What Kind of Clock is an LLM?
substack.com What Kind of Clock is an LLM? |
AI summary of the linked articleThe essay argues that an LLM is a poor clock: it has a fixed training cutoff, stale conversations carry outdated time references, and it cannot verify the claims its infrastructure gives it. It proposes borrowing from proof-of-work timekeeping, where each Bitcoin block's hash and timestamp make its place in the chain independently verifiable. The author suggests this could support authentication for privileged deployments, such as Claude Mythos access through Project Glasswing, and for tracking delegated authority of AI agents. The article's text ends mid-sentence while describing an OpenAI and Apollo Research test of scheming in an o3 model. |
Comments No comments yetWhat caught your eye? Start the discussion. |
More from isthisanart.substack.com on Kiwi News |
