26.7K engagements

"NOW the OpenAI-DeepMind race is on and is what it should be. Three questions to watch: X. Which company can be first to get out the thing that they just announced to actual user X. Which can do it in a way that is general enough to be broadly useful X. Which can do it in a way that withstands scientific scrutiny"
@GaryMarcus on X 2025-07-21 18:39:56 UTC 184.6K followers, 12.2K engagements

"what did I say was impossible certainly not AGI. i have said it will come in the next 10-20 years. all the failures that i predict XX years ago persist and not have solved. it will take to time to remedy them. you have taken post hoc the worst predictions from the past and strung together some half-baked argument around them and around a misrepresentation of me. A for history; F for logic"
@GaryMarcus on X 2025-07-21 00:47:44 UTC 184.5K followers, XXX engagements

"lets not assume this prematurely; especially after @kashhill carefully reported a case of someone without apparent prior history of psychosis experiencing chat-induced delusions"
@GaryMarcus on X 2025-07-18 19:16:47 UTC 184.6K followers, 8875 engagements

"The glorious Age of AI Agents that people like @sama and @DarioAmodei promised is turning out be a bust. All of this was entirely predictable. Shame on anyone who believed them"
@GaryMarcus on X 2025-07-11 18:52:12 UTC 184.4K followers, 67.3K engagements

"@SunIsMissing @TheZvi that part is clear. also seems likely that OpenAI announced first but werent necessarily only ones to do well. deepmind and maybe Harmonic too see also Terry Taos Mathstodon post"
@GaryMarcus on X 2025-07-20 21:29:28 UTC 184.5K followers, XXX engagements

"@BasedBeffJezos @elonmusk You have a better prescription for his woes Also regulation is not necessarily deceleration; it often helps industries e.g by increasing consumer confidence hence sales and revenue. My real issue is with the recklness rather than speed per se"
@GaryMarcus on X 2025-07-20 01:34:55 UTC 184.4K followers, XXX engagements

"is there a word like cope for taking something entirely out of context"
@GaryMarcus on X 2025-07-19 19:26:19 UTC 184.4K followers, 11K engagements

"no. but i also have almost no idea what they did in terms of either training (including eg augmentation) or architecture nor how generalizable it will be and Sam made it clear that it wont be available for inspect for many months. i already note that it came with a linguistic cost. see my take below for why I am reserving judgment:"
@GaryMarcus on X 2025-07-20 02:34:24 UTC 184.6K followers, 6987 engagements

"WATF @grok This is incredibly irresponsible. cc @elonmusk"
@GaryMarcus on X 2025-07-17 15:46:43 UTC 184.5K followers, 36.8K engagements

"Hottest way to score cheap likes on Ai Twitter today is to quote me as saying the OpenAO IMO result is impressive (it is and I really did say that) without the full context. Here is the full context:"
@GaryMarcus on X 2025-07-19 19:44:03 UTC 184.5K followers, 8137 engagements

"1. @GoogleDeepMind DeepMind makes a fair point X. Willing to bet their proofs will be cleaner. X. Willing to bet their system uses less compute to achieve what is likely (roughly) the same score. X. Their conduct unlikely OpenAIs was sportsmanlike"
@GaryMarcus on X 2025-07-20 17:28:28 UTC 184.6K followers, 30.3K engagements

"@hehehe52318711 @littmath here are the six that OpenAIs system addressed:"
@GaryMarcus on X 2025-07-19 16:37:19 UTC 184.5K followers, 1062 engagements

"Re-upping this in light of JD Vances two recent meetings with Rupert Murdoch. Standing by my original prediction"
@GaryMarcus on X 2025-07-21 05:58:37 UTC 184.6K followers, 16.9K engagements

"Large language models (LLMs) are becoming less "intelligent" in each new version as they oversimplify and in some cases misrepresent important scientific and medical findings a new study has found"
@GaryMarcus on X 2025-07-19 14:11:08 UTC 184.5K followers, 11.5K engagements

"Sam I dont take issue with OpenAIs achievements I take issue with the hype and a history of misleading statements (some detailed below). Youve built something impressive and in many ways groundbreaking. I would be singing OpenAI's praises it's amazing how you scaled Transformers - if it werent for the hype much of it fanned by you that has massively overstated what the technology can do today and in the near future going back to the Rubik's cube demonstration that misrepresented what you had done. I think this hype is harming the world. If I seem frustrated its because of the damage and the"
@GaryMarcus on X 2025-06-11 01:36:19 UTC 184.6K followers, 140.4K engagements

"welp serious problem for scaling test time compute"
@GaryMarcus on X 2025-07-22 20:02:19 UTC 184.6K followers, 8299 engagements

"excellent discussion of how a big pile of fanfic became fodder for apparent LLM-induced delusions"
@GaryMarcus on X 2025-07-18 14:13:28 UTC 184.3K followers, 7496 engagements

"Advice from a machine learning software engineer @ Google re coding agents and where they do and dont help:"
@GaryMarcus on X 2025-07-20 03:04:17 UTC 184.6K followers, 9120 engagements

"Nope. Contra Eric Schmidt world-class AI mathematicians will NOT emerge in the next year and top-tier AI programmers will not NOT emerge in two years. Any more than we had a million robotaxis five years ago. (cc: @metaculus)"
@GaryMarcus on X 2025-07-19 00:00:52 UTC 184.6K followers, 25.2K engagements