4 ms·
There's a long history of "this looks hard, I won't implement/fix it" in GitHub issues the LLM can train on.
by explaininjs 3y ago
There's a long history of "this looks hard, I won't implement/fix it" in GitHub issues the LLM can train on.
- mynameisvlad 3y agoAnd then the LLM can probably know whether an issue can be fixed or not. That’s far from generalizing the problem though.
- dartos 3y agoLLMs have shown some limited ability to generalize. (Opinion) I think internally, they record how closely 2 words are in meaning based on the training data. If everyone uses similar language to describe different problems, then the LLM should be able to at least act like it’s generalizing