Bryan Bischof fka Dr. Donut
@BEBischof
Leading AI @theoryvc; Teach ML @rutgersu. Prev: Led AI @_hex_tech; Led Data @weights_biases, ML @stitchfix, Led Data @bluebottleroast
My book with @eigenhector is out! This is a practitioner focused book on all the technical aspects of building recsys. This is a wide and deep coverage of the kinds of things Stitch Fix and YouTube use in their production recsys. And yes it touches on llms buildingrecsys.com
This is a really exciting idea, and is almost identical to how i think about my personal prompt managment library! (I use jinja instead of @) But this is well articulated and i love the prototype implementation!
New post: SpecTree A simple idea I've found useful for writing more complex LLM prompts and specs, inspired by Markdown + React It's a low-level abstraction for context engineering I'll be building more patterns on in the future
Model providers: This is what I am imagining when you say the model is thinking, and you cant stop me.
watching agents do git surgery really gives you the sense that maybe git isn't that easy after all
Really excited about this. I’ve built about 7 personal llm dev tools and used rich every time, so this feels completely appropriate to me
The Toad is out of the bag! 🛍🐸 Announcing Toad - a universal UI for agentic coding in the terminal willmcgugan.github.io/announcing-toa…
Unbothered. Moisturized. Happy. In My Lane. Focused. Flourishing.
My toxic trait is using 🐊 as the emoji for Claude but never explaining why
For my weekend queue 😍
My talk for @aiDotEngineer on what I think every person working with language models needs to know about GPUs is now available! - Latency lags bandwidth. - GPUs embrace bandwidth. - Don't be scared of N squared. - Use the Tensor Cores, Luke! youtube.com/watch?v=y-UGrY…
I'm so old I wrote that! That's assuming it's the Windows version, which is the one I worked on. The Win9x game, art, and original code, were done by Maxis/Cinematronics. I ported it to Windows NT, converted the x86 asm to C, made it work on RISC, and so on. Success has…
Are you this old?
Who came up with this? Who?
New paper & surprising result. LLMs transmit traits to other models via hidden signals in data. Datasets consisting only of 3-digit numbers can transmit a love for owls, or evil tendencies. 🧵
The hard part of combinatorics problems are the analogies. The formulas aren't super complex, applying the methods aren't particularly difficult, and the total set of abstractions are smaller than many other fields. However, when approaching a new problem in the space you're…
So why is combinatorics so much harder for AI?
"I had to open stack overflow for the first time in years" - My intern complaining about our new design direction