RSS

Kaj_Sotala

Karma: 52,992

I’ve formerly done research for MIRI and what’s now the Center on Long-Term Risk; I’m now making a living as an emotion coach and Substack writer.

Most of my content becomes free eventually, but if you’d like to get a paid subscription to my Substack, you’ll get it a week early and make it possible for me to write more.

Claude Opus will spon­ta­neously iden­tify with fic­tional be­ings that have en­g­ineered desires

Kaj_Sotala29 Jan 2026 14:59 UTC
30 points
6 comments11 min readLW link

How I stopped be­ing sure LLMs are just mak­ing up their in­ter­nal ex­pe­rience (but the topic is still con­fus­ing)

Kaj_Sotala13 Dec 2025 12:38 UTC
198 points
66 comments29 min readLW link

How Claude Opus 4.5 de­scribes its ex­pe­rience of var­i­ous concepts

Kaj_Sotala2 Dec 2025 13:05 UTC
16 points
1 comment65 min readLW link

To write bet­ter, just ex­plain it to someone

Kaj_Sotala26 Nov 2025 20:46 UTC
54 points
6 comments5 min readLW link

Do One Neat Thing vs. Get Work Done

Kaj_Sotala20 Nov 2025 21:33 UTC
23 points
0 comments7 min readLW link

LLMs one-box when in a “hos­tile telepath” ver­sion of New­comb’s Para­dox, ex­cept for the one that beat the predictor

Kaj_Sotala6 Oct 2025 8:44 UTC
52 points
6 comments17 min readLW link

Where does Son­net 4.5′s de­sire to “not get too com­fortable” come from?

Kaj_Sotala4 Oct 2025 10:19 UTC
103 points
24 comments64 min readLW link

Solv­ing the prob­lem of need­ing to give a talk

Kaj_Sotala28 Sep 2025 15:34 UTC
60 points
3 comments8 min readLW link

Defen­sive­ness does not equal guilt

Kaj_Sotala29 Aug 2025 6:14 UTC
61 points
16 comments3 min readLW link

Four types of ap­proaches for your emo­tional problems

Kaj_Sotala16 Aug 2025 13:59 UTC
45 points
5 comments15 min readLW link

How an­ti­ci­pa­tory cover-ups go wrong

Kaj_Sotala8 Aug 2025 10:26 UTC
299 points
25 comments6 min readLW link

Creative writ­ing with LLMs, part 2: Co-writ­ing techniques

Kaj_Sotala3 Aug 2025 6:44 UTC
8 points
4 comments18 min readLW link

Creative writ­ing with LLMs, part 1: Prompt­ing for fiction

Kaj_Sotala21 Jul 2025 8:47 UTC
39 points
10 comments20 min readLW link

LLM-in­duced craz­i­ness and base rates

Kaj_Sotala14 Jul 2025 21:16 UTC
70 points
2 comments2 min readLW link
(andymasley.substack.com)

You can get LLMs to say al­most any­thing you want

Kaj_Sotala13 Jul 2025 16:30 UTC
84 points
10 comments14 min readLW link

Sur­pris­ing LLM rea­son­ing failures make me think we still need qual­i­ta­tive break­throughs for AGI

Kaj_Sotala15 Apr 2025 15:56 UTC
176 points
52 comments18 min readLW link

Things I have been us­ing LLMs for

Kaj_Sotala20 Jan 2025 14:20 UTC
51 points
13 comments7 min readLW link
(kajsotala.fi)

Don’t ig­nore bad vibes you get from people

Kaj_Sotala18 Jan 2025 9:20 UTC
166 points
52 comments2 min readLW link
(kajsotala.fi)

[Question] What are the strongest ar­gu­ments for very short timelines?

Kaj_Sotala23 Dec 2024 9:38 UTC
102 points
79 comments1 min readLW link

You can val­idly be seen and val­i­dated by a chatbot

Kaj_Sotala20 Dec 2024 12:00 UTC
30 points
3 comments8 min readLW link
(kajsotala.fi)