RSS

ryan_greenblatt

Karma: 19,553

I’m the chief scientist at Redwood Research.

AIs will greatly change en­g­ineer­ing in AI com­pa­nies well be­fore AGI

ryan_greenblatt9 Sep 2025 16:58 UTC
45 points
9 comments11 min readLW link

Trust me bro, just one more RL scale up, this one will be the real scale up with the good en­vi­ron­ments, the ac­tu­ally le­git one, trust me bro

ryan_greenblatt3 Sep 2025 13:21 UTC
151 points
30 comments8 min readLW link

At­tach­ing re­quire­ments to model re­leases has se­ri­ous down­sides (rel­a­tive to a differ­ent dead­line for these re­quire­ments)

ryan_greenblatt27 Aug 2025 17:04 UTC
98 points
2 comments3 min readLW link

My AGI timeline up­dates from GPT-5 (and 2025 so far)

ryan_greenblatt20 Aug 2025 16:11 UTC
162 points
14 comments4 min readLW link

Re­cent Red­wood Re­search pro­ject proposals

ryan_greenblatt, Buck, Julian Stastny, joshc, Alex Mallen, Adam Kaufman , Tyler Tracy, Aryan Bhatt and Joey Yudelson
14 Jul 2025 22:27 UTC
91 points
0 comments3 min readLW link

Jankily con­trol­ling superintelligence

ryan_greenblatt27 Jun 2025 14:05 UTC
69 points
4 comments7 min readLW link

What does 10x-ing effec­tive com­pute get you?

ryan_greenblatt24 Jun 2025 18:33 UTC
55 points
10 comments12 min readLW link

Pre­fix cache un­trusted mon­i­tors: a method to ap­ply af­ter you catch your AI

ryan_greenblatt20 Jun 2025 15:56 UTC
32 points
1 comment7 min readLW link

AI safety tech­niques lev­er­ag­ing distillation

ryan_greenblatt19 Jun 2025 14:31 UTC
61 points
0 comments12 min readLW link

When does train­ing a model change its goals?

Vivek Hebbar and ryan_greenblatt
12 Jun 2025 18:43 UTC
71 points
2 comments15 min readLW link

OpenAI now has an RL API which is broadly accessible

ryan_greenblatt11 Jun 2025 23:39 UTC
43 points
1 comment5 min readLW link

When is it im­por­tant that open-weight mod­els aren’t re­leased? My thoughts on the benefits and dan­gers of open-weight mod­els in re­sponse to de­vel­op­ments in CBRN ca­pa­bil­ities.

ryan_greenblatt9 Jun 2025 19:19 UTC
63 points
11 comments9 min readLW link

The best ap­proaches for miti­gat­ing “the in­tel­li­gence curse” (or grad­ual dis­em­pow­er­ment); my quick guesses at the best ob­ject-level interventions

ryan_greenblatt31 May 2025 18:20 UTC
71 points
19 comments5 min readLW link

AIs at the cur­rent ca­pa­bil­ity level may be im­por­tant for fu­ture safety work

ryan_greenblatt12 May 2025 14:06 UTC
82 points
2 comments4 min readLW link

Slow cor­po­ra­tions as an in­tu­ition pump for AI R&D automation

ryan_greenblatt and elifland
9 May 2025 14:49 UTC
91 points
23 comments9 min readLW link

What’s go­ing on with AI progress and trends? (As of 5/​2025)

ryan_greenblatt2 May 2025 19:00 UTC
75 points
8 comments8 min readLW link

7+ tractable di­rec­tions in AI control

Julian Stastny and ryan_greenblatt
28 Apr 2025 17:12 UTC
93 points
1 comment13 min readLW link

To be leg­ible, ev­i­dence of mis­al­ign­ment prob­a­bly has to be behavioral

ryan_greenblatt15 Apr 2025 18:14 UTC
57 points
19 comments3 min readLW link

Why do mis­al­ign­ment risks in­crease as AIs get more ca­pa­ble?

ryan_greenblatt11 Apr 2025 3:06 UTC
33 points
6 comments3 min readLW link

An overview of ar­eas of con­trol work

ryan_greenblatt25 Mar 2025 22:02 UTC
32 points
0 comments28 min readLW link