Natasha Jaques ✓

@natashajaques · Seattle, WA · joined 26 Jun 2009

Assistant Professor leading the Social RL Lab https://t.co/ykwfJG84Bj @uwcse and Staff Research Scientist at @GoogleAI.

35 167Followers

1 142Following

1 624Posts total

454KViews on collected posts

Latest posts

Natasha Jaques ✓@natashajaques15 Sep

LLMs give eerily similar responses to the same prompt, even across different model families. Ask it to name a “well-received book” 6 times, and it will name “To Kill a Mockingbird” every time. In this work, we propose a method to address this problem with multi-agent RL https:/

5.8K views · 91 likes · 14 reposts · 5 replies Open on X →

Natasha Jaques ✓@natashajaques15 Sep

🧬 "In the course of evolution, nature has gone to endless trouble to see that every individual is unlike every other individual." Yet LLMs are forcing us all to be alike by giving us mode-collapsed, homogeneous responses. We propose a multi-agent RL method to fix this. 🧵 https:/

10.3K views · 50 likes · 8 reposts · 4 replies Open on X →

Natasha Jaques ✓@natashajaques14 Sep

Gave a recent talk about this work, which you can check out here if you're interested: https://t.co/Cu5QILMxBH 13.7K views · 80 likes · 8 reposts · 0 replies Open on X →

Natasha Jaques ✓@natashajaques12 Sep

No, RL post-training on random rewards does not improve model capabilities, except under very particular circumstances. Yes, RL post-training can teach models capabilities that aren’t already present in the base model’s pass@k distribution. While these findings might be obvious 36.7K views · 313 likes · 28 reposts · 6 replies Open on X →

Natasha Jaques ✓@natashajaques12 Sep

Can RL post-training improve model capabilities with random rewards? And can it teach skills outside the base model's distribution? Our new paper, Demystifying Reinforcement Learning Post-training of Language Models, unpacks why it succeeds or fails. https://t.co/pCEi7cw4uT 🧵 56.4K views · 259 likes · 21 reposts · 4 replies Open on X →

Natasha Jaques ✓@natashajaques11 Sep

In a random soup of programs mutating over time, self-replicating programs will eventually come to dominate. But will they learn to cooperate with each other? In Kunal’s latest paper on Autopoietic Game Theory, we examine exactly this question. We set up a system in which 14.9K views · 229 likes · 33 reposts · 5 replies Open on X →

Natasha Jaques ✓@natashajaques11 Sep

Can self-interested, self-improving, self-replicating agents learn to cooperate? Our new paper, Tapes Together Strong, shows they can: when social behavior, computation, and reproduction share one energy budget, cooperation evolves from scratch. https://t.co/Lwh8rdFtp9 🧵 https:

GIF

81.7K views · 567 likes · 94 reposts · 30 replies Open on X →

Natasha Jaques ✓@natashajaques20 Aug

Recursive Self-Improvement through Multi-agent RL Post-training and Unsupervised Environment Design (UED)… but it actually works! Delighted to finally release this paper, which trains a single LLM to act as both an Environment Designer to build new multi-turn RL training 56.9K views · 426 likes · 56 reposts · 13 replies Open on X →

Natasha Jaques ✓@natashajaques20 Aug

Continuous self-improvement needs an ever-expanding supply of training environments (goals). SPADE: one model self-plays the Environment Designer and the Reasoning Agent, writing executable, agentic environments that get harder as it improves. Environment scaling on its own. ♠️

0:30

177.7K views · 700 likes · 127 reposts · 20 replies Open on X →

Against accounts of the same size

9 posts from the last 90 days, next to the 10K–100K follower range. shown widely, but few of those viewers react.

Median views36 668this account924median for 10K–100K

Reach, %104.27%this account3.62%median for 10K–100K

Engagement, %0.87%this account1.52%median for 10K–100K

Metric	This account	Median for 10K–100K	Ratio
Median views per post	36 668	924	39.7×
Reach (views ÷ followers)	104.27%	3.62%	28.8×
Engagement rate	0.87%	1.52%	0.57×

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

177.7K20 Aug

56.9K

81.7K11 Sep

14.9K

56.4K12 Sep

36.7K

13.7K14 Sep

10.3K15 Sep

5.8K

Last 9 collected posts, oldest on the left. The scale is logarithmic: one post can outrun the rest a hundred times over.

Engagement rate per post

0.50%20 Aug

0.87%

0.87%11 Sep

1.80%

0.51%12 Sep

0.95%

0.64%14 Sep

0.64%15 Sep

1.91%

Reactions — likes, reposts, replies and quotes — divided by views. Median for 10K–100K accounts is 1.52%.

What the audience does

Likes48.7%2 715 in total

Reposts7.0%389 in total

Replies1.6%87 in total

Quotes1.1%64 in total

Bookmarks41.7%2 324 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.