Rémi 〰️(@remilouf) 's Twitter Profileg
Rémi 〰️

@remilouf

LLMs & structured generation @dottxtai. @OutlinesOSS 〰️ . Alumni @ENS_ULM & @UniOfOxford. I wander.

ID:35468732

linkhttp://dottxt.co calendar_today26-04-2009 13:56:14

4,2K Tweets

5,9K Followers

1,2K Following

Clémentine Fourrier 🍊(@clefourrier) 's Twitter Profile Photo

I discovered at ICLR 2024 that a lot of what I take for granted about LLM evaluation is actually not that widely known...

So I made a blog!
- how do we do currently do LLM evaluation? ⚖️
- most importantly, what is it actually useful for? 🤔

huggingface.co/blog/clefourri…

account_circle
Erik Meijer(@headinthebox) 's Twitter Profile Photo

As the stragglers are still heatedly debating F# vs C# vs OCaml vs Haskell, the language designers github.com/githubnext/cop… have already moved on to the next generation of AI-based programming languages githubnext.com/projects/specl….

Skate towards where the puck is going, not where it

account_circle
Paul Simmering(@paul_simmering) 's Twitter Profile Photo

There are 10 Python libraries for structured LLM output. I've tried them all. My top recommendations: outlines and instructor. Read my full review here:
simmering.dev/blog/structure…

account_circle
Rémi 〰️(@remilouf) 's Twitter Profile Photo

The urge to write a blog post about the future of interacting with auto-regressive language models when I have to do a bunch of admin.

account_circle
Rémi 〰️(@remilouf) 's Twitter Profile Photo

Using an audio transcription tool for my calls and, rereading one, I see that I have a tendency to go on a tangent inside a tangent inside a tangent and go back to the main point until the next nested tangent. I feel sorry for everyone who’s had to go through this.

account_circle
Rémi 〰️(@remilouf) 's Twitter Profile Photo

I see a lot of 'delve' in documentation these days and if that means that there would otherwise be little to no documentation I’m all for it.

account_circle
Will Kurt(@willkurt) 's Twitter Profile Photo

Working with closed APIs, all you can do is 'prompt hacking' and hope for improvements.

With open models, you have so much more control over what the model is doing and how performance can be improved. You can do actual engineering.

This is key to unlocking the largest gains.

account_circle