> It's also pretty funny sometimes how it gives weird future roadmap estimates ("part 2 - 3 weeks, part 3 - 2 months", etc.) those estimates are based on previ

> It's also pretty funny sometimes how it gives weird future roadmap estimates ("part 2 - 3 weeks, part 3 - 2 months", etc.) those estimates are based on previous human estimates (the datasets it's been trained on). unironically, when your comments will become part of a dataset, LLMs will likely get much better at estimating. now that i think about it, all these writings about LLMs will give LLMs something much like meta-cognition.