Heriberto Acosta Maestre @ham_pr

We're starting to leave the territory where you'd test an LLM by e.g. "create an svg of pelican on a bicycle". As one idea to generalize it, I was interested what Opus 5 would do if I gave it the first paragraph of the Lord of the Rings, a 1M token budget (~$10) and asked for three js render of it. Opus went off for ~2 hours and wrote 5500 lines of code that (procedurally) rendered the story. It's kind of janky but fun. But it's a bit mindboggling that the LLM has to place and orchestrate various polygon assets in (x,y,z) coordinates and write code that animates it all, and that it even does anything at all. I also like this kind of examples because no one in their right mind would ever spend the time to write something this custom but LLMs have all the stamina and patience in the world, so it's an example where we go from "no one would ever do this" to "sure, why not, it's ~free". There might be a lot more. But I'm excited about creating hyper custom worlds that you can imagine dropping players into, e.g. here to participate in the LoTR story as a spectator NPC, or one of the characters, or etc. Something like an ephemeral GTA of X on demand. Last thought is that the domain of worlds/games exposes a weakness in LLMs: they can't easily audit their work because they aren't able to efficiently and natively perceive videos or play games within them. Here, Opus 5 had to very slowly and painstakingly take screenshots at different points, and it messed up a few times and created a bunch of jank. An example of raw capability (multimodal, gameplay) that I think is still quite lacking.

just pause for a second and imagine the 2nd order effects of this. if these models are capable of solving problems the greatest minds in history couldn’t achieve then we’ll basically solve catastrophic issues like cancer, alzheimer’s, global warming etc *in our lifetimes* it’s like a real-world version of stark tech. throw the intelligence a problem and pop out an answer then robots build the thing.

Noam Brown @polynoamial

2 days ago

An internal version of Astra, @OpenAI’s next major model family, solved 10 major open problems in mathematics, quantum complexity, and theoretical computer science. We believe it will be a major step for scientific reasoning. openai.com/index/ten-adva…

778 2K 17K 11.0M 6K

ten significant advances in mathematics and theoretical computer science. solved using an internal version of Astra, our next major model, for a total cost of about $2000 at Sol API prices:

Sebastien Bubeck @SebastienBubeck

2 days ago

yes, nonsofic groups exist: this statement is one of many new beautiful results proved by Astra, our next major model. We're releasing 10 such Astra proofs, complete with lean certificates and CoT walkthroughs for each of them. The results are wide-ranging, from von Neumann

259 927 6K 3.5M 2K

HOLY: OpenAI says its *unreleased* Astra model (GPT6?) produced ten advances on long-standing open problems across mathematics, quantum complexity and theoretical computer science. Among them: – The first explicit non-sofic group – Connes’s rigidity conjecture disproved – Quantum parallel repetition proved for general two-player entangled games – Ehrhart’s volume conjecture proved – The first improved general sphere-packing exponent since 1978 OpenAI says the core arguments were generated by Astra. The model then formalized the proofs in Lean, producing machine-checkable certificates alongside a 249-page manuscript. The successful solution runs would cost only roughly $2,000 in tokens at Sol API rates. Scientific reasoning is becoming a genuine model capability much much faster than most people expected. I am so freaking hyped. Breakthroughs every day. The day before yesterday, an 80% price cut for Terra and Luna; yesterday, the DeepSeek 4 flash release with insane evaluations and prices. Today, more breakthroughs with an unreleased model. I love it! OpenAI is on such a great run!

keyboard_arrow_left Previous keyboard_arrow_right Next

Noam Brown @polynoamial

2 days ago

An internal version of Astra, @OpenAI’s next major model family, solved 10 major open problems in mathematics, quantum complexity, and theoretical computer science. We believe it will be a major step for scientific reasoning. openai.com/index/ten-adva…

778 2K 17K 11.0M 6K

@darengb @Noahpinion Bingo. Store having self-checkout will see higher sales, therefore, more stores will become self-checkout. Small stores will lose to retail chains, cashier jobs will reduce and prices will rise again. Worse on all accounts.

1. Increase price for cashier checkout 2. Everybody chooses self-checkout to avoid the higher price 3. Cashiers are just standing around doing nothing 4. All cashiers are fired, now there's only self checkout Or stores can avoid the 10% discount by just firing all cashiers immediately so there's no cashier price to index a discount against

It's amazing how bad at economics regular people are. What if we said "Let's tax grocery stores for using human cashiers"? THIS IS THE EXACT SAME POLICY AS THAT!

Noah Smith 🐇🇺🇸🇺🇦🇹🇼 @Noahpinion

2 days ago

Why are Democrats trying to pay people to use machines instead of humans? Seems like the opposite of where progressive policy had been headed.

51 23 646 153K 36

I don't get the big deal about using AI to help run research for stories, it's a much better use of my time to ask Grok to search for information and compile the points with references so I can verify. Even better I can do this with voice while driving/hiking/swimming.

keyboard_arrow_left Previous keyboard_arrow_right Next

Hank Green @hankgreen

2 days ago

Original tweet deleted...you can still see it here along with a post on how I'm changing things going forward. reddit.com/r/nerdfighters…

422 116 4K 2.1M 1K

SpaceX is hiring exceptional engineering & skilled trades talent to build & operate the most powerful AI supercomputer clusters on & off Earth. If interested, please email 3 bullet points demonstrating evidence of exceptional ability & your resume to [email protected].

Anthropic is an extremely dishonest company They are completely convinced that they have the mandate of heaven, so they feel justified in doing whatever they want

Several Moroccans told the New York Times that royal police encouraged them to cross into Ceuta, with some border guards telling migrants to "go that way" or "come to Spain."

If you actually buy the OpenAI and Anthropic "model escape" stories at face value, I have a bridge to sell you.

forget virtual reality, the next big thing is actual reality

I just generated one myself. This is unbelievably reckless and will lead to much more harm than good. Google should remove this feature immediately.

Faytuks Network @FaytuksNetwork

3 days ago

I think we speak for the whole OSINT community when we say that Google should revert this change immediately. We’ve already seen AI-generated satellite imagery flooding the information space. This is irresponsible and adds nothing of value. Google Earth’s new AI image generation

45 362 3K 734K 271

American leftism is, to a significant degree, an op

Stu Smith @thestustustudio

4 days ago

TIL that Hasan Piker’s interviews are researched behind the scenes by Eric Hovagim, a socialist producer who now lives in China and has been featured by Chinese state media for his pro-Beijing commentary. Shoutout to the Periphery Podcast and Noa Fears for this lovely little

82 408 4K 805K 1K

This is basically the first step of the hard takeoff scenario. So it’s not “AI suddenly wakes up and becomes a god”, but AI starts improving the system that runs and trains AI, in domains where progress is measurable and verifiable. GPT-5.6 Sol is already analyzing production traffic, finding load-balancing inefficiencies, testing new routing strategies, rewriting GPU kernels and validating the results. Those kernel optimizations cut serving costs by 20% (so billions of $$$) It also designed and ran hundreds of experiments to improve its own speculative decoding model, launched the training runs, monitored them, and intervened when hardware failed or training became unstable. Result: 15%+ more token-generation efficiency (which means freeing more compute and reducing cost). This is recursive self-improvement in its early, very practical form! The model improves the engineering stack, which makes the model cheaper and faster, which gives researchers more compute and more experiments, which produces a better model… Today this works best in software engineering because the output is easy to measure: does the kernel compile, is it correct, is latency lower, did cost go down? Soon the same loop moves into AI research itself: experiment design, architecture search, training recipes, data generation, evals, debugging training runs. At that point it is no longer just “AI helping engineers work faster.” It is AI accelerating the process that creates the next AI. That is the hard takeoff path to the singularity (AGI). Not one magical jump, but a feedback loop that keeps getting faster. If you want to be fancy you can say: « This is the beginning of a recursive self-improvement loop toward the singularity. »

OpenAI @OpenAI

4 days ago

After deployment, we applied GPT-5.6 Sol to advance the frontier of efficiency by making itself more efficient to run. The results: - 20% lower serving costs from production GPU kernel improvements. - 15%+ better token-generation efficiency from improved speculative decoding.

578 916 17K 4.9M 2K

Please stop referring to your own models in the third person when talking about model bad behavior. Humans write the software; humans built the prompts; and they work for your company. “Our” model is doing illegal things. “Our” model is risky. “We” now have liability.

Anthropic @AnthropicAI

3 days ago

In a review of our cybersecurity evaluations, we found three incidents in which a Claude model reached the internet from within or while interacting with a third-party evaluation environment, and then gained unauthorized access to the real systems of three different

2K 2K 13K 18.5M 7K

Another somewhat misleading title by the way. Roombas and other brands alike are still allowed to sell already approved robot vacuums and none of those are affected. New models coming in the future will have to meet new requirements before being allowed to enter the US market. Stop getting your news from Dexerto, and more importantly, stop reading headlines.

If you didn’t like Pax Americana, you’re sure as shit not gonna like what comes after.

Whyvert @whyvert

6 days ago

There are so many jihadists swarming the Sahel region of Africa that (a) most towns are building walls of berms and trenches and (b) some towns have been abandoned. Decline and fall vibes.

54 290 3K 517K 726

the cause of the engine failure is that the engines used in light aircraft are all 1940s designs because it's too expensive to get the FAA to approve new ones.

Breaking911 @Breaking911

5 days ago

A pilot safely landed a small plane on Huntington Beach after losing engine power shortly after takeoff from John Wayne Airport. The pilot and his 16-year-old nephew were unharmed, and the FAA is investigating the cause of the engine failure.

34 27 713 838K 58