Hacker Newsnew | past | comments | ask | show | jobs | submit | mkl's commentslogin

Lots of people are talking about that, and have been for a while. Autoformalisation is clearly going to be a big deal, so mathematicians have been discussing it seriously, and using it where resources allow. A fine-tuned distilled model that could do it on high-end consumer hardware could really help.

Edit: There's also quite a bit of learning needed to use the tools, and to understand enough to confirm that the theorem being verified is what you think. And of course a lot of maths can't yet be expressed in Lean as the foundations haven't been built up enough.


Qwen3.8-Flash-Next loads in 60gb on quant4. Thats pretty close to consumer hardware.

Is it any good at autoformalisation? I think it's likely to take focused fine-tuning to get something small enough that is still good at that.

so far unsloth only has https://unsloth.ai/docs/models/qwen3.8/train which is the 27B model. It also getting pretty close to what you'd need.

I wouldn't be surprised if someone is trying it.


you dont have to take headlines literally.

Most people do, so the literal interpretation matters a lot


They're not claiming controlling robot arms with language models is a sensible or efficient thing to do, they're trying it to see what happens.

i am tempted to think using it for robotics will become the standard because humans like to have one size fits all solutions.

Trying is stupid.

Trying is fun!

The official list is at https://improbable.com/ig/winners/?amp=1#ig2026 but Ars Technica has a lot more detail.

I like that so many of the winners attended the ceremony, one should never take yourself too seriously.

Why do Sol and Terra use 26 input tokens when the others use 16?

I don't know and I'm really Interested in the answer.

I heard a rumor that Luna is a slightly different architecture from Sol and Terra, which makes me wonder if Luna and Astra might be more related to each other than to Sol and Terra.

Bit of a big leap to make from a token count though!

I just checked the tiktoken library and couldn't see any changes relating to Luna: https://github.com/openai/tiktoken


It's just a play on the name, and it's pretty common. claude.ai has nothing to do with Anguilla, John Romero's rome.ro has nothing to do with Romania, twitch.tv has nothing to do with Tuvalu, etc.

Indeed, and Bit.ly has nothing to do with Libya, nor Lemmy.ml with Mali (both failed states). I posit that domain hacking is an ugly, shortsighted, unserious habit that we should drop.

Only with "display: grid;"?

I get what you mean, but those are the ankles. The knees are higher up and bend the same way as a person's. https://en.wikipedia.org/wiki/Bird_feet_and_legs


A pelican wouldn't/couldn't pedal with its ankles. It doesn't make any more physiological sense than a human doing that.

In NZ it's easy, and some GPs can do it as of earlier this year.

IDK I'm just going by a friend moving there and not being able to get any of her existing medication prescribed, even though she's been diagnosed for literally 40 years.

Isn't this well known? The human retina has fewer blue-sensitive cone cells, and very few in the fovea, which sees the sharpest details: https://en.wikipedia.org/wiki/Cone_cell#Distribution


From the paper's abstract (emphasis mine): "Wavelength influences multiple aspects of visual performance, yet its role in spatial resolution remains incompletely understood due to confounding factors such as luminance differences, chromatic aberration, and intraocular scatter."

They then did some modeling to estimate how well the results could be explained by scattering, chromatic aberration and, neural sampling (the cone cell density factor that you mentioned). They concluded that none of them could completely explain the observed results, but their results seemed to suggest the bulk of the effect is due to optical effects, with neurological factors as a secondary cause.

Refining existing knowledge by examining phenomena in more detail is how science should be done. The pressure to focus on discovering entirely new things is a major cause of the replication crisis.


Ah, another over-claiming university press release. That explains it!


I don't think it over-claimed anything? Maybe over-simplified for certain purposes. But to expect a press release to be a complete accounting of the research it reports on is to misunderstand the purpose of press releases. A press release that says everything that's in the paper would be pointless and redundant.

Here's a neat demonstration of this https://www.schulz-hildebrandt.com/9-purple-dots-illusion/ (I find the effect works much better on my phone screen than desktop)


I certainly thought so. I have astigmatism and I cannot for the life of me read blue LED/neon signs at any reasonable distance at night. Any other colour, no problem. Blue is just a blurry mess.


Well, the title here at HN implies something else than the study shows.

If blue light actually impaired vision, that would be very notable. It would mean that removing the blue 'channel' for an RGB image would improve acuity. This is not what the study showed, however.

It is mainly just the basic fewer cones bit you mentioned. Which, btw, is easily experienced if you ever look from a distance at neon or led signs that only emit blue light. They'll persistently look out of focus when compared to their surroundings.


> If blue light actually impaired vision, that would be very notable.

It is extremely noticeable for me at night because I can never fully focus on signs using blue LEDs, while red LEDs give me no problem.


Yeah. I worked in the optical industry to pay for college. That was decades ago and one of the Drs I worked with and I had discussions on this. It was interesting at the time as I knew very little about optics when I started.


Blue Blocker glasses were a thing since the 80s


I get that the guy needed to write a paper on something to get his PhD (in psychology, which might explain some of this as well, I thought standards were a little higher in the hard sciences), but don't know why it deserves a high profile. At least, based on merit.


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: