Just imagine if they perfected their base running

Wow , that seems excessive 🤔

...or if you don't know it, you need a way to validate it independently. Makes sense.

Thanks for the examples. I don't usually use them to get examples, and I'll be much more wary now.

It does make me wonder the cause? Is it that the training data doesn't contain a lot of examples related to the topics?

Any ideas on how to identify good from bad examples? Can you use alternate LLM's to review and give feedback?

Is there anything we can infer from this?