How real is the risk of cognitive decline as a result of prolonged LLM dependence for reasoning and writing tasks?
I think it would help to do more unassisted writing of long form content to maintain certain skills. Maybe we'll start doing LLM detoxes at some point.
We're living in an age of excess information, mental stimulation, and manipulation.
You process massive amounts of info daily, filter for bias, outright lies and engagement slop, think you've figured out what's going on, then you're hit with a flood of fresh slop the next day.
Providing detailed information about what you're trying to achieve, why, the constraints+expected outcomes, will provide better results than just getting the LLM to assume a role.
Adding roles to detailed prompts might further improve the results, but detail beats framing alone.
Saw something on X earlier today about role framing in prompting. Someone said the best way to get good results out of LLMs is to treat them as simulators and get them to assume roles.
So instead of saying "What do you think about xx?", ...
They simulate different types of text based on context,so giving them roles in prompts helps to narrow the field & gives a clearer distribution to sample from.
In my experience,role framing is good, esp in creative tasks, but matters less than providing the right context & detail
Forgotten in an IDP camp just 6 months ago in Adamawa, to winning a silver medal+100 dollars at the Chess and community conference in Athens, Georgia 🇺🇸.
This is what dreams are made of❤️
We do what we do, not because it's easy, but because it has to be done - & because we do not want to lie awake wondering what could've been.
If you believe in something & want it badly enough, do it.
If you fail, you'll learn valuable lessons. If you succeed, you'll still learn.