New type of LLM psychosis: “LLM type 2 bipolatity”
In December I had very high levels of anxiety (started taking meds, difficulty sleeping) as part of the Claude Code moment. Sort of depression
Now I am having difficulty sleeping because I want to keep programming with ClawdBot, eating badly, and I unilateraly stopped taking the anxiety medication. Sort of mania.
Let’s if this bipolarity disorder continues and see what my doctor thinks.
Why you think Anthropic has low odds of succeeding?
“We are updating Grok’s constitution. We think it should be able to play many characters, like the waifu anime girl Ani, the truth-seeking AI assistant Grok, and the unhinged version MechaHitler.”
What annoys me about alignment research is that it could be used for create evil models just by trivially plugging a minus sign in front of it.
I have less confidence than most here in how much automated AI researchers will accelerate the takeoff. The difficulty of improving time horizons seem like a big data & compute problem. And it is also not clear how much you’ll be able to iterate in other vectors.
But! I am relatively more confident that for a given level of intelligence, automated AI researchers will be brutal in optimizing the cost to run it. We should expect an acceleration of the price of intelligence. See NanoGPT and PostTrainBench.
This could also be a way the bubble pops. At some point, running more expensive models stop making economic sense for a given task.
Imagine if by Summer 2028 we can run a Claude Opus 6-level model in your own Mac. Will Claude Opus 7 be that much better if you’re not doing advanced research? What about Summer 2029?
No wonder we are hearing about the possibility of a 2026 Anthropic IPO.
Of course being right is better than being wrong. Ideally he should know the exact date of the arrival of the Superintelligence and organize finances for that.
But it seems to me that he has the best shot of creating the AI god with his current process.
Altman doesn’t own equity in OpenAI and he’s doing it for the glory. He genuinely believes he might give birth to the AI god. Why should he do anything different, from his vantage point
I am fairly confident That GPT-5.1, which I’m confident is a check-out of GPT-4o, has more than 60% of its training flops in post-training.
If openai created another GPT-4 pre train, they’d post-train it all over again.
Of course they’ll do it. But just not that often. Likely once a year or something like that.
Many many things in the world are explained by society having an unconscious visceral forecast of AGI timelines in the late 2020s. Many actors behaviors make more sense if you assume they have late 2020s timelines.
Why the end of neoliberalism and the post-war world? Late 2020s timelines. They understand having raw power will be relatively more important than having non-zero-sum relationship with your allies and trading partners.
Why the 2nd Cold War? Late 2020s timelines. U.S. and China know that being ahead during the late 2020s probably means being ahead forever.
Why did Russia invade Ukraine? Late 2020s timelines. Russia knows they couldn’t do it later.
Why Gen Z isn’t obsessed with work but they are obsessed with health and positional goods? Late 2020s timelines. They want to be healthy to await the arrival of the cure of aging and know they won’t have a long career.
Why politics has become more raw, polarized, with each side thinking losing is the end of the world? They want their side in power during the 2020s.
Why long-term interest rates bottomed the same month GPT-3 launched? Late 2020s timelines (ok, this is a joke)
Why the arrival of crypto and gambling culture and ticket prize (e.g.: creator and only fans and hustlers). Late 2020s timelines.
Why are governments running big fiscal deficts with no end in sight? Late 2020s timelines.
This is a bit “wtf happened in 1971?” I mean, the Intel 4004 was launched in 1971, the release of the first microprocessor. And I think what explains this is that this is all part of the same trend of Moore’s Law pushing for automation, just that we are getting to the part of the exponential where the numbers get insane.
Think about theory of the firm: the firm is the largest portion of the economy that is better run through an authoritarian central planning regime rather than using the market prices to orient and organize production.
What we have seen with informatics over the past few decades is exactly that bigger is getting better. For years now the small-cap factor no longer works. J.P. Morgan Chase & Co. is world’s largest bank and it’s outperforming the industry. Amazon is capable to coordinate more than 1.5 million full-time employees worldwide.
As AGI accelerates this trend, there’s no reason to imagine we won’t see further consolidation. Yeah, sure, some people like Pepsi and other people like Coca-Cola. But likely there won’t be 2,000 different soda brands that each one needs to be individual oversight by humans.
If you can organize production more through central planning through informatics and AGI, I dunno there will be much work left for humans to do.
And obviously, people on LW are überbulls on ASI. The view is that it’ll get millions of times smarter, whatever you define, than humans.
I think we should just assume that Claude Code has already been attacked by multiple fronts.
Someone posted on X’s fintwit:
Claude Code virality is first big AI hit which inspires lots of short ideas but almost no longs. Who benefits besides anthropic or ai infra trade that’s already max long?
To me, this is more evidence of dystopia. Maybe I am more optimistic than the average LessWronger who believes in apocalypse, but this updates me towards a very weird future.
This hints to a future where AI is extremely deflationary.
My take is that because intelligence no longer will be what defines us as humans, reconnecting with our bodies is a nice way to find purpose.
Is AI taking artists jobs?
Thank you for replying.
They would be permanently shut out of all Western trade and technology sharing.
That’s not true. Many countries um the West were literally fascists during WWII. I can totally imagine worlds where China and the West get along after that.
All critical semiconductor manufacturing in Taiwan will be destroyed by the US or the local Taiwanese military before China can get to it, and most of it is already in the process of being successfully transferred to the US. I also expect that most of the human talent would be taken to the US.
Why should China care, if they are mostly cut from the output of that anyway?
3) Even if the US did not directly intervene, the US and their allies would start massive rearmament and reindustrialisation programmes and maximally utilise their advantage in AI and other critical technologies in future.
Trump just asked for a $1.5T military budget. It’s already happening.
4) Regarding point 4, if American AI victory is inevitable due to their computing advantage, China might still get a better deal in the current scenario, where it is perceived as merely an economic competitor and geopolitical challenger, rather than a direct adversary, as it would be in the event of an invasion of Taiwan.
American AI victory is not inevitable. And not taking this bet would relinquish China to the permanent undetclass of nations, which i’m sure Beijing doesn’t want.
GPT-4 was pre-trained in 2022. GPT-4o was pre-trained in 2024. Since then, models likely have the same size. Clearly something is happening that no one wants to spend 100x more in a pre-train run. Likely because you need high-qualitt non-synthetic data.
Trade and invest in the U.S. is giving resources to the U.S. now in exchange for pieces of paper that might or might not be valuable in 10 years in a radical y different world.
See my comment trying to pushback on Daniel and Eli. I feel we both are on similar conclusions.
It really helps to think about the world as if people were rational working under full information.
It really makes sense for China to invade Taiwan as soon as this summer.
1- China won’t be able to win the long timelines world many policymakers are planning for. The David Sacks/Jensen Hwang argument is that by precluding China from the best AI accelerators, they’ll eventually have their own domestic supply the U.S. can’t control. When that happens, the U.S. will have a robot army they can cheaply deploy to defend the Taiwan strait.
2- Taiwan is really central for U.S. AI development. Really. Not only TSMC, but Micron, Mediatek, Alchip, all the ODMs like Foxconn.
3- China needs to really consider the possibility of recursive self-improvement. Waiting one year means another $500B in AI equipament for AI Labs.
4- The governance of the CCP over China is not guaranteed under a strong AI world.
5- Trumpism might be gone at the end of the decade and president Gavin Newsom might be much more willing to fight over Taiwan.
Thank you for the effort. Big fan of the authors.
The authors don’t seem to address the possibility that we are seeing a temporary acceleration of AI, because the labs are ramping methods that are much more expensive to scale, but they are doing so from very low baselines.
Here some evidence for you.
1- The acceleration in ECI precedes coding helping researchers in at least 18 months. Based on my anecdotes, I doubt any researcher at an AI lab is being accelerated by AI since they got access to models like 4.5 Sonnet and GPT-5-Codex. Epoch says: “AI capabilities accelerated in 2024! According to our Epoch Capabilities Index, frontier model improvement nearly doubled, from ~8 points/year to ~15 points/year.” I don’t think there’s any reason to believe that AI-aided R&D acceleration has happened in any meaningful way, other than maybe Sholto’s comment.
2- One place where has been an acceleration is on my spending on AI. I am now spending more than one thousand dollars in tokens and the marginal task of my job I am automating with AI costs what I used to pay for AI during an entire month. Toby Ord argues that the costs of AI are increasing exponentially: “the hourly costs for some models are now close to human costs.” While the evidence is small and we need further work, if each jump makes the marginal task exponentially more expensive, but for a fixed level of intelligence, we get prices 90% cheaper per year, one could imagine a point where we achieve the AGI at 2028, but only can deploy it economically in 2030. And a world where we achieve the Automated Coder in 2031, but only can deploy it economically in 2035.
3- Despite the METR and ECI indexes of capabilities per unit of time following an exponential with even an acceleration, the underlying trends have changed massively. a- Pretraining scaling has slowed down massively since the GPT-4.5 debacle. b- Massive efforts have been done to create human cured data around the matters we care about. SemiAnalysis say the labs are spending single-digits billions on human generated data. Beren argues most algorithimic progress is data progress. Obviously, replacing the corpus of text from random dudes debating in a 2007 forum to all the intermediate steps of a math proof by a math PhD improves the models. Obviously, this can’t scale and is an one-off improvement. b- Inference-time scaling has been improving the models considerably. To the point, I consider OpenAI models like GPT-5.2-Codex-High unusable, given how slow they are. Not only that, but gains from inference-time scaling must be paid every time they are executed. I don’t think we can continue to scale inference time compute into the back-half of the decade. c- Toby Ord also argues that RL is in on the order of 1,000,000x less compute efficient than pre-training. He says “I estimate that at the time of writing (Oct 2025), we’ve already seen something like a 1,000,000x scale-up in RL training and it required ≤2x the total training cost. But the next 1,000,000x scale-up would require 1,000,000x the total training cost, which is not possible in the foreseeable future.” Regardless of the level, I feel anyone paying attention feels the same way. Ilya argues that RL is learning from a straw.
3a- Google DeepMind Co-founder and CEO, Nobel Prize winner, Demis Hassabis said he is spending most of his time on world models. Facebook AI Research co-founder Yann LeCunn says “LLMs are a dead” and is working on world models. I feel that the “straight-line on charts” crowd, which I am definitely part of, ignore the important aspects of empiricism in the construction of human knowledge. We won’t create a LLM with a one-month time horizon and it will reason from first principles how to cure cancer. That’s exactly the opposite lesson from Samuel Albaine’s Compute Theory of Everything.
4- The authors don’t address that they are making a somewhat unverifiable prediction. The largest tasks inside the METR are on the order of 16 hours. I’d argue that the complexity of benchmarking translates to the complexity of improving the models themselves.
4a- I can imagine doing RLVR with endless goals, like Stockfish is always getting better in chess. Maybe we can have LLMs that are ever increasing better in creating better matrix factorization algorithms. I struggle to find which types of such algos we could have where overshooting human capability would be insanely singularity good.
4b- RL doesn’t seem to generalize. My market Will a large language model beat a super grandmaster playing chess by EOY 2028? is at 44% and the trend is down. Maxin Saplin’s leaderboard of LLM chess has Gemini 3 Pro merely at 1033 rating, vs 1500 for a “class C player”. While I have no doubt that if the labs wanted, they could RLVR chess into their LLMs, I think chess is a good example that you can’t do insane amounts of RL in one direction and expect good things in other directions.
5- I’d argue “significantly more important than the internet” singularity requires solving one or more of continual learning and simulation (a.k.a. world models). Computers will only get better in matters that involve the real world quickly if they aren’t bounded by the real world.
All that said, I confess the straight lines on a chart are immensely persuasive and hard to not extrapolate for many years through the Lindy Effect.
I plan to brief and lobby a presidential candidate with high single digits probability of getitng elected in my country about AI safety. What books or essays you guys think I should read as a last sprint ditch to get read to the task?