Aman Karmani (@tmm1) 's Twitter Profile
Aman Karmani

@tmm1

building Cursor @anysphere. full stack tinkerer and perf nerd. formerly vp of infra @github + ruby-core committer. founder @getchannels + ffmpeg committer.

ID: 15591045

linkhttp://github.com/tmm1 calendar_today24-07-2008 23:41:00

5,5K Tweet

8,8K Followers

4,4K Following

Birchlabs (@birchlabs) 's Twitter Profile Photo

pytorch 2.6 is out! highlights: - flex attention: better compilation of blockmask creation, better support for dynamic shapes - cuDNN SDPA: fixes for memory layout - CUDA 12.6 - python 3.13 - MaskedTensor memory leak fix

pytorch 2.6 is out!
highlights:
- flex attention: better compilation of blockmask creation, better support for dynamic shapes
- cuDNN SDPA: fixes for memory layout
- CUDA 12.6
- python 3.13
- MaskedTensor memory leak fix
Kanishk Gandhi (@gandhikanishk) 's Twitter Profile Photo

2/13 We identify 4 key cognitive behaviors that enable successful learning: Verification (checking work), Backtracking (trying new approaches), Subgoal Setting, and Backward Chaining (working backwards from a goal). Qwen naturally exhibits these, while Llama mostly lacks them.

2/13 We identify 4 key cognitive behaviors that enable successful learning: Verification (checking work), Backtracking (trying new approaches), Subgoal Setting, and Backward Chaining (working backwards from a goal). Qwen naturally exhibits these, while Llama mostly lacks them.
Benjamin F Spector (@bfspector) 's Twitter Profile Photo

Aaryan Singhal Dan Fu hazyresearch (3/7) How does ThunderMLA achieve this boost? By applying classical computing concepts (job scheduling, interpreters) to modern AI inference. Our megakernel is essentially an interpreter running a tiny virtual instruction set directly on the GPU. 🚀

Jarred Sumner (@jarredsumner) 's Twitter Profile Photo

> "Half of [typescript-go's performance gain] comes from shared-memory concurrency" JavaScript needs threads. Worker isn't good enough.

Cursor (@cursor_ai) 's Twitter Profile Photo

We've shipped several quality-of-life improvements to Cursor's UI! We know the little bits of polish and delight matter lots in a tool that you use every day. Details below...

We've shipped several quality-of-life improvements to Cursor's UI!

We know the little bits of polish and delight matter lots in a tool that you use every day. Details below...
Cheng (@zcbenz) 's Twitter Profile Photo

It is wild that with GPU programing you start measuring performance in units of µs. I wrote some notes on how I tracked down some of my suboptimal code that delayed the kernel execution for 40µs but made whole program 4x slower. github.com/ml-explore/mlx…

Darshan Sen (@raisinten) 's Twitter Profile Photo

With this change, Node.js LTS v22 will let you profile both JavaScript and native frames in your Node.js app by using samply! github.com/nodejs/node/pu…

With this change, <a href="/nodejs/">Node.js</a> LTS v22 will let you profile both JavaScript and native frames in your Node.js app by using samply!

github.com/nodejs/node/pu…
CUE (@cue_lang) 's Twitter Profile Photo

📣 We have just released CUE v0.13.0-rc.1 - with further fixes to the new evaluator, and speeding up the CLI with many inputs 🎉 github.com/cue-lang/cue/r… We plan to do v0.13.0 next week - test out this release and report any issues with the new evaluator, now on by default!

TypeScript (@typescript) 's Twitter Profile Photo

Today we're announcing previews of the native port of TypeScript! These nightly previews will be available on npm and the VS Code marketplace! npm install -D @​typescript/native-preview npx tsgo --version Read up more on our blog! devblogs.microsoft.com/typescript/ann…

Cursor (@cursor_ai) 's Twitter Profile Photo

Cursor 1.0 is out now! Cursor can now review your code, remember its mistakes, and work on dozens of tasks in the background.

Zeno Rocha (@zenorocha) 's Twitter Profile Photo

I was in a car, and didn't have my computer with me. I could have asked someone on the team to take a look, but why not try to solve it myself? so I fired up the Cursor for Slack, and asked it to fix the issue.

I was in a car, and didn't have my computer with me.

I could have asked someone on the team to take a look, but why not try to solve it myself?

so I fired up the <a href="/cursor_ai/">Cursor</a> for Slack, and asked it to fix the issue.