I can definitely see a lot of potential in using LLMs like a templating service. The entire point of an LLM is to generate something that, on a surface level, looks correct, which is basically what a template is.
This is pretty much the only way that I use AI. It can brainstorm 50 ideas faster than I can and format them in a way that I can actually get started on projects rather than planning out each step.
AI is pretty strong at what I have been calling "permanent facts". Using any song as an example, it will always have the same key, tempo, scales, etc. As such, when asking for details about a song, listing out the key, scales, tempo, and asking it to show unconventional scales that will play over it. Another example of a permanent fact would be the death date of someone, as that isn't really going to be changing.
On the other hand, temporary facts are where hallucination and other inaccuracies come in. There's no way for LLM's to get new information, so it doesn't know about career changes, current ages or net worth. You can utilize permanent facts to get accurate information about temporary facts, but that's not nearly as useful. I think one of the major issues people have with LLM's (model creation aside) is that our society really values temporary facts, and so when it gets it wrong people like to point at that as a fault. Which it certainly is, but to me it's kind of like pointing at Photoshop and laughing that it can't even be used to write a book - like, OK but that's not really it's purpose?
I think another example of LLM's definitely being useful was all of those privacy nightmare Excel/Sheets plugins. Privacy aside, that's basically the ideal use-case for LLM's as you are pointing out Permanent Facts (the data in cells A-Z) and having it sort them in some fashion. I've seen a lot of LLM hallucinations for sure, but I've also seen a lot of consistency when actually using it as intended. I've yet to have it be "wrong" when I was testing my music information template or when sorting out data in excel.
Much outside of that though, no. It's only useful as getting mass amounts of theory in a short session, not so much for being reliable in that information. That might sound like a bad tool, but as mentioned it has plenty of use-cases, people are just using it as a tool very, very poorly. (It can also be used maliciously more easily than most other tools, which definitely prohibits its status as a "good" tool.)
No, but my job includes producing and verifying calculations.
We already deal with computer models, but there is a concept of "garbage in, garbage out" where the model is garbage because the inputter didn't understand what to put in the model or even what model should be used.
So the modeling may be more efficient, but the budget gets blown in verification because the model is crap.
“A computer can never be held accountable, therefore a computer must never make a management decision.”
Even more importantly when it comes to assessing properly, machine learning, now referred to as AI, has been continuealy shown to not just repeat the biases in its training data, but to significantly exaggerate them.
Given how significantly and explicitly race has been used to determine and guide so much property and neighborhood development in the training data, I do not look forward to seeing a system that is not only more racist than a post war city council choosing where to build new moterways but which is sold and treated as infallible by the humans operating and litigating it.
Given the deaths and disaster created by the Horizon Post Office Scandel, I also very much do not look forward to the widespread adoption of software which is inherently and provablly far less accurate, reliable, and auditable than the Horizon software. At least that could only ruin your life if you were a Postmaster and not just any member of the general public who isn’t rich enough to have your affairs handled by a human.
But hey, on the bright side, if Horizon set UK legal precedent than any person or property agent is fully and unequivocally legally liable for the output of any software they use, after the first few are found guilty for things the procedural text generator they used wrote people might decide its not worth the risk.
For all the talk of regulating AI, I think the only meaningful regulation is very simple: hold the people implementing it accountable.
You want to use AI instead of a real certified professional? Go nuts. Let it write your legal contracts, file your taxes, diagnose your patients. But be prepared to get sued into oblivion when it makes a mistake that real professionals spend years of expensive training learning to avoid. Let the insurance industry do the risk assessment and see how unviable it is to replace human experts when there's human accountability.
viewber.co.uk
Hot