Lutra is constantly learning and improving! 🌟

If you’re curious and want to experience it firsthand, reply to here or send me a DM. We’d love to get your feedback and see how Lutra can assist you! 🚀

Erik Demaine has a cool collaborative whiteboard project on Github that you can run locally.
It has a cool 'time travel' feature where you can rewind back to an older state of the whiteboard!
github.com/edemaine/cocre…

If you’re developing agents, which framework are you using — PDCA (plan do check act) or OODA (observe orient decide act) or YOLO (just execute! no plans)?

Introducing RepoQA for evaluating LLMs’ repository understanding!

🌐 Leaderboard of 25+ models: evalplus.github.io/repoqa.html
⚙️ GitHub: github.com/evalplus/repoqa
🎨 Supporting 5 programming languages (more coming soon)
🚀 Evals openai/vllm/anthropic/HF/gemini models in one command!

🧵

you wouldn’t last an hour in the asylum where they raised me

Excited to share our work at Google DeepMind!

We propose Naturalized Execution Tuning (NExT), a self-training method that drastically improves the LLM's ability to reason about code execution, by learning to inspect execution traces and generate chain-of-thought rationales 🧵👇

More work from Google on AI for SWE, here automatically fixing build errors! The cool thing about fixing builds is you can check if the build succeeds before showing the user the fix. Results in a measurable shortening of code submission time too!

A GitHub flaw lets attackers upload executables that appear to be hosted on a company's official repo, such as Microsoft's—without the repo owner knowing anything about it.

The following URLs, for example, make it seem like these ZIPs are present on Microsoft's source code repo:

Is there a piece of locally running software on Apple silicon MacBook which is a combination of what Nvidia broadcast and Otter AI does, ie, locally cancels noise in outgoing audio and transcribes audio in outgoing and incoming audio?

Bonus points if it also generates insights

Question about the 'Infini-attention' paper:
Does anyone have good intuition about why their idea of aggregating memory across segments works (with much lower memory consumption compared to memorizing transformer)?
arxiv.org/pdf/2404.07143…

Feels great to see my friends and teammates get promoted after all the hard-work they put in :)