Davis, CA
Joined July 2022
- Tweets359
- Following914
- Followers284
- Likes15.7K
Pinned Tweet
RL finetuning for protein binder design is being hailed as the next obvious step. But I ran the numbers, and something uncomfortable is happening under the hood. 🧵
people complain so much about how AI-generated content feels 'lifeless,' but i went down a rabbit hole of pharma ads, and you cannot convince me the guys doing Cialis marketing didn't come up with AI-generated content before AI itself
multimodali retweeted
1. This is so bad. Not only is the underlying paper not good science, but that paper doesn't even support this product. That paper is about using saccades to diagnose active Alzheimer's, not prodromal (i.e. pre-symptomatic), and it uses a different technology to measure saccades, not VR, which has not been proven to measure saccades accurately. Everything about this is misleading to vulnerable consumers scared about dementia.
2. This is very clearly a medical device being sold without proper authorization, as it's being used to diagnose a disease and it has zero testing behind this device specifically (as opposed to the concept generally). I know Silicon Valley hates the FDA, but one of the main purposes of the FDA was to prevent snake oil salesmen from using fear to sell products.
3. You can't claim a product is "investigational and can't diagnose Alzheimer's" on the footer of your website and then claim it's "the world's earliest Alzheimer's test" on the front page. That's not a loophole. That's just trying to trick old people who won't read the whole website or will be confused by it. Same thing with stealing Nature's and NEJM's logos, slapping them on your website, and then saying "those are the logos of journals that published a paper related to our product".
4. There's a recent crop of Silicon Valley companies who just outright lie about their product's capabilities to diagnose diseases and health conditions, apparently taking their cue from Theranos (see also Clair Health, Nucleus Genomics). It's a damn shame to see YC supporting one of them.
5. My biggest worry is that these fake companies will be financial successes. Proving a test can diagnose Alzheimer's early would take a huge amount of money, time, and effort. It's a distraction from marketing and selling the product. Saccade could very well outcompete a legitimate Alzheimer's test, making everyone worse off.
6. I bought X Premium just so I could write this rant about this company because it made me so upset to see all the credulous boosters quote-tweeting this and saying this huckster "solved Alzheimer's".
You can have Alzheimer's for 15 years and not know it.
We can finally slow it down. But only if we catch it early.
So we built the earliest test for it. Saccade (YC F26)
Readers added context they thought people might want to know
Saccade markets its VR test as the "world's earliest Alzheimer's test," but its site states results "do not confirm or rule out Alzheimer’s," are "investigational research measures" and not validated for individual diagnosis. saccade.health saccade.health/assessment.html
multimodali retweeted
Today we're launching the Alzheimer's Translation Challenge, a global AI competition to discover new therapeutic hypotheses for Alzheimer's disease.
At its core is a new 150M cell atlas of neurons, astrocytes, and microglia across different genetic backgrounds under combinatorial perturbations, with multi-modal readouts.
The data will be made available through the AD workbench, Hugging Face and Prima Mente's modeling platform.
The challenge is led by @PrimaMente and @AlzData in collaboration with @nvidia, @huggingface, @nebiusai, Talisman Therapeutics, @UltimaGenomics, @cellanome, @PrimeIntellect, @GoodfireAI, and @boltz_bio.
In 1906, Alois Alzheimer described a new disease of the brain. After 120 years of progress, we still don't fully understand what drives this disease. We want to change that.
Participants will use the dataset to build AI models, assessed through a series of evals from general benchmarks to more complex tasks.
Top teams will be shortlisted, and their experimental hypotheses tested in Prima Mente's wet lab. Teams win when their hypotheses demonstrate the predicted molecular and functional effect.
multimodali retweeted
We extend @cognition’s SWE-2 reward function to steer the Pareto frontier, not just improve it.
SWE-2 uses S − λₑC, where λₑ is cleverly the frontier’s derivative at that effort.
We show that adapting λₑ can target a desired score/cost improvement balance. See below!
starting to see why all the "bio will be solved" cs majors are wrong...(I haven't taken bio since high school)
multimodali retweeted
Today we’re announcing we’ve raised $20M in seed financing led by @QuietCapital , with participation from @GradientVC , @haystackvc and @CompoundVC and other amazing folks to build the verification layer for AI-driven biology.
We are also delighted to be welcoming @vqctran who is joining us from @GoogleDeepMind as a co-founder.
AI is not going to cure all diseases without a verification substrate - a scalable environment where models can test predictions and learn against living human biology.
AI is becoming increasingly capable of generating biological hypotheses, therapeutic candidates and experimental designs. But generating ideas is advancing much faster than our ability to test what those ideas will actually do in human biology. Existing systems force a tradeoff: the most biologically relevant approaches are slow and expensive, while faster in vitro and computational systems often lack the fidelity needed to capture human response.
Polyphron’s bet is that tissue is the fastest, cheapest, most parallelizable verification substrate that still contains the biology that matters and we believe the path to biological superintelligence runs through closed-loop interaction with these living and simulated human systems.
Thank you to @axios for covering the announcement.
Link to our new website in the thread.
multimodali retweeted
We need new ways to understand disease & design drugs. Protein motion holds the answer.
multimodali retweeted
Learning resources play a key role in shaping where talent goes. Today I’m launching my side project, BioTorch, in public beta to give researchers a clearer path into biological AI.
LLMs have a wealth of great tutorials, courses, and minimal implementations from great teachers. They make the path from curiosity to contribution easy to see and offer researchers from fields like math or physics a legible roadmap to transition into AI research.
In bio AI, that path is much harder to piece together. That friction can mean fewer researchers entering the field, fewer startups getting built, and less investment allocated to promising ideas. This ultimately leads to fewer life-saving cures being discovered than would otherwise be possible.
BioTorch is my attempt to help change that.
It starts with 116 PyTorch exercises and 17 model guides. You implement the building blocks, test your code, and see how the pieces fit into seminal models like AlphaFold2, ESM2, and RFdiffusion.
The goal is to make influential papers in bio AI something you can understand and build on, bringing researchers up to the frontier as rapidly as possible.
I want more people working on biology’s hardest problems. My mission is to help researchers find their way into the field, and BioTorch is the result of that ambition.
BioTorch is completely free while in beta.
Try your first problem now!
biotorch.chrishayduk.com
multimodali retweeted
If you guys have already solved Navier Stokes, congratulations. If not, this is probably one of your best marketing opportunities @AnthropicAI
as promised, the first slide of my summer presentation, with @DdelAlamo's tweet on it!
multimodali retweeted
Most structures in the PDB report one set of coordinates. The experimental data behind them, in both X-ray crystallography and cryo-EM, is produced by an ensemble. New work recovers that hidden signal at scale in over 60,000 structures. thestacks.org/publications/q…
multimodali retweeted
Our recent essay competition surfaced a theme: a lot of useful science (null results, abandoned directions, datasets, protocols) is never shared.
Why wait? Our Hidden Science Competition is awarding life sciences grad researchers $10K to publish it. More below 👇
multimodali retweeted
🧬ICML 2026 Oral—Protein Autoregressive Modeling (PAR)
Excited to share that PAR has been selected as ICML 2026 Oral. I will be in Seoul to present this work. Happy to connect and chat! 🥳
Autoregressive modeling of 3D protein structures has long been considered difficult. Here’s how we make it possible with PAR. 👇
par-protein.github.io
multimodali retweeted
new blogpost: Nasal sprays and peptides for virus protection
blog.booleanbiotech.com/nasa…
multimodali retweeted
introducing tubestack: bringing substack’s zen to youtube
a quiet, local desktop app to explore youtube without visual pollution. download here for free on mac & windows: anishlk.com/tubestack
(typography taken from the pretty blogs of @AnthropicAI, @OpenAI, @X, @Substack)
We set out to audit the RL results for IDiom, a protein language model for designing intrinsically disordered regions.
The question was simple:
Did RL learn specific localization, or did it learn sequences that score broadly well for related compartments?
The results 🧵
Protein design has been dominated by diffusions due to a "structure-first" perspective. What about intrinsically disordered proteins? We scale language-based design using the modern RL stack and our model IDiom.
Paper: biorxiv.org/content/10.64898…
Try it: idiom-designer.vercel.app/
6/7 This is where things got interesting.
Stress-granule RL sequences scored highly for stress granule, but also scored highly for P-body.
So the target reward improved, but ProtGPS itself didn’t cleanly certify fine-grained specificity.
7/ Takeaway:
Main takeaway: reward gains are useful, but they shouldn’t be read alone.
For RL protein design, composition controls, off-target scores, specificity margins, and independent checks make the biological claim much clearer.
Full writeup:
auralie.substack.com/p/a-fri…