kzrdude

It's from the European part of South America.

The names, I thought they were going to keep Sol, Terra, Luna for a while (while increasing versions). Are names like Sol and Astra really burned as one-offs? I think that's a waste of a good model name. Hope they keep them going with updates, as nicknames for the various model sizes on offer.

Some kind of linter should flag these with a warning, I think

I live in a city (a quite small one) and if I say that I'm going into the city, then it means I leave my neighbourhood (it's in the same city) and I'm going to the city centre.

I don't think that's a problem, and it doesn't mean that Britain is not in Europe, IMO. Comes naturally with human perspective. "Where did you go this summer?" A) "I stayed in England"; B) "I travelled around Britain"; C) "I went to Europe". All perfectly understandable, I think..

The OP is about formalizing an existing result, not coming up with a new proof for FLT.

And not to mention, the hamster is standing at the wrong end of the table.

I've long observed that kind of behaviour in google translate (which makes sense, they have been using ML for a long time.)

That should be expected based on the scaling laws that we expect; larger models are more intelligent and cost more. Now it's very unfortunately that they don't publish the size of their models.

Yes

FLT was proven in 1995 by Andrew Wiles (with help of Richard Taylor).

This is not even a new proof, or at least they don't claim that it is. It's the formalization (in Lean) of an existing proof. That means, they are 'porting' the proof to a theorem proving programming language.

And the multiple Numberphile appearances of Ken Ribet are interesting too! He is incredibly well spoken.

- https://www.youtube.com/watch?v=nUN4NDVIfVI (The bridges to Fermat's Last Theorem)

- https://www.youtube.com/watch?v=NPOw4iIxN6o (podcast)

The part about prove2.me was interesting. That means that a co-working tool was instrumental in the project, and I think AI companies will take note of this. Is this proof specific or will we need to give agents access to JIRA or similar tools to solve large projects in the future?

The Uno "diffusion adaptor" will take a while for me to understand, but sounds very interesting.

https://huggingface.co/IFM/K2-Horizon-7B-Uno

This one is her latest paper: [20] J. Stadlmann, Bounded gaps between primes, Forthcoming

Ok, so the rumour was exactly true: there was a withheld prime gaps improvement, that "an AI company" was holding onto until release of a model.

I see that Muse Spark 1.3 (max) beats GPT-6 Astra on some benchmarks:

Test: Muse Spark 1.3 / GPT 6 Astra

DeepSWE v1.1: 75.4% / 74.1%

AutomationBench: 49.4% / 41.4%

Is that enough to bring this discussion down to earth again?

Bridge ends in midair

Wind sketches the farther bank

The far bank draws near

by gpt-5.6-sol

You probably mean over 30-40 km/h, not 60.

In general, all cars are too wide now because they have gradually maximized their width up to a limit. The old Italian Fiat cars were much more nimble, of course.

I don't think LLMs are properly self-referential. They reference a frozen training reality, which is not itself, but the old description of itself and the old world. Being aware would probably include continuously updating yourself (learning) from experience, including experience of oneself.

We also had GLM-5.3 flash and Qwen 3.8 Flash Next, everyone's getting flashed and I think it's a good trend.

Almost suspect that the rate of improvement to post-training is so fast that small models have an advantage - it takes much more compute to train a bigger model, so the flash models are just running in circles (well, not exactly of course) around the larger models right now.

As another non-native speaker, I misunderstood what the person meant.

Colibri did that first for GLM-5.2 https://github.com/JustVugg/colibri

And there are `Mference` and `SwiftLM` too, I think they are doing the same use case.

It's worth listening to Lawrence Lessig, speaking in 2011, "Republic Lost" on youtube: https://www.youtube.com/watch?v=Ik1AK56FtVc (It's a "Talk at Google")

I think he laid it out very clearly, but never found traction to try to change this (I don't think US Congress wants to be changed, though).

I don't think I can see the difference. I just have my skin crawl because I'm expecting to see something off and generally have a bad feeling about it.

By the way; I found that DeepSeek says that it is even an "ideal complement to modern MoE architecture"

https://arxiv.org/abs/2601.07372v1

It is compatible, and Qwen3.8 Flash Next is MoE

They are still industry leaders. They'll have to try to maintain that.