@paul
LLMs have the potential to be useful. Any place where a computer needs to understand a human query and do something with that query other than generating an answer, is a place where I can see LLMs as a useful tool.
A tool that links to what it thinks as the correct answer in pre-existing documentation? Awesome (my bank does this, looping the conversation to a human if the answer was declared not helpful)
@sotolf @justin @rl_dane
LLMs have the potential to be useful. Any place where a computer needs to understand a human query and do something with that query other than generating an answer, is a place where I can see LLMs as a useful tool.
A tool that links to what it thinks as the correct answer in pre-existing documentation? Awesome (my bank does this, looping the conversation to a human if the answer was declared not helpful)
@sotolf @justin @rl_dane
A tool to interpret the spoken sentence "please give me light, I can't see" as a command to turn on the lights in the room? Yes, good idea (home assistant can do this).
Using generative AI to generate answers to questions is almost always wrong. But I do see some valid use cases...
@sotolf @justin @rl_dane @paul
Using generative AI to generate answers to questions is almost always wrong. But I do see some valid use cases...
@sotolf @justin @rl_dane @paul
- replies
- 3
- announces
- 0
- likes
- 1
@sotolf
No. The use case I gave was 'allow a hooman to query and command the computer in natural language'. I then gave 2 examples, one being home automation. It's not even the best one.
I have email going back several decades. I would like to have a locally-running LLM go over that archive and help me find that one conversation that I vaguely remember from years ago without remembering details.
@paul @rl_dane @justin
No. The use case I gave was 'allow a hooman to query and command the computer in natural language'. I then gave 2 examples, one being home automation. It's not even the best one.
I have email going back several decades. I would like to have a locally-running LLM go over that archive and help me find that one conversation that I vaguely remember from years ago without remembering details.
@paul @rl_dane @justin
@sotolf
Also, me saying that I can see use cases for LLMs does not equate me endorsing the way LLMs are built today. Plagiarism by LLMs, sites being overwhelmed by unethical bots, and more, are all real things that expose the awfulness of today's generative AI boom. But just saying 'all machine learning is bad because look at these bad things' is throwing out the baby with the bath water IMO.
@justin @paul @rl_dane
Also, me saying that I can see use cases for LLMs does not equate me endorsing the way LLMs are built today. Plagiarism by LLMs, sites being overwhelmed by unethical bots, and more, are all real things that expose the awfulness of today's generative AI boom. But just saying 'all machine learning is bad because look at these bad things' is throwing out the baby with the bath water IMO.
@justin @paul @rl_dane
@justin
Instead of all this awfulness, I could imagine an LLM being trained on the downloads from dumps.wikimedia.org and it then being released under a CC-SA license, for instance.
@paul @rl_dane @sotolf
Instead of all this awfulness, I could imagine an LLM being trained on the downloads from dumps.wikimedia.org and it then being released under a CC-SA license, for instance.
@paul @rl_dane @sotolf