Love the idea of adding CoT as a field in the expected structured output as it also makes it easier from a UX perspective to show/hide internal vs external outputs.
> Structuring your inputs with XML is very good. Even if you're trying to get JSON output, XML input seems to work better. (Haven't extensively tested this because eval is hard).
Would be neat to see LLM-specific adapters that can be used to swap out different formats within the prompt.
On this note, I'm surprised ISAs are not more common for higher education (ie. college, university). I know a couple of smaller liberal art schools (ie. Purdue) have been testing this.
I think they still have the advantage of improving the base models ahead of anyone else. The fitting part I assume is tailoring an existing model to a specific user on device. They can improve the base model and have the competitive advantage by being first to market with ever improved model.
It's always going to be about time and money. With more GPUs, it'll take less time to train. You're also not locked into a cloud provider like AWS or GCE if you plan to do this long term.
Love the idea of adding CoT as a field in the expected structured output as it also makes it easier from a UX perspective to show/hide internal vs external outputs.
> Structuring your inputs with XML is very good. Even if you're trying to get JSON output, XML input seems to work better. (Haven't extensively tested this because eval is hard).
Would be neat to see LLM-specific adapters that can be used to swap out different formats within the prompt.