Liam Fedus ✓
@LiamFedus · San Francisco, CA · joined 16 Oct 2012
Building industrial-scale science at @periodiclabs Past: VP of Post-Training @OpenAI; Google Brain
40 991Followers
1 245Following
1 320Posts total
6MViews on collected posts
Derniers posts
Recherche de nouveaux posts…
Our infrastructure goal is science per GPU-hour.
On our workloads, we achieved:
4.1× training throughput vs. our Megatron baseline
2.5× faster decoding
95%+ cluster utilization
We’ve been fortunate to build alongside the strong open-source community, and we contribute back http
41.6K views · 374 likes
· 23 reposts · 6 replies
Open on X →
“What did we actually make?”
Answering this can take hours. Lab data + midtraining + RL took X-ray diffraction analysis success from 2.7% to 55.3% (~20×) on 134 difficult samples, scored by model judges calibrated against human experts. We see great scaling properties with https
50.6K views · 306 likes
· 16 reposts · 5 replies
Open on X →
We built high-throughput materials labs in Menlo Park to create a loop between experiments and models. The labs generate fresh data, the models learn from it, and then help us decide what to try next.
Using only 1,300 H200s, plus months of our experimental data, we mid-trained h
1.6M views · 5.1K likes
· 521 reposts · 274 replies
Open on X →
0:16
Neon required three pieces: the labs, the research, and the infra.
It’s an early example of how our unique experimental data, combined with highly efficient infrastructure, can be used to train specialized scientific models with strong performance on relatively modest compute ht
329.2K views · 842 likes
· 66 reposts · 18 replies
Open on X →
Great work from our mid-training, RL, and infra teams (not to mention a ton of lab data) to pull this off!
20.3K views · 157 likes
· 5 reposts · 11 replies
Open on X →
Does your girlfriend sit front row for your technical talks and leave with 400 photos? Mine does.
81.8K views · 463 likes
· 2 reposts · 22 replies
Open on X →
Data centers are power-limited and OpenAI's Jalapeño sets new frontiers for tokens per second per watt. Impressive and fast work from OpenAI's ASIC program. https://t.co/GFY3vGP0po
11.9K views · 117 likes
· 3 reposts · 7 replies
Open on X →
Elon's companies are winning in part by co-designing hardware all the way down to the materials. A new alloy unlocked single-piece gigacasting for the Model Y, and a new steel replaced 301 stainless on Starship.
Materials are the deepest layer of the stack, but currently one of
48.9K views · 490 likes
· 22 reposts · 9 replies
Open on X →
@LiamFedus @ekindogus @periodiclabs Congratulations Liam & Dogus! Using AI to accelerate real world scientific explorations is an important pursuit
16.7K views · 181 likes
· 7 reposts · 3 replies
Open on X →
@LiamFedus @ekindogus @periodiclabs We're not going to reason our way to Dyson spheres. Excited for you and all the best to the team!
107.2K views · 666 likes
· 23 reposts · 12 replies
Open on X →
@LiamFedus @ekindogus @periodiclabs Congrats on the launch! Doing things in the physical world is underrated by AI people.
37.4K views · 256 likes
· 8 reposts · 3 replies
Open on X →
Today, @ekindogus and I are excited to introduce @periodiclabs.
Our goal is to create an AI scientist.
Science works by conjecturing how the world might be, running experiments, and learning from the results.
Intelligence is necessary, but not sufficient. New knowledge is http
3.7M views · 4.2K likes
· 442 reposts · 423 replies
Open on X →
Face aux comptes de taille comparable
9 posts des 90 derniers jours, à côté de la tranche de 10K–100K abonnés. diffusé largement, mais peu de ces spectateurs réagissent.
Vues médianes48 872ce compte924médiane pour 10K–100K
Portée, %119.23%ce compte3.62%médiane pour 10K–100K
Engagement, %0.85%ce compte1.52%médiane pour 10K–100K
| Indicateur | Ce compte | Médiane pour 10K–100K | Rapport |
|---|---|---|---|
| Vues médianes par post | 48 872 | 924 | 52.9× |
| Portée (vues ÷ abonnés) | 119.23% | 3.62% | 32.9× |
| Taux d'engagement | 0.85% | 1.52% | 0.56× |
Autres comptes de cette tranche → Comparer avec un autre compte → Comment ces repères sont établis →
Growth & engagement
How the posts we collected actually performed: views and reaction rate post by post, what the audience did with them, and where the follower count goes.
Views per post
3.7M30 Sep
37.4K
107.2K
16.7K
48.9K23 Aug
11.9K25 Aug
81.8K26 Aug
20.3K27 Aug
329.2K15 Sep
1.6M
50.6K
41.6K
6.8K
Last 13 collected posts, oldest on the left. The scale is logarithmic: one post can outrun the rest a hundred times over.
Engagement rate per post
0.15%30 Sep
0.71%
0.66%
1.14%
1.08%23 Aug
1.07%25 Aug
0.60%26 Aug
0.85%27 Aug
0.29%15 Sep
0.38%
0.65%
0.99%
1.24%
Reactions — likes, reposts, replies and quotes — divided by views. Median for 10K–100K accounts is 1.52%.
What the audience does
Likes65.6%13 236 in total
Reposts5.6%1 138 in total
Replies3.9%797 in total
Quotes3.1%623 in total
Bookmarks21.7%4 384 in total
Share of every reaction we collected for this account. Replies mean argument, reposts mean endorsement, bookmarks mean the post was worth keeping.
The follower curve appears once this account has two daily snapshots — we take one a day, and this one is on its first.
Comptes similaires
Bun
@bunjavascript
82K followers
· 4.1K posts
JLL
@JLL
82K followers
· 18.6K posts
ALTSTEIN ✓
@Altsteinn
82K followers
· 145.9K posts
مصرف الراجحي أعمال ✓
@alrajhibankBus
82K followers
· 42.3K posts
Grace ✓
@gracecamille_
82K followers
· 24.9K posts
WCCB Charlotte ✓
@WCCBCharlotte
82K followers
· 243.5K posts
Jesús Fernández-Villaverde ✓
@JesusFerna7026
82K followers
· 7.7K posts
Avinash Gowariker
@avigowariker
82K followers
· 1.8K posts