Alex Rubinsteyn

@alexr

personalized cancer immunotherapy = genomics + immunology + machine learning + oncology (pirl.unc.edu)

1.204 Folgt6.084 Follower1.645 Posts

X Bluesky

Ist das dein Profil?Übernimm es — dann gehört dir alexr.dirty.social.Profil übernehmen

Alex Rubinsteyn@alexr· 12d

5. I'm in awe, and excited to see what other things we will learn from the AIs. There are a number of problems I've spent a long time thinking about, and maybe I will learn how to answer them soon.
6. But -- what then? What happens when AI has answered the handful of my favorite problems that I've spent the last 15 years thinking about? There are a lot of nice problems that I like, but it's not so easy to find a favorite problem.
7. There's a lot more to say here, but I think we mathematicians and theoreticians will have our work cut out for us: to keep a leash on these collossi of thought and reasoning, and to keep their abilities intelligible to humankind.

Alex Rubinsteyn@alexr· 12d

4. I am disappointed by the writeup of this proof (sorry Lijie -- I should've taken a look at it earlier!). It writes in a way that's characteristic of a lot of ChatGPT-generated proofs, in which it elaborates at length on "boilerplate" setup, but then nonchalantly introduces what I consider to be the technical crux of the result: the particular Uhlmann transformation/dilation used to "align" all the states together. The technically interesting parts are buried deep into the paper, in Section 4, and introduced without any fuss or fanfare, as if this were the obvious thing to do.

Normally, when I read a paper like this, I get immediately suspicious. What are they trying to hide by not explaining the new ideas up front and center? I wish OpenAI had spent a couple more prompts to clean up the writeup (I haven't had a chance to look at the other writeups so can't comment on their writing quality).

Yes, there's a Lean proof. But that doesn't give me any understanding. That will just take time, I guess.

Alex Rubinsteyn@alexr· 12d

3. Now, to the math. In 2016 I proved a polynomial-decay theorem; the exponential-decay theorem was left open and has since remained one of my favorite problems. I always intended to come back to it. Actually, a month ago I tried to set GPT 5.5 on it, and it didn't make very much progress.

A couple days ago Lijie Chen (@wjmzbmr1
) sent me and a couple others a writeup. Life was busy so I didn't get a chance to look, but I guess now the cat's out of the bag so I probably should opine a bit.

I presume the proof is correct (there's supposedly a Lean formalization, after all), but it will take me some time to digest it. I am gratified that it starts from where my paper left off, but goes beyond the limitations of my proof strategy by using some tricks and techniques that are probably known, in some collective fashion, to operator theorists and functional analysts. The reasoning document furnished by OpenAI is interesting but opaque: it states the problem, and then there's a leap of intuition on how to find the right purification using the "resolvent", and then does some exotic matrix entropy calculations to show that it works out.

I don't understand it yet. Maybe it'll take me an afternoon to check all the calculations, but what would still be missing is _why_ this was an approach that would've made sense in the first place. Is there some broader context or theory within which this would've been the obvious thing to do? What other results can be proven using these techniques? What is it telling us about quantum information or operator theory? I have no idea. I spent about an hour this morning asking ChatGPT these questions, but it's somewhat frustrating because it speaks with a mishmash of physicist, operator algebraist, quantum information theorist-lingo, plus the usual LLM breezy lilt that annoys everybody.

I would love to hear from experts who find these calculations familiar. I might write about this again when I understand better.

Alex Rubinsteyn@alexr· 12d

2. And then -- the kicker -- something that I personally spent a couple years on in grad school, leading to some of my proudest work: quantum parallel repetition theorems. I spent many hours, days, nights, weekends in cafes, in my office, at home, trying to understand Ran Raz's classical parallel repetition theorem and whether I could prove a quantum version of it.

This period of struggle was important for me. I learned a lot of mathematics from it, and most importantly I learned that I could solve hard problems that (some) people cared about. I wouldn't want to trade that experience for anything.

Alex Rubinsteyn@alexr· 12d

For better or worse, X is really the place for most technical topics. I don't think he'd actually get better engagement here but he might post anyways out of idealism (or at least, that's what I do!)

Some initial thoughts, and a complicated mix of feelings.

1. Wow. I mean, Erdos problems are cool (I genuinely mean that), I didn't know about the Jacobian conjecture before it got disproved. But this newest batch from OpenAI hits home in a way the previous announcements did not.

New circuit lower bounds? A simple, easy-to-describe non-sofic group? Hardness of approximation for CVP without needing a unique games-like conjecture? I didn't just hear about these problems from my friends or from seminars. I feel their importance in my bones; I deeply care about the answers to these questions.

Alex Rubinsteyn@alexr· Jul 2

Here’s what the output back to the grid looks like, I guess we sent out 30KWh today. I have never cared much about energy storage before and now it feels like all that matters

Alex Rubinsteyn@alexr· May 15

PIRL symposium today

Alex Rubinsteyn@alexr· Apr 30

Area under Precision-Recall curve for unreleased mhcflurry 2.3.0 affinity sub-model evaluated on held-out mono-allelic mass spec data (95 alleles) as a function of batch size. Vanilla mhcflurry uses batch_size=64 (and lots of other sub-optimal 2010s ML hyperparams)

Alex Rubinsteyn@alexr· Apr 17

Alex Rubinsteyn@alexr· Mar 31

PuppyVaxx AI powered mRNA vaccines made FOR dogs / BY dogs (raising at $1B valuation)

Alex Rubinsteyn@alexr· Mar 30

Previously unfathomable depths of obsessive data curation If you're an MHC sequence, I will find you.

Alex Rubinsteyn@alexr· Mar 19

Broadening the landscape of fusions we’ve talked about today to CIC-DUX4, linking it back to NUT carcinoma via CIC-NUTM1 and comparing to EWSR1-FLI

Alex Rubinsteyn@alexr· Mar 19

Favorite direction that’s come up twice is activation lethality: stabilize the driver oncogene so it accumulates and kills the cancer cell (or at least makes it easier to target) What’s the opposite of a PROTAC?

Alex Rubinsteyn@alexr· Feb 13

Trial was revised many times, some of which look deeply confounding eg v6 to v7 changed trial from observational to interventional and from immunotherapy alone to immunochemotherapy

Alex Rubinsteyn@alexr· Feb 5

Alex Rubinsteyn@alexr· Jan 15

so rare for glamslop to actually get defeated

Alex Rubinsteyn@alexr· Jan 15

Threw an unfinished TCR analysis library and bunch of notebooks with patient-specific functionality into Claude Code and...it generalized the new features perfectly into a very reasonably structured tool. Had a few nitpicks, now they're fixed. Uh, hard to go back.

Alex Rubinsteyn@alexr· Jan 13

Alex Rubinsteyn@alexr· Jan 11

Slow signs that bsky can right its course and fill this currently poorly occupied niche

Alex Rubinsteyn@alexr· Jan 8

Weird

Alex Rubinsteyn@alexr· Dec 30

LLM-as-judge going great (been slowly grading immunology assay designs by different LLMs, was curious if I could just automate that work by providing the prompt, answers of each model, and a grading rubric)

Alex Rubinsteyn@alexr· Nov 25

Got stuck on a few different experiment design queries with ChatGPT 5 Pro Finally got the Gemini app and tried Gemini 3 and…it perfectly answered my question with every relevant experimental detail across a few different papers/patents

Alex Rubinsteyn@alexr· Oct 30

Re #2 -- how much did care improve around administration of ICIs over the past 10 years? Well, if you just look at the survival curves from this paper stratified by start year, you see monotonic increase of survival asymptotes from ~2016-2020

Alex Rubinsteyn@alexr· Oct 30

I think there is still some residual evidence of the claimed interferon effect from the paper but it's small & cohort size isn't big enough to really tease it out confidently. Only with the wrong comparison group can you go from p=0.270 to "published in Nature"

Alex Rubinsteyn@alexr· Oct 30

Anyways, re: #1 / #2 -- the effect of just splitting out early ICI patients from later ones gives you pretty much all the claimed effect (without the gymnastics around using vaccination status to pull out the later patients)

Alex Rubinsteyn@alexr· Sep 26

Tried going back to Claude today and it’s…totally useless for biomed? Got this for many different questions in a row but can eventually get an answer if I write a long shpiel about why it’s really safe to tell me about an IHC CRO.

Alex Rubinsteyn@alexr· Sep 20

Very preliminary figure from Steve Vensko in our lab but still wanted to share it. % melanoma patient samples (n=27) expressing CTA vs. % cells in non-zero samples expressing the CTA

Alex Rubinsteyn@alexr· Sep 18

Is there something wrong with the HLA-II entries of the HLA Ligand Atlas? Looked up the model PRAME epitope A2/SLLQHLIGL and found it's mostly detected on HLA-II alleles -- which doesn't make sense of a teeny little 9mer. (it's also weird to see it so much on healthy samples)

Alex Rubinsteyn@alexr· Sep 18

Hominid re-diversification

Alex Rubinsteyn@alexr· Sep 16

Elementary schools teach non-commutative integer multiplication!?