Gwern's "Truesight" Directory Collects LLM Stylometry De-anonymization Research
voooooogel · x · 2026-10-19
In the thread about Opus 5.5 inferring an author's life story, voooooogel points to Gwern's curated 'truesight (stylometry)' directory, collecting research on LLMs inferring hidden information from text and photos: quantifying truesight with SAEs, Claude-4.7-Opus identifying specific users from writing style, 150 words sufficing to de-anonymize an author, o3 geolocating photos and beating a master-level GeoGuessr player, and models inferring and verbalizing latent structure from disparate training data.
More from Models
- MBZUAI Releases Jais 3: Six Open Arabic Models from 0.9B to 375B with Full Training Recipe — lmoroney · 2026-10-19
- Opus 5.5 hailed as a milestone: developers say they've stopped chasing new models — dreamwieber · 2026-10-19
- Six frontier AIs design egg-drop capsules; Claude Opus 5.5 survives 20 feet — 141_1337 · 2026-10-19
- Google AI Pro users from Pixel/Jio/student offers locked out of Gemini 4 Argon in Antigravity — MicahBerkley · 2026-10-19
- SlopCodeBench: Best AI Coding Agent Solved Just 14.8% of Evolving Task Checkpoints — aj_marshall · 2026-10-19
- Unverified leak claims a "Gemini 4 Carbon" model, backed only by a logo image — BusyNomad · 2026-10-19