RSS

Should our jour­nal pub­lish AI-drafted manuscripts?

Dan MacKinlay16 Sep 2026 8:09 UTC
4 points
0 comments27 min readLW link
(danmackinlay.name)

Who com­putes?

Dan MacKinlay16 Sep 2026 8:08 UTC
4 points
0 comments18 min readLW link
(danmackinlay.name)

Phan­tom trans­fer works via ex­tremely sub­tle se­man­tic cues

16 Sep 2026 4:52 UTC
16 points
0 comments20 min readLW link

Is METR A Mean­ingful Check On An­thropic?

SE Gyges16 Sep 2026 0:34 UTC
48 points
9 comments6 min readLW link
(www.verysane.ai)

Why would AI cause hu­man ex­tinc­tion?

jacob15 Sep 2026 23:49 UTC
9 points
0 comments10 min readLW link

Inoc­u­la­tion Mid­train­ing with Learned Neologisms

15 Sep 2026 21:52 UTC
52 points
0 comments6 min readLW link
(arxiv.org)

Quick notes from teach­ing tech­ni­cal pro­files how to talk in public

Camille B. 15 Sep 2026 21:52 UTC
62 points
2 comments6 min readLW link

Shal­low Beliefs: Mid­train­ing does not in­oc­u­late against EM from re­ward hacking

15 Sep 2026 21:50 UTC
43 points
1 comment3 min readLW link

Why I’m do­ing the Su­san Calvin Project

Haoxing Du15 Sep 2026 19:06 UTC
16 points
0 comments4 min readLW link
(susancalvinproject.substack.com)

Any AI pause will have defec­tors. How to en­sure their in­car­cer­a­tion ac­tu­ally pre­vents them from covertly con­tribut­ing to AI re­search from be­hind bars?

Connor Williams15 Sep 2026 18:31 UTC
3 points
0 comments1 min readLW link
(connorsscratchpad.substack.com)

You Don’t Have to Trust the AI Labs (in or­der to take their call for reg­u­la­tion se­ri­ously)

ixotope15 Sep 2026 18:04 UTC
11 points
0 comments11 min readLW link

Co­op­er­a­tion with AIs seems to be a low-hang­ing fruit for bet­ter eval practices

Clément Dumas15 Sep 2026 17:48 UTC
92 points
10 comments8 min readLW link

Align­ment & Suc­ces­sion: Mo­ral­ity Lives in the Hu­man Individual

L Rudolf L15 Sep 2026 15:38 UTC
24 points
2 comments16 min readLW link

Why Fo­cus on Ex­tinc­tion?

alkjash15 Sep 2026 14:37 UTC
8 points
2 comments1 min readLW link
(radimentary.wordpress.com)

Moloch Does My Hair

spookyuser15 Sep 2026 10:31 UTC
16 points
0 comments19 min readLW link

For most peo­ple “in­tel­li­gence” is not goal achievement

Pato15 Sep 2026 7:52 UTC
13 points
2 comments1 min readLW link

How we might ac­tu­ally pace the fron­tier: A pro­posal for AI com­pa­nies to do pub­lic pac­ing ex­er­cises.

Dewi Erwan15 Sep 2026 5:36 UTC
9 points
0 comments4 min readLW link
(blog.dewierwan.com)

As­tra ap­pears to perform be­lief-prop­a­ga­tion-like in­fer­ence with­out CoT

MBaert15 Sep 2026 2:48 UTC
84 points
0 comments10 min readLW link

One co­or­di­nate breaks abliter­a­tion on Gemma-3

Abhishek Mishra15 Sep 2026 2:36 UTC
7 points
0 comments14 min readLW link

How to think about LLM effort

Tao Lin15 Sep 2026 2:31 UTC
16 points
0 comments2 min readLW link