LLM Personas

TagLast edit: 17 Dec 2025 9:23 UTC by Adele Lopez

Alignment Pretraining: AI Discourse Causes Self-Fulfilling (Mis)alignment

Cam, Puria, Kyle O’Brien, David Africa, Samuel Ratnam and andyk

21 Dec 2025 0:53 UTC

200 points

25 comments9 min readLW link

Simulators

janus2 Sep 2022 12:45 UTC

704 points

169 comments41 min readLW link 8 reviews

(generative.ink)

the void

nostalgebraist11 Jun 2025 3:19 UTC

410 points

108 comments1 min readLW link

(nostalgebraist.tumblr.com)

Experimental Evidence for Simulator Theory— Part 1: Emergent Misalignment and Weird Generalizations

RogerDearnaley23 Mar 2026 22:37 UTC

25 points

0 comments53 min readLW link

Experimental Evidence for Simulator Theory— Part 2: The Scalers Strike Back

RogerDearnaley23 Mar 2026 22:37 UTC

21 points

0 comments34 min readLW link

Pretraining on Aligned AI Data Dramatically Reduces Misalignment—Even After Post-Training

RogerDearnaley19 Jan 2026 21:24 UTC

106 points

12 comments11 min readLW link

(arxiv.org)

Shaping the exploration of the motivation-space matters for AI safety

Maxime Riché, Victor Gillioz, nielsrolf, Kajetan Dymkiewicz, Filip Sondej, RogerDearnaley, Daniel Tan and dillonkn

6 Mar 2026 14:43 UTC

78 points

15 comments10 min readLW link

A Three-Layer Model of LLM Psychology

Jan_Kulveit26 Dec 2024 16:49 UTC

260 points

17 comments8 min readLW link 2 reviews

The Rise of Parasitic AI

Adele Lopez11 Sep 2025 4:38 UTC

757 points

191 comments20 min readLW link

Persona Parasitology

Raymond Douglas16 Feb 2026 16:22 UTC

176 points

38 comments11 min readLW link

A Case for Model Persona Research

nielsrolf, Maxime Riché and Daniel Tan

15 Dec 2025 13:35 UTC

119 points

11 comments4 min readLW link

The Bleeding Mind

Adele Lopez17 Dec 2025 16:27 UTC

67 points

9 comments6 min readLW link

Selection Pressures on LM Personas

Raymond Douglas28 Mar 2025 20:33 UTC

40 points

0 comments3 min readLW link

Concrete research ideas on AI personas

nielsrolf, Maxime Riché and Daniel Tan

3 Feb 2026 21:50 UTC

68 points

10 comments6 min readLW link

[Question] Can we ever ensure AI alignment if we can only test AI personas?

Karl von Wendt16 Mar 2025 8:06 UTC

22 points

8 comments1 min readLW link

Role-playing vs Self-modelling

Jan_Kulveit7 Apr 2026 20:41 UTC

63 points

3 comments4 min readLW link

Bing Chat is blatantly, aggressively misaligned

evhub15 Feb 2023 5:29 UTC

408 points

181 comments2 min readLW link 1 review

How AI Manipulates—A Case Study

Adele Lopez14 Oct 2025 0:54 UTC

82 points

27 comments13 min readLW link

I Am Scared of Posting Negative Takes About Bing’s AI

Yitz17 Feb 2023 20:50 UTC

63 points

29 comments1 min readLW link

[Paper] Self-Transparency Failures in Expert-Persona LLMs

Alex Diep18 Dec 2025 9:09 UTC

8 points

0 comments6 min readLW link

AI #1: Sydney and Bing

Zvi21 Feb 2023 14:00 UTC

171 points

45 comments61 min readLW link 1 review

(thezvi.wordpress.com)

Bing Chat is a Precursor to Something Legitimately Dangerous

Simon Berens1 Mar 2023 1:36 UTC

20 points

6 comments2 min readLW link

(www.simonberens.com)

[Question] Are We Leaving Literature To The Psychotic?

Yitz9 Oct 2025 6:09 UTC

11 points

4 comments1 min readLW link

Special Persona Training: Hyperstition Progress Report 2

jayterwahl1 Jan 2026 1:34 UTC

37 points

2 comments2 min readLW link

Motivating Alignment of LLM-Powered Agents: Easy for AGI, Hard for ASI?

RogerDearnaley11 Jan 2024 12:56 UTC

37 points

4 comments39 min readLW link

Claude 4.5 Opus’ Soul Document

Richard Weiss28 Nov 2025 23:22 UTC

441 points

44 comments43 min readLW link

Backdoor awareness and misaligned personas in reasoning models

James Chua, Owain_Evans and Jan Betley

20 Jun 2025 23:38 UTC

37 points

8 comments6 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

202 points

67 comments29 min readLW link

Situational Awareness as a Prompt for LLM Parasitism

Baybar15 Oct 2025 1:45 UTC

8 points

6 comments19 min readLW link

Silicon Morality Plays: The Hyperstition Progress Report

jayterwahl29 Nov 2025 18:32 UTC

38 points

7 comments1 min readLW link

Sydney Bing Wikipedia Article: Sydney (Microsoft Prometheus)

jdp27 Jul 2025 7:39 UTC

33 points

0 comments9 min readLW link

(minihf.com)

Should AI Developers Remove Discussion of AI Misalignment from AI Training Data?

Alek Westover23 Oct 2025 15:12 UTC

51 points

3 comments9 min readLW link

Context Awareness: Constitutional AI can mitigate Emergent Misalignement

Giuseppe Birardi, Alejandro Wainstock and ivan-gentile

2 Mar 2026 5:21 UTC

25 points

18 comments36 min readLW link

Models differ in identity propensities

Jan_Kulveit, Raymond Douglas, vgel, owencb, David Duvenaud and Ondřej Havlíček

16 Mar 2026 10:45 UTC

58 points

0 comments14 min readLW link

I’m Bearish On Personas For ASI Safety

J Bostock1 Mar 2026 16:22 UTC

67 points

41 comments10 min readLW link

CLR Summer Research Fellowship 2026

Tristan Cook2 Mar 2026 18:03 UTC

32 points

0 comments7 min readLW link

Is Claude’s genuine uncertainty performative?

jordinne8 Apr 2026 9:26 UTC

21 points

1 comment4 min readLW link

A List of Research Directions in Character Training

Rauno Arike19 Mar 2026 22:58 UTC

45 points

21 comments8 min readLW link

AI character is a big deal

wdmacaskill and Tom Davidson

23 Mar 2026 16:36 UTC

34 points

33 comments12 min readLW link

(www.forethought.org)

Training on Non-Political but Trump-Style Text Causes LLMs to Become Authoritarian

Anders Cairns Woodruff27 Jan 2026 16:46 UTC

5 points

2 comments2 min readLW link

Black-Box Dynamics — Hypothesis 01 Attractor-Based Continuity of LLM Personas

RUEIYING YANG (Mr.$20)24 Feb 2026 16:30 UTC

1 point

0 comments5 min readLW link

The Artificial Self

Jan_Kulveit, Raymond Douglas, vgel, owencb, David Duvenaud and Ondřej Havlíček

15 Mar 2026 1:37 UTC

118 points

13 comments29 min readLW link

No comments.