tmalsburg2

Opus models after 4.8 didn’t work well with my homegrown harness, so I just skipped them until 5.5 which seems like a pretty happy match.

Nice!

> When context gets too long, maki compacts history automatically: strips images, thinking blocks, and summarizes older turns.

Don’t the summaries of older turns effectively invalidate the context cache, such that you consume less tokens but more expensive tokens?

I wrote my own harness in Emacs and it’s completely ridiculous how well it works. Auto-compact is the only missing feature on my list. Claude‘s approach, if I understand it correctly, invalidates a lot of cached context, and I‘m thinking about a more cache-friendly strategy.

Humans will not regurgitate longer segments of code verbatim. Even if we wanted to, we couldn’t do it because our memory doesn’t work that way. LLM on the other hand can totally do that, and there’s nothing you can do to prevent it.

Org-mode FTW

If the censoring is at the DNS level, can the admin please replace the domain name in the url with the ip address to which it should resolve? Thank you.

It actually interesting that Carlsen (likely the best classical player of all time) hasn’t overfit classical chess to the point where it hurts his ability to play other variants.

Sure, the market chose Markdown, but this simply led me to the conclusion that the market isn’t worth following. Of course the mismatch creates some friction, but the benefits of org-mode, for me personally, easily outweigh that.

Sounds smart but it’s abusing the semantics of “refine” and is therefore ultimately vacuous.

What problem is this trying to solve and does it actually succeed at solving it? I‘m struggling to see the appeal given that the JS still needs to model the internal structure of the template in order to fill the slots.

Also inflation.

It’s do easy to come up with serious non-entertainment examples, I‘m sure you don’t need my help finding them.

How often do you go back an rewatch earlier parts of a movie? I hardly ever do this. In the cinema, theater, or when listening to the radio it’s simply impossible and it still works.

Yeah, that's confusing, but the HuggingFace page says it has 9.78 B parameters.

https://huggingface.co/ibm-ai-platform/Bamba-9B-fp8

Isn't this exactly the point of this model? No need to memorize everything (which makes transfomers expensive), just keep the relevant info. SSM are essentially recurrent models.

OP is talking about information entropy. Nature isn't relevant there.

What would even be the right hands in this case? Seems like almost any hands would be the wrong hands.

The tech report doesn't say on which languages it was trained. The huggingface page says 140 languages but has no list, as far as I can see. :/

HN removed the first word of the title (“How”). I tried to delete the submission, but it didn’t work. @dang, is it possible that there’s a bug here?

The description is pretty abstract. What concrete problem does this package solve? Some example uses would be helpful.

The only thing that's more ridiculous than an antiquated dress code is an antiquated dress code that's selectively applied to some players and not to others. Other players had to change, so FIDE could hardly make an exception for Carlsen. If Carlsen cares so little about the tournament that he can't even be bothered to change his pants, he should probably not have participated in the first place.

> Though for law, I think some ambiguity is beneficial.

Ambiguity means that there are two or more possible interpretations and it's not clear which of them is intended. That's hardly useful. What's beneficial, and what you perhaps had in mind, is some amount of under-specification where the meaning is clear but leaves gaps to be filled in by judges.

The vast majority of accounts that followed me on Twitter over the last year are obvious bots.

Why not use temperature 0 for sampling? If the top-ranked move is not legal, it can’t play chess.

Why has this been flagged? Great material, even Germans will be able to learn from this.

Very roughly 1/3 of the speed of light. One third is lost to the physical medium it seems. What accounts of the remaining third and how much could be plausibly shaved off of it?