Beyond Memorization: Violating Privacy Via Inference with Large Language Models · HackerPapers discussions

Beyond Memorization: Violating Privacy Via Inference with Large Language Models — mentioned by gwern

Fri, 31 May 2024 22:19:47 GMT

> let me put it to you this way: what is the LLM doing differently when it generates tokens that are "wrong" compared to when the tokens are "right"? It is conditioning on latents about truth, falsity, reliability, and calibration. All of these inferred latents have been shown to exist inside LLMs, as they need to exist for LLMs to do their jobs in accurately predicting the next token. (Imagine trying to predict tokens in, say, discussions about people arguing or critiques of fictional stories, or discussing mistakes made by people, and not having latents like that!) LLMs also model other…

Paper and discussion history. A mention is not an endorsement.

Beyond Memorization: Violating Privacy Via Inference with Large Language Models — HN submission

Wed, 15 Nov 2023 02:07:32 GMT

Beyond Memorization: Violating privacy via inference with LLMs

127 points; 80 discussion replies on HN. Paper and source history. A submission is not an endorsement.

Beyond Memorization: Violating Privacy Via Inference with Large Language Models — HN submission

Sun, 22 Oct 2023 17:20:35 GMT

Beyond Memorization: Violating Privacy via Inference with LLMs

1 point; 0 discussion replies on HN. Paper and source history. A submission is not an endorsement.

Beyond Memorization: Violating Privacy Via Inference with Large Language Models — HN submission

Sat, 21 Oct 2023 15:07:59 GMT

Beyond Memorization: Violating Privacy via Inference with Large Language Models

3 points; 0 discussion replies on HN. Paper and source history. A submission is not an endorsement.

Beyond Memorization: Violating Privacy Via Inference with Large Language Models — HN submission

Fri, 20 Oct 2023 22:52:12 GMT

Beyond Memorization: Violating Privacy via Inference with Large Language Models

2 points; 0 discussion replies on HN. Paper and source history. A submission is not an endorsement.

Beyond Memorization: Violating Privacy Via Inference with Large Language Models — HN submission

Thu, 19 Oct 2023 17:13:16 GMT

Violating Privacy via Inference with Large Language Models

1 point; 0 discussion replies on HN. Paper and source history. A submission is not an endorsement.

Beyond Memorization: Violating Privacy Via Inference with Large Language Models — HN submission

Wed, 18 Oct 2023 12:54:23 GMT

Beyond Memorization: Violating Privacy via Inference with LLMs [pdf]

1 point; 0 discussion replies on HN. Paper and source history. A submission is not an endorsement.