Jason Wei ✓

@_jasonwei · sf · joined 22 Oct 2020

ai researcher @meta, past: openai, google 🧠

112 797Followers

736Following

1 264Posts total

1.1MViews on collected posts

Against accounts of the same size

8 posts from the last 90 days, next to the 100K–1M follower range. shown widely, but few of those viewers react.

Median views45 457this account4 116median for 100K–1M

Reach, %40.30%this account1.71%median for 100K–1M

Engagement, %0.66%this account1.47%median for 100K–1M

Metric	This account	Median for 100K–1M	Ratio
Median views per post	45 457	4 116	11.0×
Reach (views ÷ followers)	40.30%	1.71%	23.6×
Engagement rate	0.66%	1.47%	0.45×

Others in this range → Compare with another account → How these benchmarks are built →

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

380.2K16 Jul

9.4K

4.2K

14.2K

208.1K13 Jul

26.9K14 Jul

66.5K12 Aug

297.8K17 Aug

7.9K

1.6K18 Aug

6.6K19 Aug

64K

Last 12 collected posts, oldest on the left. The scale is logarithmic: one post can outrun the rest a hundred times over.

Engagement rate per post

1.07%16 Jul

0.64%

3.08%

0.52%

0.08%13 Jul

0.62%14 Jul

0.71%12 Aug

0.47%17 Aug

0.90%

0.92%18 Aug

0.74%19 Aug

0.40%

Reactions — likes, reposts, replies and quotes — divided by views. Median for 100K–1M accounts is 1.47%.

What the audience does

Likes57.8%5 871 in total

Reposts5.6%573 in total

Replies2.9%299 in total

Quotes1.8%183 in total

Bookmarks31.8%3 234 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.

Latest posts

Cognitive reward shapes in sports and career Sports are amazing environments to learn. When you play a sport for thousands of hours, you start to see the world through that sport. It is a simple fact—your biological neural network is being conditioned to respond to the behavior 64K views · 211 likes · 21 reposts · 20 replies 19 Aug 2026 @_jasonwei but it also seems like larger models have a stronger cognitive core, not just more world knowledge - in the sense of being able to integrate new information as well as grok fundamentals better, and this seems to have continued to improve even going from opus to fable. 6.6K views · 44 likes · 2 reposts · 2 replies 19 Aug 2026 @_jasonwei great arguments, but i partially disagree, knowledge scales with parameters, but tool use doesn't, because it is a behavior, not entirely cognition. i have built RL environments to eval this: single tool use, composing/multi tool use, and tool creation across four do 1.6K views · 12 likes · 2 reposts · 1 replies 18 Aug 2026 @_jasonwei one way to get past the larger models thing i've been using is to periodically use larger models to make deterministic "compound tools" (made out of more primitive tools) that the smaller models cannot screw up. pretty handy form of @erikdunteman "muscle memory" - se 7.9K views · 61 likes · 5 reposts · 4 replies 17 Aug 2026 When language models first started using tools well, I was sympathetic to the narrative that instead of scaling up language models, all we needed was a strong enough "cognitive core", say 1B parameters, and anything else could be done with tool use, like browsing the internet or 297.8K views · 1.2K likes · 112 reposts · 74 replies 17 Aug 2026 What's left for humans in a world where machine intelligence has so many advantages? I recently got a Tesla, and using full self-driving has been a wake up call to just how many advantages AI has over humans. The few times I disengaged it because I thought it was going into the 66.5K views · 394 likes · 36 reposts · 31 replies 12 Aug 2026 On HealthBench Pro, Muse Spark 1.1 achieves similar performance with GPT-5.6 Sol (maybe slightly better) at a fraction of the cost. Affordable health superintelligence is our north star! 26.9K views · 133 likes · 16 reposts · 15 replies 14 Jul 2026 We benchmarked Muse Spark 1.1 and GPT-5.6 Sol on HealthBench Professional, OpenAI's benchmark of 525 real clinician tasks 🏥🩺 Muse Spark 1.1 tops our board: better overall score than GPT-5.6 Sol, statistically on par on the length-adjusted score at a fraction of the cost https:// 208.1K views · 114 likes · 13 reposts · 17 replies 13 Jul 2026 @_jasonwei We actually have a justification for this very intuition "I would never surpass the full ability of someone else because they were playing to their strengths which I didn’t have" -- we show that RL can adapt/calibrate CoTs to the model's own capability, while SFT often 14.2K views · 64 likes · 8 reposts · 1 replies 16 Jul 2025 @_jasonwei this is how bill ackman would post he's joining meta 4.2K views · 127 likes · 0 reposts · 1 replies 16 Jul 2025 @_jasonwei absolutely iconic to post this right now 9.4K views · 60 likes · 0 reposts · 0 replies 16 Jul 2025 Becoming an RL diehard in the past year and thinking about RL for most of my waking hours inadvertently taught me an important lesson about how to live my own life. One of the big concepts in RL is that you always want to be “on-policy”: instead of mimicking other people’s 380.2K views · 3.5K likes · 358 reposts · 133 replies 16 Jul 2025