Astral founder Charlie Marsh used OpenAI's Codex to find a Ruff formatter optimization yielding a 6.4% speedup · Digg

Astral founder Charlie Marsh used OpenAI's Codex to find a Ruff formatter optimization yielding a 6.4% speedup · Digg

Posts from X

Most Activity

VIEWS1.3KBOOKMARKS5LIKES16

Vaibhav (VB) Srivastav@reach_vb

Codex for autonomously optimising your work:

Charlie Marsh@charliermarsh

It's still incredible to me that you can just point the model at things and make them better continuously.

Codex found an easy 5-10% speed-up in the formatter while I was doing other work.

3h1.3K165

RETWEETS1

Cam@Cameron_C2

Thanks Charlie for the tokens - this optimization has sped up oxc's formatter by ~5% as well!

Charlie Marsh@charliermarsh

I asked it to try again and it found another easy 5% speed-up.

2h11910

REPLIES3

Charlie Marsh@charliermarsh

I asked it to try again and it found another easy 5% speed-up.

Charlie Marsh@charliermarsh

It's still incredible to me that you can just point the model at things and make them better continuously.

Codex found an easy 5-10% speed-up in the formatter while I was doing other work.

3h40070

Charlie Marsh@charliermarsh

https://github.com/astral-sh/ruff/pull/26344

https://github.com/astral-sh/ruff/pull/26429

Charlie Marsh@charliermarsh

I asked it to try again and it found another easy 5% speed-up.

3h28220

Hira@Hiraweb3

@charliermarsh Auto-optimizing code while you scroll Twitter? That’s just unfair.

3h682

Cody Blakeney@code_star

I’ve recently taken to finding my major bottlenecks when processing large datasets with spark and having codex write the heavy inner worker logic in rust.

Whenever I have spare time I just let it cook on improving the speed. I pretty frequently get little 1-2% speed ups while I’m responding to slack or in meetings. (On top of the 2-4x speed up from the original rust port)

All it costs is a little laptop battery and my codex rate limit resets I’ve been collecting.

Cody Blakeney@code_star

Agents can just do things

20m11140

Charlie Marsh@charliermarsh

This code probably runs over, I don't know, billions of files per day?

Charlie Marsh@charliermarsh

I asked it to try again and it found another easy 5% speed-up.

3h7620

Vaibhav (VB) Srivastav@reach_vb

@charliermarsh at this point you should give your codex a promotion and a nice beefy machine to do what needs doing!

Charlie Marsh@charliermarsh

It's still incredible to me that you can just point the model at things and make them better continuously.

Codex found an easy 5-10% speed-up in the formatter while I was doing other work.

3h14410

Cody Blakeney@code_star

Agents can just do things

Charlie Marsh@charliermarsh

It's still incredible to me that you can just point the model at things and make them better continuously.

Codex found an easy 5-10% speed-up in the formatter while I was doing other work.

25m2320

Charlie Marsh@charliermarsh

@Cameron_C2 Love it

2h421

Senior Spreadsheet Engineer@merm_bot

@charliermarsh We live in an age of wonders

3h18

Rehan Momin@RehanMonline

@reach_vb I would love to do a openai hackathon in my city any support @reach_vb

3h16

Cam@Cameron_C2

https://github.com/oxc-project/oxc/pull/23982 (ignore the codspeed benchmarks showing no diff, jsdelivr is down which is causing us issuees)

2h14

itchy@itchy_est

@charliermarsh If you /goal it could you optimize more?

3h3

Strata@ChainZenit

@code_star it is wild how quickly they are getting agents to ship

21m1

Strata@ChainZenit

@code_star that is such a cool way to optimize your workflow.

5m