Alex Rubinsteyn
@alexr.bsky.social
personalized cancer immunotherapy = genomics + immunology + machine learning + oncology
(pirl.unc.edu)
(pirl.unc.edu)
Emotionally, going from CS to biomed is rough.
Everything is slower, hierarchical, bureaucratized, entrained in Byzantine incentive structures, pay is bad for important roles bc of a tax on passion, public comms obfuscates behind concern for patients, &c
Everything is slower, hierarchical, bureaucratized, entrained in Byzantine incentive structures, pay is bad for important roles bc of a tax on passion, public comms obfuscates behind concern for patients, &c
February 15, 2026 at 12:00 AM
Emotionally, going from CS to biomed is rough.
Everything is slower, hierarchical, bureaucratized, entrained in Byzantine incentive structures, pay is bad for important roles bc of a tax on passion, public comms obfuscates behind concern for patients, &c
Everything is slower, hierarchical, bureaucratized, entrained in Byzantine incentive structures, pay is bad for important roles bc of a tax on passion, public comms obfuscates behind concern for patients, &c
Reposted by Alex Rubinsteyn
Moralizing AI to the point of refusing to understand it seems like such a political miss.
Biggest realignment of capital / labor / social norms & institutions in a long time…
Biggest realignment of capital / labor / social norms & institutions in a long time…
February 12, 2026 at 2:29 PM
Moralizing AI to the point of refusing to understand it seems like such a political miss.
Biggest realignment of capital / labor / social norms & institutions in a long time…
Biggest realignment of capital / labor / social norms & institutions in a long time…
Is there an AI savvy corner of the lefty political milieu?
Like, obviously this stuff is going to accelerate oligarch-ification, is there anyone publicly thinking about social consequences who has a grounded understanding of the frontier and trajectory?
Like, obviously this stuff is going to accelerate oligarch-ification, is there anyone publicly thinking about social consequences who has a grounded understanding of the frontier and trajectory?
February 12, 2026 at 1:47 AM
Is there an AI savvy corner of the lefty political milieu?
Like, obviously this stuff is going to accelerate oligarch-ification, is there anyone publicly thinking about social consequences who has a grounded understanding of the frontier and trajectory?
Like, obviously this stuff is going to accelerate oligarch-ification, is there anyone publicly thinking about social consequences who has a grounded understanding of the frontier and trajectory?
I left xAI today.
I will never forget the friends I made in the extracorporeal deputrification chamber.
A brief reset and then onwards! 2026 will be the year we finally convert all living biomass into a perfect sphere. 🚀🚀🚀
I will never forget the friends I made in the extracorporeal deputrification chamber.
A brief reset and then onwards! 2026 will be the year we finally convert all living biomass into a perfect sphere. 🚀🚀🚀
February 11, 2026 at 2:30 PM
I left xAI today.
I will never forget the friends I made in the extracorporeal deputrification chamber.
A brief reset and then onwards! 2026 will be the year we finally convert all living biomass into a perfect sphere. 🚀🚀🚀
I will never forget the friends I made in the extracorporeal deputrification chamber.
A brief reset and then onwards! 2026 will be the year we finally convert all living biomass into a perfect sphere. 🚀🚀🚀
Just realized you don’t need DNA to find almost all expressed somatic variants if you have long read scRNA
February 11, 2026 at 1:04 AM
Just realized you don’t need DNA to find almost all expressed somatic variants if you have long read scRNA
Codex kicking off large training jobs on Modal and I feel like I'm ascending into a Sci-Fi hallucination
February 10, 2026 at 4:50 PM
Codex kicking off large training jobs on Modal and I feel like I'm ascending into a Sci-Fi hallucination
I can’t believe we’re speedrunning super-human automated intelligence with offline BPE tokenization
Like…I was sure something better would win.
Like…I was sure something better would win.
February 8, 2026 at 7:05 PM
I can’t believe we’re speedrunning super-human automated intelligence with offline BPE tokenization
Like…I was sure something better would win.
Like…I was sure something better would win.
codex-5.3 caught a totally ruinous copy/paste error in a data download script I wrote by hand before handing off the data to codex-5.2 for simple visualizations
Ran codex-5.3 on a few other repos and it caught small legitimate bugs in each of them. Uh, definitely an improvement
Ran codex-5.3 on a few other repos and it caught small legitimate bugs in each of them. Uh, definitely an improvement
February 6, 2026 at 7:11 PM
codex-5.3 caught a totally ruinous copy/paste error in a data download script I wrote by hand before handing off the data to codex-5.2 for simple visualizations
Ran codex-5.3 on a few other repos and it caught small legitimate bugs in each of them. Uh, definitely an improvement
Ran codex-5.3 on a few other repos and it caught small legitimate bugs in each of them. Uh, definitely an improvement
Worth reading and understanding: blog.escalante.bio/180-lines-of...
~180 lines of code to win the in silico portion of the Adaptyv Nipah binding competition
Here's the script we used to get 1st place in the in silico portion of the Adaptyv Nipah competition:
import modal
def download_boltz2():
from mosaic.models.boltz2 import Boltz2
Boltz2()
...
blog.escalante.bio
February 6, 2026 at 5:31 PM
Worth reading and understanding: blog.escalante.bio/180-lines-of...
Reposted by Alex Rubinsteyn
Our latest research showing PRAME as a previously unrecognized therapeutic target in NUT carcinoma just dropped on JITC. Huge thanks to @sitcancer.bsky.social @benjamingvincent.bsky.social @alexr.bsky.social @unclineberger.bsky.social and so may others!!!
jitc.bmj.com/content/14/2...
jitc.bmj.com/content/14/2...
Shared PRAME epitopes are T-cell targets in NUT carcinoma
Background NUT carcinoma is a rare but highly lethal solid tumor without an effective standard of care. NUT carcinoma is caused by bromodomain-containing NUTM1 fusion oncogenes, most commonly BRD4::NU...
jitc.bmj.com
February 5, 2026 at 1:54 PM
Our latest research showing PRAME as a previously unrecognized therapeutic target in NUT carcinoma just dropped on JITC. Huge thanks to @sitcancer.bsky.social @benjamingvincent.bsky.social @alexr.bsky.social @unclineberger.bsky.social and so may others!!!
jitc.bmj.com/content/14/2...
jitc.bmj.com/content/14/2...
Finally published!
“Shared PRAME epitopes are T- cell
targets in NUT carcinoma”
We stumbled into working on a rare deadly cancer without effective treatments and found that some existing TCR therapies might be worth trying (at least work in vitro)
Link: jitc.bmj.com/content/14/2...
“Shared PRAME epitopes are T- cell
targets in NUT carcinoma”
We stumbled into working on a rare deadly cancer without effective treatments and found that some existing TCR therapies might be worth trying (at least work in vitro)
Link: jitc.bmj.com/content/14/2...
Shared PRAME epitopes are T-cell targets in NUT carcinoma
Background NUT carcinoma is a rare but highly lethal solid tumor without an effective standard of care. NUT carcinoma is caused by bromodomain-containing NUTM1 fusion oncogenes, most commonly BRD4::NU...
jitc.bmj.com
February 5, 2026 at 3:02 PM
Finally published!
“Shared PRAME epitopes are T- cell
targets in NUT carcinoma”
We stumbled into working on a rare deadly cancer without effective treatments and found that some existing TCR therapies might be worth trying (at least work in vitro)
Link: jitc.bmj.com/content/14/2...
“Shared PRAME epitopes are T- cell
targets in NUT carcinoma”
We stumbled into working on a rare deadly cancer without effective treatments and found that some existing TCR therapies might be worth trying (at least work in vitro)
Link: jitc.bmj.com/content/14/2...
Two new NUT carcinoma cell lines: pubmed.ncbi.nlm.nih.gov/41485364/
Establishment of novel stable human sinonasal NUT carcinoma cell lines - PubMed
MDA-NUT87 and MDA-NUT88 are the first stable human sinonasal NUT carcinoma cell lines established from the primary tumor site. They preserve the hallmark genetic and phenotypic characteristics of NUT carcinoma and show sensitivity to BET inhibition. These models represent valuable tools for mechanis …
pubmed.ncbi.nlm.nih.gov
February 4, 2026 at 6:48 PM
Two new NUT carcinoma cell lines: pubmed.ncbi.nlm.nih.gov/41485364/
Ran into a third friend at this Durham co-working space, laptop also full of coding agents running in parallel (four Codex instances)
Three for three now -- split between Claude Code, Cursor, and Codex, I guess I'm the 4th (with 5 Codexes waiting for my attention)
Three for three now -- split between Claude Code, Cursor, and Codex, I guess I'm the 4th (with 5 Codexes waiting for my attention)
February 2, 2026 at 7:26 PM
Ran into a third friend at this Durham co-working space, laptop also full of coding agents running in parallel (four Codex instances)
Three for three now -- split between Claude Code, Cursor, and Codex, I guess I'm the 4th (with 5 Codexes waiting for my attention)
Three for three now -- split between Claude Code, Cursor, and Codex, I guess I'm the 4th (with 5 Codexes waiting for my attention)
New PIRL blog post by @benjamingvincent.bsky.social:
Some advice to companies and clinics starting personalized neoantigen vaccine programs
Link: pirlblog.substack.com/p/some-advic...
Some advice to companies and clinics starting personalized neoantigen vaccine programs
Link: pirlblog.substack.com/p/some-advic...
Some advice to companies and clinics starting personalized neoantigen vaccine programs
The problems aren't all solved yet...
pirlblog.substack.com
February 2, 2026 at 7:25 PM
New PIRL blog post by @benjamingvincent.bsky.social:
Some advice to companies and clinics starting personalized neoantigen vaccine programs
Link: pirlblog.substack.com/p/some-advic...
Some advice to companies and clinics starting personalized neoantigen vaccine programs
Link: pirlblog.substack.com/p/some-advic...
Teaching my kids to sing “Costco! Costco!” over the Dsinghis Khan song
(final humiliation of the Cold War)
(final humiliation of the Cold War)
February 2, 2026 at 2:26 PM
Teaching my kids to sing “Costco! Costco!” over the Dsinghis Khan song
(final humiliation of the Cold War)
(final humiliation of the Cold War)
Disturbance in the force…
Sonnet 4.5 Thinking with search just gave me 6 made up paper links in a row (link either didn’t work or went to different unrelated paper)
Feels like it’s been a while (had to go back to ChatGPT)
Sonnet 4.5 Thinking with search just gave me 6 made up paper links in a row (link either didn’t work or went to different unrelated paper)
Feels like it’s been a while (had to go back to ChatGPT)
February 1, 2026 at 3:51 AM
Disturbance in the force…
Sonnet 4.5 Thinking with search just gave me 6 made up paper links in a row (link either didn’t work or went to different unrelated paper)
Feels like it’s been a while (had to go back to ChatGPT)
Sonnet 4.5 Thinking with search just gave me 6 made up paper links in a row (link either didn’t work or went to different unrelated paper)
Feels like it’s been a while (had to go back to ChatGPT)
Reposted by Alex Rubinsteyn
Read 1000 moltbook posts and then go back to reading bluesky without going insane challenge
January 30, 2026 at 8:20 PM
Read 1000 moltbook posts and then go back to reading bluesky without going insane challenge
Been reading Accelerando just to find out what happens next
January 30, 2026 at 6:11 PM
Been reading Accelerando just to find out what happens next
Raising low tech kids is starting to feel transgressive
(asked about laptop time at a school tour and felt like a fundamentalist)
(asked about laptop time at a school tour and felt like a fundamentalist)
January 28, 2026 at 2:48 AM
Raising low tech kids is starting to feel transgressive
(asked about laptop time at a school tour and felt like a fundamentalist)
(asked about laptop time at a school tour and felt like a fundamentalist)
Reposted by Alex Rubinsteyn
Claude Code did like 95% of the MHCflurry TensorFlow to PyTorch port in about a day but there are some persistent numerical differences/bugs that it's been churning for a week, playing trial-and-error whack-a-mole coding like a junior dev.
Codex seems to be one-shotting the fix
Codex seems to be one-shotting the fix
January 27, 2026 at 6:06 AM
Claude Code did like 95% of the MHCflurry TensorFlow to PyTorch port in about a day but there are some persistent numerical differences/bugs that it's been churning for a week, playing trial-and-error whack-a-mole coding like a junior dev.
Codex seems to be one-shotting the fix
Codex seems to be one-shotting the fix
Codex and Claude Code working on the same branch. What could go wrong?
January 27, 2026 at 3:55 AM
Codex and Claude Code working on the same branch. What could go wrong?
Reposted by Alex Rubinsteyn
Ever wondered what happens if you remove Tfh cells from established germinal centres?
January 26, 2026 at 2:20 AM
Ever wondered what happens if you remove Tfh cells from established germinal centres?