ainch

6 ms·

61.
by ainch 4mo ago
In some sense, science is the most extreme form of compression - Newtonian mechanics explains an incredible number of phenomena in a few lines of text.
62.
by ainch 4mo ago
AISI did also say that GPT-5.5, which has been public for months, scores basically the same as Mythos on their cybersec evaluation. But there wasn't as much media about about that for some reason. https://www.aisi.gov.uk&#x2
63.
by ainch 4mo ago
Token prices have increased, but it's not really the whole story at this point, given some models will use far more tokens to complete a task than others. One of the charts in Anthropic's blog posts shows Fable at 'low'
64.
by ainch 4mo ago
They say it's because the EU's DMA would require them to open up device data to third-party assistants, and they'd no longer be able to guarantee user privacy. https://www.apple.com/newsroom/2026/06&
65.
by ainch 4mo ago
As WIRED reported[0], despite constantly writing about how an AI collapse is just about to come, Zitron privately does PR for AI firms on the side. The man is an obvious hack, and it's disappointing that he has become one of the main
66.
by ainch 4mo ago
I'd spent 6 hours solving a gnarly RL problem (mathematically solving divergence of off-policy TD-Lambda for any value of lambda or behaviour policy). As a punt I gave it to o3 (remember LLMs were 'bad at maths') - after 15 m
67.
by ainch 4mo ago
There's a growing body of evidence that most of what the brain does is constantly predicting the world around us - look into the predictive brain hypothesis if you're interested. As Ilya Sutskever has pointed out, if you read a m
68.
by ainch 4mo ago
You can fine-tune it so, given an image and a task description, it generates a corresponding set of actions.
69.
Training a Simple World Model with Jax(alexinch.com)
2 points by ainch 4mo ago|1 comments
70.
by ainch 4mo ago
Deep Blue's strength was leveraging massive compute to execute a task-specific, human-written algorithm. The problems which LLMs are tackling don't elude mathematicians because they require too much number-crunching, but because t
71.
Speeding up MuJoCo 460x with Jax(alexinch.com)
2 points by ainch 5mo ago|0 comments
72.
by ainch 5mo ago
Very cool work, the learned world state is a smart way of getting consistent generation across all the views (and not having the map vanish when you 180 like some other models). Multi-agent is such an interesting field, because it's cl
73.
by ainch 5mo ago
Thanks for the reply - you're clearly much more experienced with the internals here, but I believe we're still talking at cross-purposes. I believe you're talking about compilers like jax.jit or torch.compile performing symbo
74.
by ainch 5mo ago
At the same time, the training paradigm being scaled, Reinforcement Learning, is significantly less data-efficient than next-token prediction. You basically need to run an agent for minutes (or longer if you want good long-horizon performan
75.
by ainch 5mo ago
The price for a given level of capability will fall, but the frontier has recently been getting more expensive. If you compare GPT-5 to GPT-5.5 on the Artificial Analysis benchmark, it's ~4x more expensive, but achieves a higher score.
76.
by ainch 5mo ago
Tokens can be sold at profit, but 70% of compute expenditure goes to R&D and model training[0]. Inference needs to cover all of that as well as being profitable in a vacuum. [0] https://epoch.ai/data-insights/openai
77.
by ainch 5mo ago
I use jaxtyping as documentation, but the fact it can only be used for runtime checking (in a slightly clunky manner) and can't infer shapes based on ops really limits its utility imo.
78.
by ainch 5mo ago
I think we're talking at cross purposes. The reason I'm excited about the Pyrefly work is that it leverages the type system to infer array shapes statically, which makes it simpler to reason about them when you're writing the
79.
by ainch 5mo ago
I didn't know that, thanks for sharing. It makes sense, but then it also makes me wonder why none of the deep learning libraries (Torch, Jax/NNX, Eigen etc...) make this information available. Instead, ML people all have their own
80.
by ainch 5mo ago
The Pyrefly type checker is starting to work on this kind of shape hinting - so far it only works on Torch but I believe the plan is for it to work with other array packages (eg. JAX, NumPy) https://pyrefly.org/en/docs&
81.
by ainch 5mo ago
I've only discovered Quanta this year but it's quickly become my favourite publication. The focus on quality articles across science, and especially pure maths, feels very unique. I don't know whether it's profitable or
82.
by ainch 5mo ago
Did you publish this anywhere? Would love to read more.
83.
by ainch 5mo ago
Apple already have this working on iOS - as discussed in this recent post. https://unix.foo/posts/local-ai-needs-to-be-norm/
84.
by ainch 5mo ago
I'm glad it's 1.0, but for me the most exciting thing is the tensor shape typing they mention in the blogpost. I've been looking for a good solution to this for ages, and the Pyrefly implementation seems to have everything I&
85.
by ainch 5mo ago
I love Kraftwerk, but contributing to anti-nuclear sentiment in Germany hasn't been a major success. If only more European countries had followed the French example and developed substantial nuclear fleets.
86.
by ainch 5mo ago
Agreed, Gemini is clearly a capable model, but the tool use is lagging behind the other two. Ironically it regularly gets things wrong (ie. the current version of some software) because of an unwillingness to use web search.
87.
by ainch 5mo ago
They've said they'll open source the compiler alongside the 1.0 release.
88.
by ainch 5mo ago
In their original pitch that was definitely part of it: take Python code, add type hints, get a big speedup. As they've built it out it seems to have diverged.
89.
by ainch 5mo ago
As someone in ML who's interested in performance, I'm keen for Mojo to succeed - especially the prospect of mixing GPU and CPU code in the same language. But I do wonder if the changes they're making will dissuade Python devs
90.
by ainch 5mo ago
The AI CEOs love to pontificate about AI curing cancer, but it seems like DeepMind is the only one actively working on these research problems, while OpenAI/Anthropic largely chase enterprise/coding revenue.

More ›