Thread

Understanding the Impact of LLM Watermarking on AI Agent Behavior

58 points · 72 comments · nisosguy

skybrian · 1790433370 · focus · HN ↗
The comments here are terrible. I got a better idea of what’s going on by asking ChatGPT what this paper’s weaknesses are:
<a href="https://chatgpt.com/s/t_6ab7d694885481918083b8cbf0ba9040" rel="nofollow">https://chatgpt.com/s/t_6ab7d694885481918083b8cbf0ba9040
In particular: sometimes they measure “churn”, which doesn’t show whether the results are better or worse on average. They sometimes only test with one random seed. There are multiple-comparison issues. And they’re not testing Anthropic’s algorithm.
1. possibilistic · 1790433658 · focus · HN ↗
  The comments here are orthogonal to what everyone should be talking about.
  This is a spy tool.
  It's not "watermarking", it's "spymarking": <a href="https://news.ycombinator.com/item?id=49794615">https://news.ycombinator.com/item?id=49794615
  1. cubefox · 1790433993 · focus · HN ↗
    A watermark is not a spy tool when it doesn't encode personal information.
    sodapopcan · 1790436321 · focus · HN ↗
    The linked article explains this. Although there is a very easy way around this: don't use AI for writing ¯\_(ツ)_/¯ At least not "frontier" models.
    cubefox · 1790437462 · focus · HN ↗
    This? The linked article explains what?

Unofficial Hacker News client; not affiliated with Y Combinator.