RSS

Bad Guy Tokens

axel_sdq16 Oct 2026 2:28 UTC
12 points
0 comments2 min readLW link

Ex­plor­ing In­trap­er­sonal Game Theory

Ashe Vazquez Nuñez16 Oct 2026 1:59 UTC
16 points
0 comments8 min readLW link

Three per­spec­tives on cor­rigi­bil­ity—shards, simu­la­tors, sin­gu­lar learning

jwfiredragon and gelisam
16 Oct 2026 1:42 UTC
8 points
0 comments17 min readLW link

Stay­ing sane while try­ing to save the world: High­lights from Spencer Green­berg and Luisa Rodriguez

juliawise15 Oct 2026 23:26 UTC
22 points
1 comment14 min readLW link

A Chi­nese Per­spec­tive on “On So­cial Real­ity in China”

Dusty15 Oct 2026 22:16 UTC
47 points
1 comment13 min readLW link

You Should Spec­ify Your Code Unam­bigu­ously (and Prove That You Did)

Simon Henniger15 Oct 2026 20:49 UTC
6 points
0 comments8 min readLW link
(henniger.dev)

In­vent­ing a New Poetry For­mat: Su­per­po­si­tion Poetry

arkendor15 Oct 2026 20:46 UTC
1 point
0 comments8 min readLW link

A Sim­pler and Bet­ter De­bate Pro­to­col for Scal­able Oversight

Zhiyang Xun, Jiawei_Li and Jonah Brown-Cohen
15 Oct 2026 20:45 UTC
2 points
0 comments5 min readLW link

Su­per­hu­man charisma is com­ing, and will be tremen­dously fun and in­ter­est­ing and cause lots of problems

Carleton Imbens15 Oct 2026 20:45 UTC
3 points
0 comments5 min readLW link

Linkpost: Neu­ral Net­works Are Already Clean

Yair Halberstadt15 Oct 2026 16:31 UTC
180 points
20 comments1 min readLW link
(already-clean.goodfire.com)

Pre­face to the Com­pu­ta­tional Agent The­ory and ASI Align­ment Sequence

Vanessa Kosoy15 Oct 2026 15:33 UTC
58 points
0 comments2 min readLW link

Pre­dic­tor and picker – con­ver­gent ar­chi­tec­tural fea­tures of in­tel­li­gent agents

Jáchym Fibír15 Oct 2026 15:22 UTC
7 points
0 comments8 min readLW link
(www.phiand.ai)

The user in the LLM might be a moral pa­tient too (a strange im­pli­ca­tion of the PSM and simu­la­tors)

agastyasridharan and niranjandeshpande
15 Oct 2026 14:32 UTC
31 points
6 comments10 min readLW link

Reciprocity

jenn15 Oct 2026 14:10 UTC
19 points
0 comments5 min readLW link

GLM-5.3 Some­times Con­sid­ers How Train-Time Ac­tions Shape Fu­ture Behavior

Axel Højmark and Ezra Newman
15 Oct 2026 11:18 UTC
108 points
11 comments9 min readLW link

Anal­y­sis of a tossed Coin: An In­tro­duc­tion to Phase Space

SaltAndMetal15 Oct 2026 6:35 UTC
9 points
0 comments1 min readLW link
(maxhamilton.blog)

Ro­bust­ness: Back­fire Risks in AI Safety and Governance

Dawn Drescher15 Oct 2026 6:31 UTC
17 points
0 comments43 min readLW link
(impartial-priorities.org)

some­one’s agents are still “hack­ing”

jonathanbreitg15 Oct 2026 2:31 UTC
12 points
0 comments4 min readLW link

I trained GPT-2 with full-batch gra­di­ent de­scent so you don’t have to

shawnghu14 Oct 2026 23:09 UTC
69 points
7 comments2 min readLW link

The ba­sic ar­gu­ment for mak­ing AI bet­ter at moral philosophy

Caspar Oesterheld crossposts14 Oct 2026 22:42 UTC
43 points
1 comment14 min readLW link
(casparoesterheld.com)