kzrdude

Lots of used bookstores are online, and I’ve ordered lots of books that way.

This website for example may look like a user-to-user book sales platform but most books there are supplied by used bookstores: https://www.bokborsen.se/

Then we should change the law

That is just a policy problem. A reactive legislature could fix this in a week, rewriting the law.

For example, require the digitized copies to be put in a public trust where for example libraries can provide these books as digital loans. (Per author adjustment for copyright, if it’s not possible now, store them nonpublically in the trust until later.)

We don’t know but I think we assume that it is MoE or some almost equivalent sparse activation technology.

For some tasks yes, and we don't know how many active parameters Luna is using..probably less than 27B

Raw model size is around 27x2 GB since it's in BF16 format

It's fun to see "test time" scaling work out so well, maybe the best example of all.

There is an extra green square where no egg is present, so it's a false positive in the expected.

Great, happy that it was helpful

In the third vision bench result, Sol is 100% correct but the expected has 1 error. Seems like an oversight.

In the next bench, Sol looks like it’s correct again but the bboxes are rotated 90 degrees for some reason.

Nils van der Poel was* a long distance speed skater who set impressive records and wrote about his training regime. His book is https://www.howtoskate.se/. He trained ambitiously for 3 years, won Olympic gold medals and retired! He followed a very analytically minded path and avoided overtraining.

See also brief overview article about him:

https://runningmagazine.ca/sections/training/olympic-speed-s...

(*) He changed his name since, to a more common name, so using past tense seemed humorously appropriate.

I think (maybe I missed something) that identical size and quant versions of Qwen 3.5 and 3.8 should run at the same speed. It’s the exact same architecture.

Some weeks ago a new official Gemma 4 release was posted that corrected some of the chat template problems. So the official release files on hugging face should be the way to go.

If they don’t loudly promise to not scan or store requests they handle then I believe they already do it, or will.

Depends on how you work with it. Reading the meandering and repetitive thinking is disturbing and taxing, so we can’t do that. So thinking has to be (mostly) hidden and just becomes waiting time.

It selects tokens but they expand to embedding vectors which are huge, also in memory and attention requirements, I think?

I don't get it, for pi.dev it's easy to replace or append to the system prompt using a text file. Per user or per project.

I would guess it's the 'vast' and the 'unprecedented' for example.

Is dflash still free there after the 0731 upgrade?

So what's needed to solve this is that someone publicly and popularly benchmarks using the chat template. So that they have an incentive to look good using it.

Is the chat template used at all when they benchmark the model?

So you happen to know how to demonstrate this higher capacity, what do we mean by that? Can understand bigger codebases? And how do we explain it, what part of the model does it?

I think that's called test-time scaling i.e using more tokens at infer time to squeeze out higher model performance. That's must be part of the explanation for good benchmark results.

I just think, look at how much happens in a generation or two right now. No wonder languages can change quickly.

It's believed that basque is a remnant of earlier people before agriculture and indo europeans arrived.

Any pointers to more info about that? Sounds interesting.

I find it funny and satisfying that Python has converged to it's own "printf", by which I mean print(f"").

Yes, it seems to be a new process, 18A. (18 Ångström, allegedly). Manufactured in North America. If the shiny video they did didn't fool me, it's a two sided wafer process, power grid on one side, then the silicon wafer (with through-silicon vias), then transistors, then signal layers.

The processor is this one: Intel Core 5 processor 320 https://www.intel.com/content/www/us/en/products/sku/246018/... in the Wildcat Lake series.

As a mac user (also a linux user), I got used to it and it always seemed better than the alternative (weird install programs).

Yes, "this one is obsessed with the golden gate bridge" will be for real this time.