arsu (@arsu) profile picture
Joined December 2007
Photo by arsu (@arsu)Photo by arsu (@arsu)Photo by arsu (@arsu)Photo by arsu (@arsu)Photo by arsu (@arsu)Photo by arsu (@arsu)
NEWS 🚨: NASA has released onboard video of Artemis II's reentry through Earth's atmosphere 🤩 Top speed: about 25,000 mph
290
6,948
555
76,036
7,356,793
RT @Dr_Atoosa: OpenAI was founded because Sam and co thought Google couldn't be trusted with AGI. Anthropic split off to be the "responsibl…
1
1
Pasada total la del skater profesional olímpico y los fotógrafos 🔝🔝🔝
500
14,373
831
106,258
7,009,473
"Amaze amaze amaze" #ArtemisII
3
55
10
388
17,352
Information overload and the hectic pace of modern life, 1924 edition
6
2
30
6,740
Replying to @kevinweil
Hi, as the owner/maintainer of erdosproblems.com, this is a dramatic misrepresentation. GPT-5 found references, which solved these problems, that I personally was unaware of. The 'open' status only means I personally am unaware of a paper which solves it.
22
142
49
2,844
688,129
this is embarrassing
67
156
97
3,342
1,353,261
For over a century, antibiotics were found by slow, painstaking, serendipitous screening. Today we share a different path: the first demonstration of deep reinforcement learning (RL) for antibiotic discovery. I’m thrilled to introduce ApexAmphion—inspired by the systems that mastered Go and StarCraft. We pair a 6.4B-parameter protein language model with PPO in a closed loop to generate → score → optimize antibiotic candidates at digital speed. In games, RL agents explore vast decision spaces and optimize toward rewards (winning). ApexAmphion brings the same logic to biology. Instead of chess moves, the agent proposes peptide sequences (candidate antibiotics). Instead of a scoreboard, it receives multi-objective rewards tied to predicted potency (minimum inhibitory concentration, MIC) and developability—properties that make a drug more likely to succeed. Iteration by iteration, the system learns to design better molecules. Highlights (ground-truth experiments): • 100/100 designed molecules active in vitro (100% hit rate) • 99/100 active against ≥2 clinically relevant pathogens (incl. MDR) • Multi-objective rewards: predicted MIC + key physicochemical properties • Rapid & steerable: tune potency and developability on demand • Amphorium: >2 million machine-annotated candidates for follow-up In other words, our ApexAmphion model has now mastered antibiotic discovery and can design promising candidates within hours, often in a single morning or afternoon. We’re compressing years of trial-and-error into hours-scale loops, expanding the space of antimicrobial discovery, and executing a programmable RL pipeline that moves from games to potential life-saving therapeutics. Incredible Team work: @Hanqun_CAO, @mdt_torres, Jingjie Zhang, Zijun Gao, @WUFang40615703, Chunbin Gu, @jure, @YejinChoinka, Guangyong Chen, Pheng-Ann Heng. Paper: arxiv.org/abs/2509.18153 #ReinforcementLearning #AI #DrugDiscovery #Antibiotics #AMR #ProteinLM #SyntheticBiology #Biotech #ApexAmphion
11
95
12
440
55,848
🧵¿Cómo se hace una predicción meteorológica? ¿Cómo funciona un modelo meteorológico? En este hilo hablaremos del complejo proceso científico y técnico en que se basa el método de elaboración de la predicción moderna del tiempo y que se usa para la predicción de una dana.
48
1,025
82
2,320
296,837
Ojo. Estoy descubriendo que hay cuentas aquí que hablan sobre temas de los que no tienen mucha idea, pero lo escriben en plan contundente y, claro, les cuela. No creo que sean muchas. Sigo investigando y os digo.
39
41
7
349
21,821
Raptor 3 (sea level variant) Thrust: 280tf Specific impulse: 350s Engine mass: 1525kg Engine + vehicle-side commodities and hardware mass : 1720kg
992
4,397
1,112
30,590
5,514,668
The world’s most powerful launch vehicle ever developed, powered by 33 Raptor engines, lifting off from Starbase
2,421
13,013
1,982
87,977
32,721,625
Elea Boa elkarlanean sortu da. Hasieratik proiektuan sinetsi eta %100a eman duen jendez inguratuta. Abestiak Santa Monican jaio, Los Angeleseko Los Feliz auzoan grabatu eta Zarautzen musikari eta lagun finen laguntzaz borobildu eta amaitu genituen. – at Santa Monica, CA
1
2
7
796
3 hilabete ‘Amesten’ kalean dela. Eskerrik asko abestia entzun, bideoa ikusi, lagunekin partekatu edo babes mezu bat bidali didazuen guztioi. ‘Amesten’ was released 3 months ago. Thanks to everyone who listened to the song, watched the video, shared it or sent me a message.
1
3
260
Gravity xkcd.com/2712/ Today’s comic is interactive! Click the image on xkcd.com and use arrow keys or tap to fire thrusters.
161
713
103
5,741
789,185
There's a new competitor to DALL-E out there: Google's Imagen. I don't have access to Imagen but I do have some of their example images. Let's do some side-by side comparisons. (left is Imagen, right is DALL-E) 1. "A blue jay standing on a large basket of rainbow macarons."
51
402
177
2,198
I train these clever creatures to save victims trapped in collapsed buildings after earthquakes. We kit them out with a rat backpack, and train them to trigger a switch when they find a victim & come back for a tasty treat 🐀 #herosnotpests #science #weirdjobs #WomenInSTEM
1,798
18,296
5,138
125,983
Conspiracy theorizing is a disease that affects people intelligent enough to detect a pattern, but not intelligent enough (or not mentally balanced enough) to not get fooled by random arrangements. Overdetection, the domain of loonies, is vastly worse than underdetection.
200
912
131
4,908
The future of Humanity (My first tweet yay)
194
3,307
546
14,790