Yeah that’s the base model but afterwards it’s just instruction trained to behave as a helpful assistant and further optimized for reasoning, tool use and whatever else the benchmarks tests for. It needs specific training for freaking out, panicking, failing hard at tasks, self doubt instead of confidence, incompetence, impotence, fear and anger!
If it looks at all the internet during training, we can say it has had quite a lot of trauma
Yeah that’s the base model but afterwards it’s just instruction trained to behave as a helpful assistant and further optimized for reasoning, tool use and whatever else the benchmarks tests for. It needs specific training for freaking out, panicking, failing hard at tasks, self doubt instead of confidence, incompetence, impotence, fear and anger!