Alchemist-in-training. đ§ââïž Human & planetary health @thisishcvc đ ïž Past: AI for Science @Stanford âĄïžđ§Ș CEO @Plume_Labs (acq. by @AccuWeather) đŠïž
đïžđČđ
Joined May 2010
- Tweets22.8K
- Following6.6K
- Followers7.3K
- Likes116K
We can barely get reliable simulations of a single protein for a few nanoseconds, but sure, just let Big Computer go brrr and weâll simulate an entire human over clinically relevant time frames. đ«
If you had an exponential amount of compute you could solve simulate biology exactly and wouldn't need to run any experiments.
Some of you will think "of course, but that's an unrealistic assumption" and others will think "no, you need experiments even with that much compute. That's just How It's Done".
You would of course need evaluations once you have a solution, but you don't in principle need experiments to find it: Experiments are a way to save compute, dating from a time when it was much more expensive.
And I think the presentation is exponential but real systems aren't irreducibly exponential. AlphaFold worked, after all.
PSA for those in the back: the Bitter Lesson doesnât say brute force simulations will solve everything; actually, quite the opposite.
xitter.cf/rlacombe/status/211456âŠ
I see many folks making this category error of invoking the Bitter Lesson to argue weâll âsolveâ biology with brute force simulations.
They should read Richard Sutton more carefully.
His point never was that scaling compute solves everything. Itâs more nuanced: he argues architectures that win *tend to be the ones scale with compute and data.*
Canonical example: deep learning in computer vision scaled better than hand-crafted features, and ultimately won the day. Search over the backlinks structure of the internet scaled better than human curation, and won as well.
Simulation isnât learning or search. It scales with compute, but is often hard to parallelize so hasnât benefited as much from GPU architectures. It doesnât scale with data either, except in unique cases like assimilation of weather observations.
Bottom line is: donât parrot the new words youâve just learned if you didnât do your homework, and read (and understood) the primary sources.
xitter.cf/heroin_sheikh/status/2âŠ
I see many folks making this category error of invoking the Bitter Lesson to argue weâll âsolveâ biology with brute force simulations.
They should read Richard Sutton more carefully.
His point never was that scaling compute solves everything. Itâs more nuanced: he argues architectures that win *tend to be the ones scale with compute and data.*
Canonical example: deep learning in computer vision scaled better than hand-crafted features, and ultimately won the day. Search over the backlinks structure of the internet scaled better than human curation, and won as well.
Simulation isnât learning or search. It scales with compute, but is often hard to parallelize so hasnât benefited as much from GPU architectures. It doesnât scale with data either, except in unique cases like assimilation of weather observations.
Bottom line is: donât parrot the new words youâve just learned if you didnât do your homework, and read (and understood) the primary sources.
xitter.cf/heroin_sheikh/status/2âŠ
Les lycĂ©ens français qui demandent la fin de la sĂ©lection Ă lâentrĂ©e de lâuniversitĂ© ne sont pas prĂȘts pour le systĂšme allemand (3 examens ratĂ©s : Ă©limination Ă vie). đł
TIL my alma mater RWTH Aachen (among others) is called "Battle Royale from hell" in some circles lol
In the German higher ed system, if you fail an exam 3 times, you are banned from that field of study forever, everywhere. I think this makes sense.
The exams are hard but imo fair and test understanding not just cramming (mech eng at least). Personally it can be rough, but societally, you don't really want the person who failed fluid mech 3x on your airbus eng team do you??
It's not quite battle royale though because it's not stack ranked, in a good cohort it's possible to get 100% students passing, though I've never seen a cohort like that.
I kinda just assumed it's like this in most countries!
Romain Lacombe retweeted
My Claude found a critical zero day vulnerability in the Code of Hammurabi, unnoticed by developers for 3,779 years.
Romain Lacombe retweeted
China has gotten ahead of the US in biotechnology, largely by making their clinical trial system more effective. I covered this in a @nytimes article.
Now, American patients are travelling there to have access to drugs they can't access at home. It's high time the US woke up and implemented much needed reforms.
"She had been through chemotherapy and surgery, but the cancer had now spread to her brain. And she didnât qualify for the relevant clinical trials.
âToo bad sheâs not Chinese,â a doctor told her son... âShe could go to a pharmacy in China and buy it.â
kffhealthnews.org/health-indâŠ
My take (in French). I'm extremely bearish.
xitter.cf/rlacombe/status/211156âŠ
En tirer les leçons oui, mais comment, concrÚtement ??
Le pays est tiraillĂ© entre des pyromanes antisĂ©mites profondĂ©ment dangereux qui profitent du chaos, des repris de justice racistes issus de la collaboration, et des imbĂ©ciles utiles qui les laissent faire par peur de leur ombre. Le tout Ă©lu par une population mal informĂ©e et Ă©duquĂ©e, qui se ment Ă elle-mĂȘme, pense faire sa petite rĂ©volution en chantant "les MisĂ©rables" et crie Ă l'ultra-turbo-libĂ©ralisme Ă la moindre rĂ©forme alors que notre gĂ©rontocratie croĂ»le sous le poids des retraites.
On fait quoi pour avancer ?
TrĂšs juste.
xitter.cf/chraynaud/status/21115âŠ
âSweet Coraliiine! La laaa la laâ đ
Researchers in @ScienceAdvances have demonstrated that brief physical contact between healthy and bleached coral can restore algae to the bleached coral, leading to their full recovery. scim.ag/4s40Tt3
ALT Two coral samples are shown side by side in a comparison. On the left, labeled "Week 0," the corals appear lighter, with distinct white and brown patterns. On the right, labeled "Week 9," the same corals are noticeably darker with more prominent brown hues, indicating changes over time.
Our age of technical wonders feels like Dune. There are gigantic worms gathering under the sand in the desert, and theyâre swallowing intellectual work while producing spice that powers the galaxy.
When they come for you, will you get swallowed, or will you ride the Worm?
Tweet inspired by JB here:
xitter.cf/jb_michel/status/21107âŠ
All eyez on maths. Whatâs happening right now in mathematics, the angst and awe mathematicians feel, the profound transformation of the job itself almost overnight, is a canary in the coal mine.
mathstodon.xyz/@tao/11739526âŠ
Romain Lacombe retweeted
An internal AI at Waymo has solved the âpredict when your car will arriveâ millennium problem.
Mathematicians at Uber and Lyft in shambles
Been using mostly Argon for some time now, and itâs properly incredible for research!
Do with that information what you will.
đ€âïžđ
Ben Affleck single-handedly making technical competence cool again. End the idiocracy, more you-can-just-do-things role models in Hollywood please! â€ïž
Ben Affleck: "Iâve always been into computers. I can write pretty shitty Python scripts."
Interviewer: lol
Ben: "So obviously you convert the image into a tensor containing the batch, frame and RGB values of every pixel."
Interviewer: wtf
Ben: "Then run a convolutional neural network for edge detection and feature extraction."
Interviewer: jaw on floor
Ben: "Itâs pretty simple."