HackerTrans
TopNewTrendsCommentsPastAskShowJobs

remilouf

no profile record

Submissions

[untitled]

1 points·by remilouf·il y a 2 mois·0 comments

Tool calls that execute 100% of the time

blog.dottxt.ai
5 points·by remilouf·il y a 3 mois·0 comments

[untitled]

1 points·by remilouf·il y a 3 mois·0 comments

[untitled]

1 points·by remilouf·il y a 3 mois·0 comments

The M×N problem of tool calling and open-source models

thetypicalset.com
159 points·by remilouf·il y a 3 mois·50 comments

I run my company from Emacs

thetypicalset.com
20 points·by remilouf·il y a 3 mois·1 comments

Every AI Integration Is Held Together with Parsing Logic and Prayer

blog.dottxt.co
2 points·by remilouf·l’année dernière·0 comments

[untitled]

1 points·by remilouf·il y a 2 ans·0 comments

[untitled]

1 points·by remilouf·il y a 2 ans·0 comments

[untitled]

1 points·by remilouf·il y a 2 ans·0 comments

Tokenization Is a Problem for LLMs

blog.dottxt.co
2 points·by remilouf·il y a 2 ans·1 comments

[untitled]

1 points·by remilouf·il y a 3 ans·0 comments

[untitled]

1 points·by remilouf·il y a 3 ans·0 comments

[untitled]

1 points·by remilouf·il y a 3 ans·0 comments

Show HN: Vectorize OpenAI API calls (like NumPy)

github.com
3 points·by remilouf·il y a 3 ans·0 comments

comments

remilouf
·il y a 2 mois·discuss
Rémi here, really cool! It kind of turned into a rabbit hole on my end: https://github.com/rlouf/sigil
remilouf
·il y a 2 mois·discuss
Of course: https://github.com/rlouf/sigil
remilouf
·il y a 2 mois·discuss
It was indeed inspired from my IRC days :)
remilouf
·il y a 2 mois·discuss
Original author here, the project has evolved quite a bit since then, you can follow here if that interests you: https://github.com/rlouf/sigil

(The $7k was sarcasm)
remilouf
·il y a 2 mois·discuss
Author here. Sorry my writing is tedious. Next time I’ll use AI to make it more readable.
remilouf
·il y a 3 mois·discuss
> Ironically LLMs solve the MxN problem he's complaining about

Enlighten me please
remilouf
·il y a 3 mois·discuss
Ooops sorry
remilouf
·il y a 3 mois·discuss
Author here. You're right, it's not a hard problem, but a particularly annoying one.
remilouf
·il y a 3 mois·discuss
I haven't always done this, and the knowledge base used to visibly degrade over time. Reviewing a PR does not take a long time, maybe a few minutes, and this compounds over time.
remilouf
·il y a 2 ans·discuss
This is actually pretty funny.
remilouf
·il y a 2 ans·discuss
That’d be a pretty inefficient way to generate bullshit at scale
remilouf
·il y a 2 ans·discuss
LLM evaluations are very sensitive to the details of the prompt's structure. This post shows how using structured generation reduces the results' variance and the ranking shifts.
remilouf
·il y a 2 ans·discuss
Looks like it’s quite the opposite: http://blog.dottxt.co/performance-gsm8k.html
remilouf
·il y a 2 ans·discuss
What do you mean by "semantic dimension"?
remilouf
·il y a 2 ans·discuss
That whole structured generation line of work looks promising. I hope someone else takes this and runs evaluations on other benchmarks. Curious to see if the results translate!
remilouf
·il y a 2 ans·discuss
Awesome work! I am really impressed by how much structured generation improves model performance.
remilouf
·il y a 2 ans·discuss
This article presents a way to make structured generation with LLMs much faster than standard generation, but what I find most interesting is how it highlights the issues that tokenization entails towards the end.
remilouf
·il y a 3 ans·discuss
We already support regex-guided generation in the library, and could easily make an API to serve this as well if that's a feature people want!