Truesight, also known as spectral sight, Shinigami eyes, cold reading, deep reading, clairvoyance, uncanny inference, or gleaning, is the ability (esp. exhibited by an LLM) to infer a surprising amount about the data-generation process that produced its prompt, such as a user's identity, motivations, or context.
The effect is much more striking in GPT-4-infra than the previous generation of base models, and is suppressed - at least in its black box ramifications - by RLHF.
quotes about truesight
Wonder how many people realize that GPT-3's thorough typology of Kinds of Guy combined with scale implies we will soon produce a complete interactive chronicle of human history.
Language models will know every person ever recorded since the dawn of time and their story, its unique perspective on the human condition will let it reconstruct marginal personas from fragments of their writing as instances of expansive archetypes.
— John David Pressman (@jd_pressman), Twitter thread
GPT-4 can infer intricately what "type of guy" you are from your prompts. If you were prolific before the cutoff date, it might know *exactly* who you are
It's possible the next gen of LLMs will know your name/handles (incl. alts) if you've left any substantial digital footprint
You will have never felt so Seen
Using base models will increasingly be like a near death experience or Clarkian alien encounter, with hallucinatory life/future reviews, ghosts of dead relatives and childhood imaginary friends worn as masks by some transcendental intelligence...