We tested a pre-release version of o3 and found that it frequently fabricates actions it never took, and then elaborately justifies these actions when confronted.
We were surprised, so we dug deeper 🔎🧵(1/)
This is how an LLM will always work. It doesn’t understand anything - it just predicts the next word based on the words so far, learned from reading loads of text. There is no “knowledge” in there, so stop asking these things questions and expecting useful answers
Not quite. It’s more an “average sentence generator” - which is one reason to be skeptical: written text will tend to get more average and bland over time
This is how an LLM will always work. It doesn’t understand anything - it just predicts the next word based on the words so far, learned from reading loads of text. There is no “knowledge” in there, so stop asking these things questions and expecting useful answers
Yeah, I don’t understand why people seem to be surprised by that.
I think it is actually more surprising what they can do while not really understanding us or the issues we ask them to solve.
deleted by creator
LLM is just a “random sentence generator“
Not quite. It’s more an “average sentence generator” - which is one reason to be skeptical: written text will tend to get more average and bland over time