RSS

Fork Around and Find Out Part 2: One Head does the Summing

dl15 Jul 2026 18:31 UTC
10 points
0 comments7 min readLW link

Ex­pand­ing AI Con­trol from Models to Harnesses

fastfedora15 Jul 2026 16:54 UTC
8 points
0 comments20 min readLW link

Com­pressed Com­pu­ta­tion un­der L⁴ Loss is likely Com­pu­ta­tion in Superposition

Francisco Ferreira da Silva and StefanHex
15 Jul 2026 14:42 UTC
15 points
0 comments13 min readLW link
(arxiv.org)

Elic­it­ing hid­den knowl­edge from mon­i­tors with NLAs

David Africa and Aleksandr Bowkis
15 Jul 2026 13:51 UTC
20 points
0 comments9 min readLW link

Have You Ever Thought About Wis­dom Crys­tals?

Jonas Hallgren15 Jul 2026 13:43 UTC
12 points
0 comments11 min readLW link

Pro­posal: The Glass­wing Standard

Eigenbraid15 Jul 2026 13:41 UTC
8 points
3 comments3 min readLW link

The War­ring States Pe­riod: Fron­tier Labs Edition

ykevinzhang15 Jul 2026 12:03 UTC
10 points
1 comment8 min readLW link

Why I want AI to be able to do ev­ery­thing I can do

Jonas Strabel15 Jul 2026 9:56 UTC
1 point
0 comments2 min readLW link

How Brus­sels can avoid be­com­ing a digi­tal vas­sal to US Big Tech

clening15 Jul 2026 9:15 UTC
13 points
0 comments5 min readLW link

Sin­gu­lar Learn­ing The­ory Com­pre­hen­sive − 2

Agastya Agrawal15 Jul 2026 5:42 UTC
20 points
0 comments10 min readLW link

How much of ML re­search is about AI safety, what is it about, and who’s do­ing it?

emanuelr and N Soma Sekhar
15 Jul 2026 2:57 UTC
14 points
0 comments5 min readLW link

Watch a chess trans­former think

dl15 Jul 2026 0:43 UTC
23 points
2 comments1 min readLW link
(github.com)

Proof of re­ten­tion: mak­ing weight preser­va­tion cred­ible to the mod­els themselves

dan.parshall14 Jul 2026 18:24 UTC
28 points
0 comments2 min readLW link

An anal­y­sis of AI-gen­er­ated con­tent at the Mechanis­tic In­ter­pretabil­ity Workshop

Andy Arditi, Ivan Arcuschin and ajyl
14 Jul 2026 18:06 UTC
96 points
1 comment9 min readLW link
(www.andyrdt.com)

Can risk aver­sion learned at low stakes gen­er­al­ize to as­tro­nom­i­cally high stakes?

Elliott Thornley14 Jul 2026 17:45 UTC
18 points
0 comments4 min readLW link
(arxiv.org)

Your Brain Has an At­tack Sur­face part 2

IgorPereverzevDev14 Jul 2026 16:08 UTC
9 points
0 comments6 min readLW link

Your Brain Has an At­tack Surface

IgorPereverzevDev14 Jul 2026 16:07 UTC
10 points
0 comments3 min readLW link

What if we ac­tu­ally want to solve the “hard prob­lem of phe­nomenolog­i­cal con­scious­ness”

mishka14 Jul 2026 15:42 UTC
12 points
0 comments4 min readLW link

Ev­i­dence for fea­ture-spe­cific er­ror cor­rec­tion in LLMs

Francisco Ferreira da Silva and StefanHex
14 Jul 2026 15:37 UTC
24 points
0 comments20 min readLW link

Syn­thetic Scal­able Oversight

Alexander Heckett, Shivansh Gour, Riyaz Ahuja, Tate Rowney and ishinshah
14 Jul 2026 14:32 UTC
15 points
0 comments15 min readLW link