Haha yeah we were looking for family holiday movies, I noped out of that list pretty quick.

But temperature is a thing for a reason.. some level of non-determinism makes them MORE useful in a lot of cases. So perhaps we can trade determinism for being less useful.. is that still providing the business value though?

The answer as always is probably.. it depends. 😶‍🌫️

Aren’t there also cases like temperature support being removed from (at least some GPT-5 models)?

Is it? thinkingmachines.ai/blog/defeati...

To be fair Reimer was the only reason there was even a game 7 in that series. Probably should’ve been over in 5.

Cool, wasn’t sure if you might be trying bazzite.gg - lotta people seem to like it for gaming

What distro are you running?

github.com/kcp-dev/kcp

Wtf only 6 hockey teams? 😆

remember when AI couldn’t count the number of Rs in “strawberry”? that was eight months ago

Wow.. bsky.app/profile/kjhe...

Do you have the source on this?

Join the open source pledge: opensourcepledge.com

It also follows immediately after the next biggest hype wave (crypto / web3) and seems like a bunch of hucksters just jumped from one to the other.

www.theverge.com/2024/4/2/241...

I’ve seen Matthews score goals in almost every way I thought possible. Pretty sure this is the first diving poke check of a goal!

If I’m remembering you and John had some pretty hard stances on preventing training of LLMs on your content, even if that cat seems to be out of the bag already. The code that those LLMs were trained on was also mostly not published for that use (SO responses, open source repos on Github, etc. )

Honest question: is there any cognitive dissonance with your view of LLMs used for writing and your use of LLMs for generating code?

You even mention that when copy/pasting code you would credit the person on stack overflow (or whatever) but the LLMs are basically trained on the same data…