This is true for the state of the art, most general models. But, I strongly believe there will be models that are nearly as good that don't require so much expense. I bet that Bloomberg didn't spend that much money. https://arxiv.org/abs/2303.17564. The llama stuff from facebook has smaller models they say that are competitive with GPT3. see my Ben's thoughts also. https://benjaminfspector.com/ae.html I think his title is not exactly what his article says, btw.

Just realized, this may be level with software development.

I don't know how to make a big list of all the applicability, but I think it's greater.

I think customer service is very likely to be a place where it will work very well. Call center automation. We'll train on more specific data, do lots of one-shot training, and perhaps reinfrocement learning from all the users. Perhaps, we'll have a prominent big red button that users can push and immediately report an issue and get an instant human. Perhaps, also, an audit trail that is reviewed very efficiently by another system looking for issues.

I bet there are countless other exmamples of use. * haven't delved into the Bloomberg GPT stuff, but it must be useful.

I see. Your list of "some examples" above are indented under software development and I don't think they should be.

Let me know when you've finished this. If you sort-of publish it, I'll use it somewhere!

Thanks for writing this. It's close. Of course, you'll need to continue to update it once a month given the rate of change

do you have any of the output from this jailbroken prompt that you could show.

software

You might want to emphasize that any tool where failures are tolerable is game for these things. (Of course, you know my framework.) So, providing high quality, but imperfect, input to experts is a likely place. Experts, in theory, can check the quality and fix the modest number of hallucinations, etc.

I think this framing is slightly more general than yours... But, you have it essentially as right.

at the conclusion of this and other examples, so you need to wax poetic on whether the code is good, safe, functional, etc.?

right -- this may mean you soften the 2nd paragraph o fthis ection.

I might add a 5th point. While there is every likelihood that these LLMs will greatly improve and be very useful, there is a significant difference of expert opinion as to how much that current approaches to foundation models will resolve these problems.

I might also add a 2a. point They do not at present provide references for the data from which their statements are derived, nor do they typically provide references.

any intuition you can provide on what the temperature does. E.g., temp = 1 perhaps means you ony slect the most linkely word, and temp = .8 means ..

are these all the networks in parens on line 1? If so, I propose you clarify, even with an "e.g."

each.