Kyle Roh

184 posts

Kyle Roh

Kyle Roh

@kyle_r0k

@stanford kid doing AI and biochem @stanfordmed @stanfordailab

Katılım Eylül 2021
213 Takip Edilen92 Takipçiler
Kyle Roh retweetledi
Beff (e/acc)
Beff (e/acc)@beffjezos·
Reminder that there are people with 160-190 IQ who actually intellectually mog Mythos (which probably takes a whole Rubin pod to run), while their brain runs on ~20W. We aren't being bullish enough on the biological substrate. Humanity is still early. It's time to bio/acc
English
157
110
2.4K
81.6K
Kyle Roh retweetledi
Arcesilaos
Arcesilaos@ArcesilaosRes·
It’s a good list but it’s quite telling that @MartinShkreli names capital allocators first and companies second. Why leave out the true GOATs - drug hunters? Here is my list: 1. Paul Janssen (haloperidol, fentanyl, risperidone, loperamide,…) 2. Leo Sternbach (basically all benzodiazepines) 3. Getrtude Elion and G Hitchings: Allupurinol and Acyclovir 4. Akira Endo (Early Statins) 5. Greg Winter (Phage display, basis of dozens of mAbs)
Martin Shkreli@MartinShkreli

pharma GOATs (includes their teammates): Wong Edelman Holman Aghazadeh Rothbaum Chen Baker Ramaswamy you want to ask the question "does a drug work" or "will a drug work", ask the people who have made the most money answering said question. don't ask an influencer

English
5
4
113
31.1K
Kyle Roh
Kyle Roh@kyle_r0k·
SpaceX is undervalued
English
2
0
0
199
MTS
MTS@MTSlive·
SITUATION DETECTED: Google DeepMind has released DiffusionGemma, an open model that generates entire blocks of text simultaneously rather than word-by-word, achieving up to 4x faster output on dedicated GPUs.
English
5
30
506
15.8K
Kyle Roh
Kyle Roh@kyle_r0k·
@bravo_abad Of course — nature never writes code sequentially
English
0
0
1
409
Jorge Bravo Abad
Jorge Bravo Abad@bravo_abad·
No scaling laws for single-cell foundation models: when bigger atlases stop teaching the model anything In language and vision, the recipe has been simple: more data, bigger models, better performance. Single-cell biology borrowed that playbook. Foundation models for transcriptomics jumped from 1 million cells to atlases of over 100 million, on the assumption that scale would unlock the same gains. Alan DenAdel and coauthors put that assumption to the test, and the result is sobering. Working from a 22.2-million-cell corpus, they pretrained 400 models across five architectures (from PCA and a variational autoencoder up to the Geneformer transformer) and ran 6,400 evaluation experiments. They varied not just dataset size (1% to 75%) but also diversity, using cell-type re-weighting and geometric sketching to deliberately enrich rare cell types and transcriptional states. The finding: performance saturates almost immediately. On cell-type classification, batch integration, and perturbation prediction, most models hit their ceiling at roughly 1% of the corpus, about 200,000 cells. Beyond that, adding millions more cells changed essentially nothing. More diversity didn't help. Even spiking in genome-scale Perturb-seq data, to give the models perturbed phenotypes rather than just healthy ones, failed to move the needle. Larger models did score better overall, but they too plateaued early on data. Two points stood out. Simple baselines (PCA, logistic regression) often matched or beat the transformers. And the strongest model, SCimilarity, won not because of size but because its contrastive training objective is aligned with the downstream task. For single-cell data, what you train on and how you frame the objective matters far more than how much you collect. This reframes a quiet but expensive habit. In drug discovery, biotech, and any pipeline leaning on cell atlases, the instinct to keep scaling pretraining corpora may be burning compute for no return. The real leverage sits elsewhere: curating high-quality, task-relevant data and matching the training objective to the actual question you're trying to answer. Paper: DenAdel et al., journal license | doi.org/10.1038/s41592…
Jorge Bravo Abad tweet media
English
15
93
384
95.8K
Adam Draper ⏻
Adam Draper ⏻@AdamDraper·
I believe this next wave of Bio is going to be bigger than anyone can imagine. Multiple $10T companies. Maybe even a 100 Trillion dollar company.
English
109
84
1.3K
308.2K
Kyle Roh retweetledi
spidey
spidey@lochan_twt·
"If you are working with AI/ML, there are probably lines of code in your computer / server that are written by me" he built vLLM
spidey tweet media
English
52
208
5.8K
730.4K
Kyle Roh
Kyle Roh@kyle_r0k·
Great!! Advances in medical technology should always deserve a standing ovation.. Fantastic results, HR 0.4 in metastatic pancreatic cancer is amazing
English
0
0
1
89
Kyle Roh
Kyle Roh@kyle_r0k·
The deranged, progressivist UK govt agenda now led to police handcuffing a murder victim instead of the attacker. An innocent victim’s life, according to its govt, is only worth a few lines of apology on X. I’m skeptical if such a govt is worth surviving. May he rest in peace..
Elon Musk@elonmusk

Release the footage, you evil bastards!

English
0
0
0
67
Kyle Roh retweetledi
Romain Lopez
Romain Lopez@_romain_lopez_·
We built a joint experimental and computational platform for scalable multi-modal single-cell chemical screens — profiling RNA, protein (including phospho-signaling), and chromatin accessibility responses to thousands of small molecule perturbations in parallel. biorxiv.org/content/10.648…
Romain Lopez tweet media
English
2
40
180
13.7K