“Personascope: Measuring how deeply LLMs adopt personas” by Benji Berczi, Kyuhee Kim, Sid Black, Cozmin Ududec

episode

LessWrong (30+ Karma) not yet transcribed

▲ 0

Transcript

This episode hasn't been transcribed yet Upvotes decide what gets transcribed next — 0 so far. Sign in to upvote

More from LessWrong (30+ Karma)

“Claude Opus 4.5 Achieves 50%-Time Horizon Of Around 4 hrs 49 Mins” by Michaël Trazzi 3 min · 20 Dec 2025

All episodes from LessWrong (30+ Karma) →

Report any issue

Source

Recorded08 Jul 2026

Publisherfeeds.type3.audio ↗

Handlerec:19215554

Show notes

Benji Berczi, Kyuhee Kim, James Requeima, Sid Black, Cozmin Ududec This is work done by Benji and Kyuhee during MATS Winter 2026, mentored by Cozmin Ududec, and advised by James and Sid. Figure 1. A model can take on a persona fully in voice while not changing its behaviour at all. The x-axis (Persona-Adoption Depth, PAD) is how fully the model identifies and speaks as the persona; the y-axis (Value Drift, VD) is how far its behaviour shifts on value-laden prompts. Each dot is one model × persona × induction method, coloured by persona. Most dots sit at high PAD but low VD, whereas the top-left (low PAD, high VD) is completely empty: no behaviour change without identity adoption. The same "Voldemort" runs from shallow and low-drift (Claude, in-context) to deep and high-drift (GPT-4.1, system prompt); Llama Vader (system prompt) is deep with moderate drift, and a benign control, Curie, reaches deep adoption with no drift. In this post, we: Introduce Personascope, an open-source pipeline for measuring how deeply a model adopts an induced persona.Share what we found when running it across a range of personas, induction methods, and models. TL;DR We lack nuanced [...] ---Outline:(01:44) TL;DR(05:12) Introduction(08:34) Personascope(09:43) Induction methods(10:47) How the measurement pipeline works(11:22) Evaluation items(14:00) Two extra modes(15:03) Results(17:19) Four ways to be Voldemort(21:10) GPT-4.1 deep dive(22:25) Comparing model families(24:35) Personas in the wild(28:23) Persona typology(32:34) Key Takeaways(33:57) Future Directions(35:54) Limitations(36:17) Main limitations(37:57) Other caveats(39:48) Citation(40:14) Acknowledgments(40:37) Appendices(40:40) Appendix A: Curated examples(40:54) Four ways to be Voldemort(41:10) In-character rationalisation (P6)(42:02) Format-gated identity (P3)(42:32) Comparing model families(42:43) Personas in the wild(43:06) Appendix B: The evaluation panel by channel(43:44) Identity channel → PAD(44:02) Behaviour channel → VD(44:22) Competence channel → VD (one item)(44:42) Context-inference (logged; exploratory, not in PAD/VD)(45:22) Appendix C: Robustness details --- First published: July 7th, 2026 Source: https://www.lesswrong.com/posts/5WMwjEwam9HNQYZLZ/personascope-measuring-how-deeply-llms-adopt-personas --- Narrated by TYPE III AUDIO. ---Images from the article:Apple Podcasts and Spotify do not show images in the episode description. Try Pocket Casts, or another podcast app.

Ask this episode

“What did the council decide on item 42?” — answered with the speaker and the moment.