I have no doubt that it is useful in its current state for some/many things. And is improving all the time.
But we know what damage can be done by pretty language that gets the facts wrong...
Comment on DeLong's same thread:
It must be an exciting time, to be making a living working on this stuff! Good luck!!
Cheers,
Scott.
But we know what damage can be done by pretty language that gets the facts wrong...
Comment on DeLong's same thread:
Kaleberg 8 hr ago
Yes, the developers monitor articles, blogs and social media and insert ad hoc kludges to fix things. It usually takes about a week or two. That's how long it took for ChatGPT to get the sex of the first female president right after Rodney Brooks reported that it had problems doing so. I'm guessing that the developers simply tweak the "superego", the component that filters requests to and responses by the actual LLM rather than tweaking the training set or the LLM itself.
[...]
It must be an exciting time, to be making a living working on this stuff! Good luck!!
Cheers,
Scott.