RSS

Rohin Shah(Rohin Shah)

Karma: 14,281

Research Scientist at DeepMind. Creator of the Alignment Newsletter. http://​​rohinshah.com/​​

Im­prov­ing Dic­tionary Learn­ing with Gated Sparse Autoencoders

Senthooran Rajamanoharan, Arthur Conmy, lsgos, Tom Lieberum, Vikrant Varma, János Kramár, Rohin Shah and Neel Nanda
25 Apr 2024 18:43 UTC
60 points
23 comments1 min readLW link
(arxiv.org)

AtP*: An effi­cient and scal­able method for lo­cal­iz­ing LLM be­havi­our to components

Neel Nanda, János Kramár, Tom Lieberum and Rohin Shah
18 Mar 2024 17:28 UTC
19 points
0 comments1 min readLW link
(arxiv.org)

Fact Find­ing: Do Early Lay­ers Spe­cial­ise in Lo­cal Pro­cess­ing? (Post 5)

Neel Nanda, Senthooran Rajamanoharan, János Kramár and Rohin Shah
23 Dec 2023 2:46 UTC
18 points
0 comments4 min readLW link

Fact Find­ing: How to Think About In­ter­pret­ing Me­mori­sa­tion (Post 4)

Senthooran Rajamanoharan, Neel Nanda, János Kramár and Rohin Shah
23 Dec 2023 2:46 UTC
22 points
0 comments9 min readLW link

Fact Find­ing: Try­ing to Mechanis­ti­cally Un­der­stand­ing Early MLPs (Post 3)

Neel Nanda, Senthooran Rajamanoharan, János Kramár and Rohin Shah
23 Dec 2023 2:46 UTC
9 points
0 comments16 min readLW link

Fact Find­ing: Sim­plify­ing the Cir­cuit (Post 2)

Senthooran Rajamanoharan, Neel Nanda, János Kramár and Rohin Shah
23 Dec 2023 2:45 UTC
18 points
3 comments14 min readLW link

Fact Find­ing: At­tempt­ing to Re­v­erse-Eng­ineer Fac­tual Re­call on the Neu­ron Level (Post 1)

Neel Nanda, Senthooran Rajamanoharan, János Kramár and Rohin Shah
23 Dec 2023 2:44 UTC
106 points
4 comments22 min readLW link

Dis­cus­sion: Challenges with Un­su­per­vised LLM Knowl­edge Discovery

Seb Farquhar, Vikrant Varma, zac_kenton, gasteigerjo, Vlad Mikulik and Rohin Shah
18 Dec 2023 11:58 UTC
147 points
21 comments10 min readLW link

Ex­plain­ing grokking through cir­cuit efficiency

Vikrant Varma and Rohin Shah
8 Sep 2023 14:39 UTC
98 points
10 comments3 min readLW link
(arxiv.org)

Does Cir­cuit Anal­y­sis In­ter­pretabil­ity Scale? Ev­i­dence from Mul­ti­ple Choice Ca­pa­bil­ities in Chinchilla

Neel Nanda, Tom Lieberum, Matthew Rahtz, János Kramár, Geoffrey Irving, Rohin Shah and Vlad Mikulik
20 Jul 2023 10:50 UTC
43 points
3 comments2 min readLW link
(arxiv.org)

Shah (Deep­Mind) and Leahy (Con­jec­ture) Dis­cuss Align­ment Cruxes

OliviaJ, Rohin Shah, Connor Leahy and Andrea_Miotti
1 May 2023 16:47 UTC
94 points
10 comments30 min readLW link

[Linkpost] Some high-level thoughts on the Deep­Mind al­ign­ment team’s strategy

Vika and Rohin Shah
7 Mar 2023 11:55 UTC
128 points
13 comments5 min readLW link
(drive.google.com)

Cat­e­go­riz­ing failures as “outer” or “in­ner” mis­al­ign­ment is of­ten confused

Rohin Shah6 Jan 2023 15:48 UTC
86 points
21 comments8 min readLW link

Defi­ni­tions of “ob­jec­tive” should be Prob­a­ble and Predictive

Rohin Shah6 Jan 2023 15:40 UTC
43 points
27 comments12 min readLW link

Refin­ing the Sharp Left Turn threat model, part 2: ap­ply­ing al­ign­ment techniques

Vika, Vikrant Varma, Ramana Kumar and Rohin Shah
25 Nov 2022 14:36 UTC
39 points
9 comments6 min readLW link
(vkrakovna.wordpress.com)

Threat Model Liter­a­ture Review

zac_kenton, Rohin Shah, David Lindner, Vikrant Varma, Vika, Mary Phuong, Ramana Kumar and Elliot Catt
1 Nov 2022 11:03 UTC
74 points
4 comments25 min readLW link

Clar­ify­ing AI X-risk

zac_kenton, Rohin Shah, David Lindner, Vikrant Varma, Vika, Mary Phuong, Ramana Kumar and Elliot Catt
1 Nov 2022 11:03 UTC
127 points
24 comments4 min readLW link1 review

More ex­am­ples of goal misgeneralization

Rohin Shah and Vikrant Varma
7 Oct 2022 14:38 UTC
53 points
8 comments2 min readLW link
(deepmindsafetyresearch.medium.com)

[AN #173] Re­cent lan­guage model re­sults from DeepMind

Rohin Shah21 Jul 2022 2:30 UTC
37 points
9 comments8 min readLW link
(mailchi.mp)

[AN #172] Sorry for the long hi­a­tus!

Rohin Shah5 Jul 2022 6:20 UTC
54 points
0 comments3 min readLW link
(mailchi.mp)