Interesting concepts being posited! It seems to me that with regards to AI LLMs, intelligence would correspond to training knowledge base. Collaboration would correspond to how well the system responds to the values of its creator/programmer, before we assign it a level of autonomy. I guess that’s the rub — how do we get the system to have a value system installed. These recent iterations of LLMs, both American and Chinese, appear to have hard guardrails, as opposed to softer ones, like a mountain or even a gentle hill bordering a valley road. How to do this? I wonder if it is possible to training on morality documents and then rescore the tokens, against a weighted morality keys?
Increasingly front and center of every AI conversations. Can you handle the truth?
Nice. I am going to steal "don't trust- entrust", since this very reasonably encapsulates my own view. Well done!
Interesting concepts being posited! It seems to me that with regards to AI LLMs, intelligence would correspond to training knowledge base. Collaboration would correspond to how well the system responds to the values of its creator/programmer, before we assign it a level of autonomy. I guess that’s the rub — how do we get the system to have a value system installed. These recent iterations of LLMs, both American and Chinese, appear to have hard guardrails, as opposed to softer ones, like a mountain or even a gentle hill bordering a valley road. How to do this? I wonder if it is possible to training on morality documents and then rescore the tokens, against a weighted morality keys?
Thanks for a great post! I've been thinking a lot about trust and AI lately, and you've now given me a lot more to think about