We built a tool for our own internal team at AnswerAI to let us learn, explore, & develop mastery with the help of AI: SolveIt.
Today, we’re releasing it to the public for the 1st time.
https://t.co/PeGT8WWcxF
@823jjd@rbhar90 But if there is a simple button you could press somewhere that means by default it just produces 'maintainable code' then why would you not just turn it on for all your users. If people for some reason wanted unmaintainable code they would then just ask the model for it.
@jeremyphoward@alistairmcleay@LukeVerswey@ericries Happy to hear it's going well! I have to say I miss hearing more from you publicly 😔. I do make sure to check out the https://t.co/kfoZknRaEP gh repo from time to time to see what new tools you guys are working on
🔓🧬First big unlock from vibe science-ing: rapid access to publicly available datasets. Sounds basic. It isn't.
If you've ever tried to pull raw data from a paper you care about, you know: the metadata is a mess, the supplementary tables are unstructured, the file formats don't match, and by the time you've got it working you've lost half a day. For people without strong bioinformatics skills, it's often a dead end entirely. 1/🧵
@Oliver__Hahn Incredible tweet! I too have found huge value in paper associated datasets, but felt the pain of cleaning them up. Nice that this might be over to a large extent.
@ChrisHayduk IMO the problematic part is the generation of good quality data, not the analysis of said data. The costs involved often make experimentation difficult to justify.
@adamlewisgreen Really interesting, thanks for sharing. How did you come across the 2022 statistics paper, and how did you know it was worth building on?
I'm rebuilding AlphaFold2 from scratch in pure PyTorch.
No frameworks on top of PyTorch. No copy-paste from DeepMind's repo. Just nn.Linear, einsum, and the 60-page supplementary paper.
The project is called minAlphaFold2, inspired by Karpathy's minGPT. The idea is simple: AlphaFold2 is one of the most important neural networks ever built, and there should be a version of it that a single person can sit down and read end-to-end in an afternoon.
Where it stands today:
- ~3,500 lines across 9 modules
- Full forward pass works: input embedding → Evoformer → Structure Module → all-atom 3D coordinates
- Every loss function from the paper (FAPE, torsion angles, pLDDT, distogram, structural violations)
- Recycling, templates, extra MSA stack, ensemble averaging — all implemented
- 50 tests passing
- Every module maps 1-to-1 to a numbered algorithm in the AF2 supplement
The Structure Module was the most satisfying part to build. Invariant Point Attention is genuinely beautiful — it does attention in 3D space using local reference frames so the whole thing is SE(3)-equivariant, and the math fits in about 150 lines of PyTorch.
What's next:
- Build the data pipeline (PDB structures + MSA features)
- Write the training loop
- Train on a small set of proteins and see what happens
The repo is public. If you've ever wanted to understand how AlphaFold2 actually works at the level of individual tensor operations, this is meant for you.
Repo: https://t.co/k25vl5th1y
In January, @jonhoo, @jjgort, and I returned to @MIT_CSAIL to teach Missing Semester, a class on topics missing from most CS programs—tools and techniques that everyone should know, like Bash, Git, CI/CD, and AI tools. Today, we’re releasing the course for free online!