Susan Branchett

Bio/Chem is the new Maker Movement, thanks to the AI Scientist revolution. I'm going to be demonstrating DIY "self-driving lab" gear (all open source and makeable for <$150) at @SynBioBeta in San Jose, May 4-7. Here's just a glimpse of a @HuggingFace LeRobot arm with an autosyringe that can autonomously "discover" color theory in minutes. CC @johncumbers

If you've ever worked with computers, you understand the disease- the delight in being able to see how much you can do.

@elorganillero Congratulations

All that work, and all people want is the infant's drawings elorganillero.com/product/yorksh…

Make critical thinking a foundational subject in education. Teach students how to think critically, analyze information, and discern fact from fiction using scientific methods, creating a more discerning and informed society.

Excited to release new repo: nanochat! (it's among the most unhinged I've written). Unlike my earlier similar repo nanoGPT which only covered pretraining, nanochat is a minimal, from scratch, full-stack training/inference pipeline of a simple ChatGPT clone in a single, dependency-minimal codebase. You boot up a cloud GPU box, run a single script and in as little as 4 hours later you can talk to your own LLM in a ChatGPT-like web UI. It weighs ~8,000 lines of imo quite clean code to: - Train the tokenizer using a new Rust implementation - Pretrain a Transformer LLM on FineWeb, evaluate CORE score across a number of metrics - Midtrain on user-assistant conversations from SmolTalk, multiple choice questions, tool use. - SFT, evaluate the chat model on world knowledge multiple choice (ARC-E/C, MMLU), math (GSM8K), code (HumanEval) - RL the model optionally on GSM8K with "GRPO" - Efficient inference the model in an Engine with KV cache, simple prefill/decode, tool use (Python interpreter in a lightweight sandbox), talk to it over CLI or ChatGPT-like WebUI. - Write a single markdown report card, summarizing and gamifying the whole thing. Even for as low as ~$100 in cost (~4 hours on an 8XH100 node), you can train a little ChatGPT clone that you can kind of talk to, and which can write stories/poems, answer simple questions. About ~12 hours surpasses GPT-2 CORE metric. As you further scale up towards ~$1000 (~41.6 hours of training), it quickly becomes a lot more coherent and can solve simple math/code problems and take multiple choice tests. E.g. a depth 30 model trained for 24 hours (this is about equal to FLOPs of GPT-3 Small 125M and 1/1000th of GPT-3) gets into 40s on MMLU and 70s on ARC-Easy, 20s on GSM8K, etc. My goal is to get the full "strong baseline" stack into one cohesive, minimal, readable, hackable, maximally forkable repo. nanochat will be the capstone project of LLM101n (which is still being developed). I think it also has potential to grow into a research harness, or a benchmark, similar to nanoGPT before it. It is by no means finished, tuned or optimized (actually I think there's likely quite a bit of low-hanging fruit), but I think it's at a place where the overall skeleton is ok enough that it can go up on GitHub where all the parts of it can be improved. Link to repo and a detailed walkthrough of the nanochat speedrun is in the reply.

Excited to share our @NatureComm work on 28 electrochemically exfoliated 2D materials, showing how stiffness anisotropy guides the formation of high-AR ultrathin nanosheets—key for flexible electronics. nature.com/articles/s4146… #2Dmaterials #FlexibleElectronics

If you think you understand it, try building it.

Come and sing choruses from Brahms's German Requiem with Leeds Festival Chorus this Saturday at St George's Leeds!

Come and see me on July 12th playing the trombone (King&I, Kirkstall Festival, Ruins Stage, 12:30) and the kazoo (Leeds Festival Chorus, St Edmund's Roundhay, 19:30) trybooking.com/uk/events/land…

"Hardware eventually fails. Software eventually works." — Michael Hartung

This MIT CS class teaches you things that all the other classes don't teach you, like... 🖥️ Shell tools and scripting 🖥️ Vim 🖥️ Data wrangling 🖥️ Command-line environment 🖥️ Version control Watch all 11 lectures for free here: bit.ly/MissSemester

Carrier multiplication below the bandgap? Yes! ⚡🔬 Our @ACSNano study shows efficient CM in 2H-MoSe₂, even below the threshold—unlocking new potential for solar energy! ☀️🚀 📖 Read: pubs.acs.org/doi/10.1021/ac… #CarrierMultiplication #UltrafastSpectroscopy #2DMaterials @tudelft

It took a third of a century, but today a dream came true: pubs.acs.org/doi/10.1021/ac…

🎊Lars Bannenberg has received funding from the Open Technology Programme within NWO's Applied and Technical Sciences domain for his research, ‘Protective coatings for optical #Hydrogen sensors to enable enhanced electrolyser and fuel cell efficiency'. 👉edu.nl/8dxtu

70 years ago this month, researchers ran the first FORTRAN program, considered the beginning of general-purpose programming: bit.ly/fortran7 Image v/@NASA

Our CodeRefinery workshop starts in one week with an introduction to version control with git and GitHub. You can still register to join online, but if you are in Espoo, FI or Oslo, NO you can also join in person: coderefinery.github.io/2024-09-10-wor…

#Otd in 1955, the term “artificial intelligence” was coined in a conference proposal: stanford.io/2WJJJGN

🚀 Excited to announce the first #Amsterdam #Julia Language #Meetup! 🎉 Join us on Sept 26 at @Xebia for networking, talks, and fun. Learn from Julia experts like Martijn Visser and Tom Lemmens. Please RSVP to help us order enough pizza! 🍕 👉 More info: meetup.com/julia-lang-ams…

In June, @EvaMemmel from @TUDelftME, won the prize for her poster “Position: Tensor Networks are a Valuable Asset for Green AI” in the #MachineLearning and Foundational #AItechniques category at our PhD poster event. You can see her poster and pitch here: youtu.be/VtIS54SZNm4