Yejin Choi ✓

@YejinChoinka · Seattle, WA · joined 05 Aug 2017

professor at Stanford, researcher at NVIDIA, adventurer at heart

30 840Followers

492Following

2 007Posts total

268.6KViews on collected posts

Derniers posts

Yejin Choi ✓@YejinChoinka25 Jan

Ah, that statement "approaches like AlphaEvolve" was intended specifically for AlphaEvolve or approaches that rely just on evolutionary search. Thanks for the pointer to 😍 EvoTune 😍, @caglarml! Let's say great minds think alike 🤣 We will update our arxiv shortly to cite your 14.9K views · 61 likes · 1 reposts · 3 replies Open on X →

Yejin Choi ✓@YejinChoinka25 Jan

Thanks @docmilanfar for sharing your thoughts. Two points here: First, you've misrepresented my argument. Saying babies are "deployed" from day 1 doesn't mean they stop learning—quite the opposite! My point is precisely that learning continues during deployment. See our 34.4K views · 179 likes · 17 reposts · 3 replies Open on X →

Yejin Choi ✓@YejinChoinka25 Jan

@YejinChoinka Very cool work! However, statement on approaches like AlphaEvolve is fixed at test time is incorrect; we proposed Evotune exactly for that purpose: https://t.co/XLsc5fBpTa There have been a few more papers came up and and build on this idea too since last year. 16.9K views · 24 likes · 0 reposts · 1 replies Open on X →

Yejin Choi ✓@YejinChoinka25 Jan

Excited to share TTT-Discover (Test-Time Training for Discovery)—seeking new discoveries on long-standing problems: ✅Erdős min overlap, ✅denoising for single-cell analysis, and ✅GPU kernels! The key insight: Scientific discovery requires learning from a long sequence of 52.5K views · 302 likes · 31 reposts · 11 replies Open on X →

Yejin Choi ✓@YejinChoinka12 Jan

Incredibly excited to share our latest work with @NVIDIAAI that has been over a year in the making: 🔥End-to-End Test-Time Training 🔥 Automated researchers need to compress years of experience into intuition 🧠– far too much to just dump into context. But how do humans do 38.1K views · 237 likes · 34 reposts · 5 replies Open on X →

Yejin Choi ✓@YejinChoinka13 Oct 2025

Instruction tuning has a hidden cost: ✅ Better at following instructions ❌ Narrower output distribution ❌ Worse in-context steerability We built 🌈 Spectrum Suite to investigate this and 🌈 Spectrum Tuning as an alternative post-training method — 42.5K views · 222 likes · 30 reposts · 3 replies Open on X →

Yejin Choi ✓@YejinChoinka13 Oct 2025

🤖➡️📉 Post-training made LLMs better at chat and reasoning—but worse at distributional alignment, diversity, and sometimes even steering(!) We measure this with our new resource (Spectrum Suite) and introduce Spectrum Tuning (method) to bring them back into our models! 🌈 1/🧵 htt

69.5K views · 195 likes · 49 reposts · 5 replies Open on X →

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

69.5K13 Oct

42.5K

38.1K12 Jan

52.5K25 Jan

16.9K

34.4K

14.9K

Last 7 collected posts, oldest on the left. The scale is logarithmic: one post can outrun the rest a hundred times over.

Engagement rate per post

0.36%13 Oct

0.60%

0.72%12 Jan

0.66%25 Jan

0.15%

0.58%

0.43%

Reactions — likes, reposts, replies and quotes — divided by views.

What the audience does

Likes58.2%1 220 in total

Reposts7.7%162 in total

Replies1.5%31 in total

Quotes0.5%10 in total

Bookmarks32.2%675 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.

Derniers posts

Growth & engagement

Views per post

Engagement rate per post

What the audience does

Comptes similaires