Jacob Andreas

@jacobandreas · Cambridge, MA · joined 25 Mar 2007

Teaching computers to read. Assoc. prof @MITEECS / @MIT_CSAIL / @NLP_MIT (he/him). https://t.co/5kCnXHjtlY https://t.co/2A3qF5vdJw

24 918Followers

956Following

2 857Posts total

200.6KViews on collected posts

Posts recentes

Jacob Andreas@jacobandreas24 Jul

👉 New preprint! (and a direction I'm excited to do more work on soon) 10.1K views · 63 likes · 1 reposts · 1 replies Open on X →

Jacob Andreas@jacobandreas23 Jul

Pluralistic alignment is thriving as a research agenda yet failing at its goal: making the AI systems people actually use more pluralistic🌈 🚨New position paper: we argue adoption in deployed models should be the fields main goal & we provide a roadmap of how to get there 🧵1/ ht 34.4K views · 121 likes · 28 reposts · 7 replies Open on X →

Jacob Andreas@jacobandreas7 Jul

@jacobandreas @evanqed @arnab_api Your name is on it too and it’s kind of weird that you wrote your post not being explicit about that.. You shouldn’t feel like there’s anything wrong with promoting your own research. 270 views · 3 likes · 0 reposts · 0 replies Open on X →

Jacob Andreas@jacobandreas7 Jul

@jacobandreas @evanqed @arnab_api one of the earliest (mech) interp papers that I remember reading! Also a good follow-up: https://t.co/8dgm091XtR 750 views · 11 likes · 0 reposts · 0 replies Open on X →

Jacob Andreas@jacobandreas7 Jul

@jacobandreas @evanqed @arnab_api Jacob-ian there, done that! 1.6K views · 4 likes · 0 reposts · 1 replies Open on X →

Jacob Andreas@jacobandreas7 Jul

If you're excited about Anthropic's J-space work, definitely worth checking out the original paper on Jacobian lenses by @evanqed and @arnab_api! https://t.co/gttqzgB4VB 37.9K views · 511 likes · 62 reposts · 10 replies Open on X →

Jacob Andreas@jacobandreas3 Jul

👉 New preprint (we had a big backlog 😅)! Revisiting adversarial imitation learning for the era of RLVR: 11.9K views · 62 likes · 8 reposts · 4 replies Open on X →

Jacob Andreas@jacobandreas3 Jul

Higher benchmark scores do not always mean better models for users. Why? We claim that RL teaches LMs to be correct but not how to be correct: code can pass tests but be unreadable; explanations can be right but unclear. How do we train LMs to be right in the right way? (1/n) 31.6K views · 131 likes · 33 reposts · 5 replies Open on X →

Jacob Andreas@jacobandreas1 Jul

👉 Preprint: understanding learning dynamics & mechanisms in LMs trained to explain / predict their own behaviors! 8.8K views · 64 likes · 7 reposts · 2 replies Open on X →

Jacob Andreas@jacobandreas1 Jul

New Paper 📄: LMs just want to explain themselves! When we SFT an LM on explanations of its own behaviors, do they learn to actually introspect, or do they merely imitate the original training distribution? We find evidence for the former. Despite training on a static set of http 36.4K views · 196 likes · 36 reposts · 6 replies Open on X →

Jacob Andreas@jacobandreas29 Jun

👉 New preprint! Automated interpretability by approximating / replacing NN components (here attention heads) with programs. 26.9K views · 168 likes · 16 reposts · 0 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

26.9K29 Jun

36.4K1 Jul

8.8K

31.6K3 Jul

11.9K

37.9K7 Jul

1.6K

750

270

34.4K23 Jul

10.1K24 Jul

Last 11 collected posts, oldest on the left. The scale is logarithmic: one post can outrun the rest a hundred times over.

Engagement rate per post

0.69%29 Jun

0.68%1 Jul

0.83%

0.55%3 Jul

0.62%

1.55%7 Jul

0.32%

1.47%

1.11%

0.48%23 Jul

0.64%24 Jul

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

What the audience does

Likes60.5%1 334 in total

Reposts8.7%191 in total

Replies1.6%36 in total

Quotes1.3%28 in total

Bookmarks28.0%617 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.

Posts recentes

Growth & engagement

Views per post

Engagement rate per post

What the audience does

Contas semelhantes