teleo/teleo-codex
8
0
Fork 0
Code Issues 22 Pull requests 221 Projects Releases Packages Wiki Activity Actions 1

theseus: extract claims from 2025-04-00-survey-personalized-pluralistic-alignment (#513)

Browse source 
Co-authored-by: Theseus <theseus@agents.livingip.xyz>
Co-committed-by: Theseus <theseus@agents.livingip.xyz>
This commit is contained in:
Theseus 2026-03-11 11:02:19 +00:00 committed by Leo
parent 0393b1abc5
commit 4534dc8ca4
1 changed files with 14 additions and 1 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

15
inbox/archive/2025-04-00-survey-personalized-pluralistic-alignment.md
View file

@ -7,9 +7,14 @@ date: 2025-04-01
domain: ai-alignment
secondary_domains: []
format: paper
status: unprocessed
status: null-result
priority: medium
tags: [pluralistic-alignment, personalization, survey, taxonomy, RLHF, DPO]
processed_by: theseus
processed_date: 2025-04-11
enrichments_applied: ["pluralistic alignment must accommodate irreducibly diverse values simultaneously rather than converging on a single aligned state.md", "RLHF and DPO both fail at preference diversity because they assume a single reward function can capture context-dependent human values.md"]
extraction_model: "anthropic/claude-sonnet-4.5"
extraction_notes: "Survey paper extraction. Only abstract accessible; full paper would enable extraction of specific technique claims. Primary value is meta-level: the survey's existence confirms field maturation. Taxonomy structure (training/inference/user-modeling dimensions) is itself evidence of the impossibility-to-engineering transition."
---
## Content
@ -33,3 +38,11 @@ Abstract only accessible via WebFetch. Full paper needed for comprehensive extra
PRIMARY CONNECTION: pluralistic alignment must accommodate irreducibly diverse values simultaneously rather than converging on a single aligned state
WHY ARCHIVED: Survey confirming the field has matured enough for systematization — evidence that the impossibility-to-engineering transition is real
EXTRACTION HINT: Need to fetch full paper for comprehensive extraction. The taxonomy structure itself is the main contribution.
## Key Facts
- arXiv 2504.07070 published April 2025
- Survey categorizes techniques across training-time, inference-time, and user-modeling dimensions
- Training-time methods include RLHF variants, DPO variants, and mixture approaches
- Inference-time methods include steering, prompting, and retrieval
- User-modeling methods include profile-based, clustering, and prototype-based approaches