RSS

Sam Marks

Karma: 6,532

Thoughts on the per­sona se­lec­tion model

Sam Marks24 Sep 2026 1:21 UTC
88 points
7 comments4 min readLW link

Train­ing Models to Pre­dict and Ex­plain Their In-the-Wild Behavior

Adam Karvonen, Subhash Kantamneni, Euan Ong and Sam Marks
4 Sep 2026 16:16 UTC
50 points
2 comments8 min readLW link

Eval­u­at­ing Ex­pla­na­tions of LLM Be­hav­ior In The Wild with Coun­ter­fac­tual Experiments

Adam Karvonen, Euan Ong, Subhash Kantamneni and Sam Marks
21 Aug 2026 19:09 UTC
74 points
6 comments7 min readLW link

Nat­u­ral Lan­guage Au­toen­coders Pro­duce Un­su­per­vised Ex­pla­na­tions of LLM Activations

Subhash Kantamneni, kitft, Euan Ong and Sam Marks
7 May 2026 20:21 UTC
209 points
35 comments8 min readLW link

Model Spec Mid­train­ing: Im­prov­ing How Align­ment Train­ing Generalizes

Chloe Li, Nevan Wichers, saraprice, Sam Marks and Jonathan Kutasov
5 May 2026 21:55 UTC
73 points
7 comments7 min readLW link
(alignment.anthropic.com)

In­tro­spec­tion Adapters: Train­ing LLMs to Re­port Their Learned Behaviors

keshavs, RowanWang, abhayesian and Sam Marks
28 Apr 2026 19:02 UTC
41 points
2 comments12 min readLW link
(alignment.anthropic.com)

Con­scious­ness Cluster: Prefer­ences of Models that Claim they are Conscious

James Chua, Owain_Evans, Sam Marks and Jan Betley
18 Mar 2026 16:06 UTC
92 points
30 comments5 min readLW link

Cen­sored LLMs as a Nat­u­ral Testbed for Se­cret Knowl­edge Elicitation

Bartosz Cywiński, Helena Casademunt, Khoi Tran, aryaj, Sam Marks and Neel Nanda
9 Mar 2026 18:50 UTC
41 points
3 comments5 min readLW link

The per­sona se­lec­tion model

Sam Marks23 Feb 2026 22:56 UTC
189 points
55 comments43 min readLW link
(alignment.anthropic.com)

Ac­ti­va­tion Or­a­cles: Train­ing and Eval­u­at­ing LLMs as Gen­eral-Pur­pose Ac­ti­va­tion Explainers

Sam Marks, Adam Karvonen, James Chua, Subhash Kantamneni, Euan Ong, Julian Minder, Clément Dumas and Owain_Evans
18 Dec 2025 20:21 UTC
155 points
11 comments8 min readLW link
(arxiv.org)

Eval­u­at­ing hon­esty and lie de­tec­tion tech­niques on a di­verse suite of dishon­est models

RowanWang, Sam Marks, Johannes Treutlein, evhub and Fabien Roger
25 Nov 2025 19:33 UTC
41 points
0 comments4 min readLW link
(alignment.anthropic.com)

Steer­ing Eval­u­a­tion-Aware Models to Act Like They Are Deployed

Tim Hua, andrq, Sam Marks and Neel Nanda
30 Oct 2025 15:03 UTC
62 points
12 comments18 min readLW link

Inoc­u­la­tion prompt­ing: In­struct­ing mod­els to mis­be­have at train-time can im­prove run-time behavior

Sam Marks, Nevan Wichers, Daniel Tan, Aram Ebtekar, Jozdien, David Africa, Alex Mallen and Fabien Roger
8 Oct 2025 22:02 UTC
184 points
37 comments2 min readLW link

Petri: An open-source au­dit­ing tool to ac­cel­er­ate AI safety research

Sam Marks7 Oct 2025 20:39 UTC
77 points
0 comments1 min readLW link
(alignment.anthropic.com)

Elic­it­ing se­cret knowl­edge from lan­guage models

Bartosz Cywiński, Arthur Conmy and Sam Marks
2 Oct 2025 20:57 UTC
69 points
3 comments2 min readLW link
(arxiv.org)

Dis­cov­er­ing Back­door Triggers

andrq, Tim Hua, Sam Marks, Arthur Conmy and Neel Nanda
19 Aug 2025 6:24 UTC
57 points
4 comments13 min readLW link

Towards Align­ment Au­dit­ing as a Num­bers-Go-Up Science

Sam Marks4 Aug 2025 22:30 UTC
129 points
15 comments6 min readLW link

Build­ing and eval­u­at­ing al­ign­ment au­dit­ing agents

Sam Marks, trentbrick, RowanWang, Sam Bowman, Euan Ong, Johannes Treutlein and evhub
24 Jul 2025 19:22 UTC
47 points
1 comment5 min readLW link

Steer­ing Out-of-Distri­bu­tion Gen­er­al­iza­tion with Con­cept Abla­tion Fine-Tuning

kh4dien, Helena Casademunt, Adam Karvonen, Sam Marks, Senthooran Rajamanoharan and Neel Nanda
23 Jul 2025 14:57 UTC
79 points
8 comments5 min readLW link

Prin­ci­ples for Pick­ing Prac­ti­cal In­ter­pretabil­ity Projects

Sam Marks15 Jul 2025 17:38 UTC
33 points
0 comments13 min readLW link