If you're going #CUDAmode, next time you take a break and turn the lights back on, read this awesome blog post series on the development of ispc, the equivalent "C-but-SPMD" tool for acceleration with Intel CPU vector units. h/t Horace He for the rec! pharr.org/matt/blog/2018…

engineers will talk your ear off about Amdahl's Law as it applies to optimizing physical systems then say things like "we're accelerating science by making this algorithm go faster" with zero irony

Slashing down CI run time is underrated.

TFW you and your boy provoke an angry letter to the editor, but that letter is from Richard Bellman, who also literally works down the hallway from you.

this too shall pass@k

The ethical way to SaaS is to be constantly teaching your users how things work under the hood while also offering a product that doesn't *require* them to understand. This also happens to be the highest quality marketing.

I'll be presenting twitter-95.com, my all-LLaMAs social network from 1995, at LLaMAPALOOZA this Friday! Come thru. lu.ma/k3ttp1ey

my coworker is totally killing it

me going from undergrad to grad school to industry

serverless pgSQL 🤝 serverless GPUs

Him

Postgres 17 has been stamped, and per tradition will be out on Thursday. Get excited! I’m already seeing some cool benchmarks. Which features or improvements are you most excited about?

Tomorrow Charles 🎉 Frye and I will be presenting on how to orchestrate Modal in Dagster. Here's a sneak peak of the high-level architecture!

brilliantly crisp analogy! looking forward to reading this one

LLMs have made exciting progress on hard tasks! But they still struggle to analyze complex, unstructured documents (including today's Gemini 1.5 Pro 002). We (UC Berkeley) built 📜DocETL, an open-source, low-code system for LLM-powered data processing: data-people-group.github.io/blogs/2024/09/…

Attending the CUDA MODE hackathon last Saturday was such a pleasure. So many amazing people with great hacks. It is a very powerful community, where you can learn and practice all the difficult CUDA and systems programming. I was amazed by all the talent! GPU MODE ON 😎

excited to see arena-style rating systems for more models!

Come to our SF office tomorrow, the 25th at 5:30 PM PT. Why? Airtrain AI, OpenPipe & Modal will be here too. Emmanuel Turlay Kyle Corbitt Charles 🎉 Frye plan to keep you entertained. Get your name on the list. Luma in the replies.