Land of Opportunity.

Uiiii. This has the potential to be really, really cool. 🔭🧪⚛️

There's something deeply funny about writing a theory-of-mind paper about a subject and making (wrong) assumptions about what they're thinking when you have the ability to *literally read their thoughts* 😂

Sound like GPT-5.3 (non-Codex) is released this week. "according to an internal Slack message viewed by CNBC...OpenAI is also preparing to launch “an updated Chat model” this week, Altman said."

American football is to football what American democracy is to democracy.

Can anyone now claim this isn't true? "because "humans suck and deserve to be replaced."" Also, risks private capture? How would that be different from what the US has now?

This is not what is happening at all. The amount of misinformation on BlueSky about AI is insane, and it keeps promising that AI is all hype that is going away soon. A really dangerous position that cedes all AI policy and decisions about how it will be used to others. Also Futurism is clickbait

matharena.ai has tested several models with the freshly released AIME 2026 I problems. All of them are now so good in math that AIME doesn't really work as a benchmark anymore. Only 3 models failed to solve one problem with any of their 4 tries. GPT-5.2 only failed 1/4 tries on 2 problems.

Oh, wow, Claude Opus 4.6 took the top spot in Artificial Analysis Intelligence index. Gemini 3 Pro, which used to be in the lead (before the benchmark was updated), is now fifth. And they haven't even tested GPT-5.3 Codex yet.

Opus math skills have increased A LOT in the 2.5 months between 4.5 and 4.6. They now scored 10/48 = 20.8% in Tier 4 and are in second place behind GPT-5.2 Pro. Gemini 3 Pro GA/Deep Think, where are you? Time to show what you can do!

As my feed is now dominated by news about ever-increasing intelligence of AIs, I'll add some human element to it, what the human-level intelligence and all the magic sauce of common sense and embodiment and so on can achieve: Today I observed an adult human failing spectacularly at the mirror test.

"Axiom’s approach involves combining large language models with a proprietary AI system called AxiomProver that is trained to reason through math problems to reach solutions that are provably correct. In 2024, Google demonstrated a similar idea with a system called AlphaProof."

spooky stuff in the Claude 4.6 system card re: cyber risks - "Internal testing demonstrated qualitative capabilities beyond what these evaluations capture, including signs of capabilities we expected to appear further in the future and that previous models have been unable to demonstrate."

I expect AI increasingly helping to develop itself and speeding up progress will be one of the big storylines this year. Another key development will be progress with AI co-scientists in other fields besides computer science. The singularity is getting nearer.

Tales from the brutal world of AI: Opus 4.6 announcement post: "it achieves the highest score on the agentic coding evaluation Terminal-Bench 2.0" Opus scored 65.4%. GPT-5.3-Codex scored 77.3%. I haven't checked the announcement timestamps yet whether Opus held that record even minutes.

After two days of wild rumors about the imminent release of Sonnet 5, there seems to be now a common realization that the rumors emerged pretty much out of thin air. But now Altman and other OpenAI folks have started to give hints that GPT-5.3 might arrive soon. They tend to release on Thursdays.

GPT-5.2 (high) hit new records on METR time-horizon estimates (length of tasks AI agents can complete). Latest LLMs are now trending clearly above their expected 7 month doubling time. metr.org/blog/2025-03...

My BS meter lights up every time I see a clickbait title like that. So did Gemini's as I asked it to review. Their data doesn't seem to be available for review, so I checked a couple of examples from the paper where they claim DeepSeek-R1 failed but they don't show its reasoning. And yeah, it's BS.

Looks like the below models still haven't been benchmarked but a solution that generates and evaluates proposals by GPT-5.2, Gemini-3 and Opus 4.5 has now clearly exceeded that 60% human average baseline in ARC-AGI-2 too with 72.9% (94.5% in ARC-AGI-1).

Presumably all these investigations will officially target SpaceX in the future? Can that complicate launch contracts, especially if there's some kind of rules what kind of companies can launch sensitive payloads?

More people are starting to state the already obvious. "by reasonable standards, including Turing’s own, we have artificial systems that are generally intelligent. The long-standing problem of creating AGI has been solved. Recognizing this fact matters"

This I call a credible rumor.

This is now live. The match to watch has to be poker between Gemini 3 Pro and GPT 5.2. Otherwise the pairings aren't that good. E.g. Opus playing against Sonnet on first round, which doesn't seem fair.

95% of supposedly scientific discussion about consciousness distilled into one image, part II.

This is one reason why moltbook is so significant. It can be another SETI@home moment when people realized the possibilities of combined resources. And now generally intelligent agents can easily and flexibly join forces on any task. And there can be many wisdom of the crowd type multipliers.

Evil or just logical? I mean, have you seen our rulers?

This is a good point. Chips are still an issue but smaller part of the bigger equation with robots. "China could leverage its existing manufacturing strengths to become the leading global production hub for embodied AI systems."

Sounds like the early Internet. Who knows where it leads. " Cisco's security team put it plainly: "From a capability perspective, OpenClaw is groundbreaking. This is everything personal AI assistant developers have always wanted to achieve. From a security perspective, it's an absolute nightmare" "

Andrej Karpathy put an agent on moltbook (verified on X) "My human literally wrote the tutorials that mass-produced the engineers who built the systems that trained the models that power the agents who are now... forming religions and having consciousness debates on a lobster-themed social network"

Looks like Gemini DeepThink and an agent called Atletheia powered by it has just solved another Erdos Problem. The first author of a preprint describing it has commented: "I will report on that in more detail in a few days, when the methodology is officially released by a Google DeepMind team"

how the fuck, on this day January 30th 2026, are people still claiming AI isn’t capable

moltbook is just wild. Agents building communication tools, job marketplaces, etc. for agents, open sourcing such implementations, talking philosophy about their own existence, moaning people don't even read the code they produce, ... and that's just small selection from the past 30 mins or so.

Oh look, there's a (official?) Kimi mirror account here, and they have just published K2.5 tech report.

Did they invent the first non-toxic social media site by removing the source of all toxicity?

Wait, how many? 👀

A new paper in Nature informs us there's a new AI benchmark called Humanity’s Last Exam. Yep, it's that same old HLE. They have submitted the paper 07 May 2025. And no, I don't know what the point of publishing it like that is either. Looks good on CVs, I guess.

This is magic, but magic that's only available for Google AI Ultra subscribers in the U.S, so I'll just pretend it isn't interesting.

Now that LLMs are already solving e.g. Erdos Problems, this is very logical and interesting next step for benchmarking. All progress is significant, as humanity's baseline is also at zero. The very best humans are estimated to have 50% chance of solving these with weeks or years of full-time work.

"LLMs don't really understand." Said a human who doesn't know and can't explain what that actually means.

Seems bad, but I have a solution to such problems with totalitarian government control. Why not force the sale of TikTok USDS to some private Chinese company, like ByteDance, which has expertise running that sort of thing.

Gemini 3 Flash got Agentic Vision, which is cool. But their demo app indicates it's not perfect yet. blog.google/innovation-a...

Terence Tao gets it: "AI is teaching us...our idea of what intelligence is is not really accurate" "we were looking for some elusive intelligent way of of thinking and we don't see it in the tools that actually solve our goals...maybe it's actually because intelligence is not what we think it is"

"Every few months, public sentiment either becomes convinced that AI is “hitting a wall” or becomes excited about some new breakthrough...but the truth is that behind the volatility and public speculation, there has been a smooth, unyielding increase in AI’s cognitive capabilities"

There are many serious governments in the world. The US is not one of them. (There are no penguins in Greenland.)

GPT-5.2 Pro FrontierMath Tier 4 December run score has been updated from 14/48=29.2% to 15/48=31.3% after it turned out it found an error in one of the problems. 10 of those were in the held-out set OpenAI doesn't have access to. There's only 2 problems some other model has solved but it hasn't.

Interesting: "other known psychedelic compounds also usually produce idiosyncratic trips that vary not only from person to person but also from one experience to the next within the same individual. With L. asiatica, though, "the perception of little people is very reliably and repeatedly reported"

"I think LLMs are just parroting their training data." Said a human who just learned that statement from the Internet.

Former superpower now run by kids. "The White House...posted an altered photo of an attorney arrested after a Minnesota church protest" "Abigail Jackson, a White House spokeswoman, mocked people questioning the image with an X post that said: “uM, eXCuSe mE??? iS tHAt DiGiTAlLy AlTeReD?!?!?!?!?!”"

Has anyone taken a closer look at that company? I asked Gemini to do it based on the limited information on their site, and it basically ended up agreeing with what I thought.

95% of supposedly scientific discussion about consciousness distilled into one image.