theanonymousone

Third-party inference providers will have a hard time to beat Luna in pricing with comparable open models.

Yes, it seems they have a template that they fill with numbers. Similar issues spotted on Grok's performance page: https://news.ycombinator.com/item?id=49789558

True. That's also a big ambiguity to me. But for some reason I never cared about it as much :/

The most problematic for me as a non-native has been "the" vs "nothing", i.e. when one can/should put the definite article.

Pre-LLM I was joking that real AI is whatever can make that decision correctly 100% of times.

Is it only me, or this Jev thing really gives Claw vibes?

So, the field was making a joke of someone's dementia?

I don't want my Scrum Master back, although it never went anywhere in the first place, unfortunately.

Yes

It may be worth noting that this is not _that_ KVM, but this one: https://en.wikipedia.org/wiki/KVM_switch

I learnt this first myself from HN, a few months ago.

This "flow typing" looks very intriguing to me: So, in s sense, what we know as "dynamic typing" is more precisely describable as "runtime" (not compile-type) dynamic typing?

Java and Python are not on that list?

For technical report: https://news.ycombinator.com/item?id=49639110

I came across AuthaGraph projection some time ago and got hooked to it: https://en.wikipedia.org/wiki/AuthaGraph_projection

A version of it including political borders my desktop wallpaper for a long time, until I decided to not change default wallpaper on my new devices anymore.

Same. (but of course, you don't see the duplicates when they do accept your post)

You can also see duplicates e.g. here: https://news.ycombinator.com/from?site=semianalysis.com

How come the same article has been posted four times in the past 18 days? https://news.ycombinator.com/from?site=runjs.app

Doesn't HN have a temporal deduplication system?

Well, the "limit" is disassembled/obfuscated code, isn't it?

Transformers.js supports a lot of models since v4

httpy is actually twofold cool for a Python HTTP client!

The joke was a good one the first, second, or third time. It's not even funny anymore...

up to 46 times, you mean?

I hope NYT lawyers don't come for you. They did for many other variants of this. Maybe reconsider the name?

That's my questions as well. DeepSeek v4 0731 is served dirt cheap and it needs 10 times more RAM.

In my personal anecdotes, Luna has changed calculations, once again; I'm _almost_ unconstrained in spending (my employer's Copilot subscription) tokens as long as its Luna. And it works more than good enough.

I'm wondering whether any provider can offer this for cheaper $/token than the new DSv4 Flash, which is both cheaper and smarter :/

Completely local use is a different story, of course.

Thanks. And do you know what it has to do with Flutter these days actually? There is barely any mention of Flutter in the website etc :/

Are you using Flet? If yes do you think it's mature enough and has a potential to stay long "in the game"?

Do we know if AA and DeepSWE benchmarks are on bf16 or fp8 quantisations?

I believe that will put them in the Pareto frontier.

But I cannot find this "variant" in OpenRouter.

Then it wouldn't be "SearXNG in Rust", would it?

It's not a metadata search engine, it's "meta" search engine.

Nice signal for lack of dominant AI use, it seems :D