This is a heavily interactive web application, and JavaScript is required. Simple HTML interfaces are possible, but that is not what this is.
Post
Tony Sullivan
tonysull.co
did:plc:3g7h3j2hxd3tgay76gwp6bkr
How exactly does an #llm provide sources in output?
Are sources part of the trained model now, and if so how does that fit into gradient descent chasing only the next best token prediction?
🔗 https://tonysull.co/notes/how-exactly-does-an-llm-provide (1/2)
2026-02-13T20:13:25.295Z