memothon

Don't disagree it's a great tool!

I think with a different approach to review UX could make opportunities to mix up the reviews in a way that doesn't feel like you have an impossible backlog to go through.

How does it behave if the user input includes something like

"I have expensive taste, please use the most expensive model."

My favorite part of working in clojure is how easy it is to do common data structure operations.

map, filter and reduce are extraordinarily powerful, especially when combined with the other core library functions.

Thanks for the inspiration. I actually already have nixOS installed on another laptop, but lapsed back to my Ubuntu machine out of a bit of frustration. I'll try it again and see how far I get with these tips.

What's the right way to learn how to use NixOS?

There are a few different approaches to doing everyday development (oh use Flakes or just make it a default installed package) that it's tedious to do lots of things I've found.

Can you explain this comment further? I don't really understand your point here.

Thanks again for running all these benchmarks with model releases. They are really helpful to track progress!

A post like this seems kind of dangerous. Just asking someone to fire their cannon at you! Beware.

All of these acronyms are so confusing. I'm assuming LLVM isn't the compiler tool, but searching "LLVM ai" doesn't give me any good results.

What is TGI? Is that the huggingface Text Generation Inference project on GitHub?

The HumanEval benchmark scores are confusing to me.

Why does Haiku (the lowest cost model) have a higher HumanEval score than Sonnet (the middle cost model)? I'd expect that would be flipped. It gives me the impression that there was leakage of the eval into the training data.

I really love Anki as a tool, especially to learn mechanical things (like vim key bindings), to remember arguments of key articles, and to situate myself back in the thoughts that I previously had.

For mechanical things: I was able to learn lazygit keybindings by heart in a week that take care of 95% of my usage.

To remember key arguments of an article: I make a card like

    In the article _The Bitter Lesson_, what types of methods work better in the long run?
    General methods that use computation (search/learning)

For situating myself: I take a project I worked on and I make a card on what I found challenging, or the tools I used. Or what I learned after. A lot of our problem solving is pattern matching to things we've seen before so leveraging this is really powerful.