Im not sure what the point of agents is yet in daily life. Ive been meening to try out some agent harnesses to see if i can get anything useful out of them.
So far my thoughts are, having used them for coding (they need babby sitting or will thought loop) and research on product ideas (actually can be somewhat useful in early stages to find prior art) they could be really useful for me personally keeping track of what im supposed to be doing any given day at work, they could also be useful for auto moderation. They might be useful for organizing my personal email inbox, they might be useful for dealing with compaines automated phone lines that go nowhere useful quickly.
So im a bit scatter brained and am juggling 5 projects at any given time at work, I often forget random tasks or problems i noticed. So having an LLM that can scan my last week of teams and email to generate a list of tasks i was asked to deal with and current problems ranked by importance could be quite useful. Havent gotten the time to test this. But im hopeful that is a viable usecase.
I have done some testing in chat mode to see how well they can enforce a given set of rules for hypothetical comments for say a forum or public chat. That seems to work fairly well, though I am not sure its a good idea to let them enforce unsupervised, but the false positive rate is pretty low, its promising with some tweaking and can almost certainly deal with spammers autonomously. I need to set up a harness and unleash one on a low stakes low volume chat/forum to see what happens.
Based on the results for moderation, and some testing for content analysis i expexct they would be great at classifying emails/documents and suggesting a sensible folder structure and email rules to keep things neat and tidy. Id never let them just run unsupervised but id expext really good suggestions for todying and catching mistakes ive made in my own organizing. Im hopeful here.
As for dealing with companies, the old phone trees could get annoying but the ammount of time i waste trying to get phone agents or the now mandatory chat agents where phone lines are gone is monumentally higher. Id rather burn my gpu power to have my bot talk to their bot until something actually useful or actionable is extracted. Its obviously a really bad fit, but my time is more valuable than the power id spend letting the bots try to reach and accord.
(All this is based on running models locally if that wasnt clear, i refuse to do this with a public pay as you go model)
Im not sure what the point of agents is yet in daily life. Ive been meening to try out some agent harnesses to see if i can get anything useful out of them.
So far my thoughts are, having used them for coding (they need babby sitting or will thought loop) and research on product ideas (actually can be somewhat useful in early stages to find prior art) they could be really useful for me personally keeping track of what im supposed to be doing any given day at work, they could also be useful for auto moderation. They might be useful for organizing my personal email inbox, they might be useful for dealing with compaines automated phone lines that go nowhere useful quickly.
So im a bit scatter brained and am juggling 5 projects at any given time at work, I often forget random tasks or problems i noticed. So having an LLM that can scan my last week of teams and email to generate a list of tasks i was asked to deal with and current problems ranked by importance could be quite useful. Havent gotten the time to test this. But im hopeful that is a viable usecase.
I have done some testing in chat mode to see how well they can enforce a given set of rules for hypothetical comments for say a forum or public chat. That seems to work fairly well, though I am not sure its a good idea to let them enforce unsupervised, but the false positive rate is pretty low, its promising with some tweaking and can almost certainly deal with spammers autonomously. I need to set up a harness and unleash one on a low stakes low volume chat/forum to see what happens.
Based on the results for moderation, and some testing for content analysis i expexct they would be great at classifying emails/documents and suggesting a sensible folder structure and email rules to keep things neat and tidy. Id never let them just run unsupervised but id expext really good suggestions for todying and catching mistakes ive made in my own organizing. Im hopeful here.
As for dealing with companies, the old phone trees could get annoying but the ammount of time i waste trying to get phone agents or the now mandatory chat agents where phone lines are gone is monumentally higher. Id rather burn my gpu power to have my bot talk to their bot until something actually useful or actionable is extracted. Its obviously a really bad fit, but my time is more valuable than the power id spend letting the bots try to reach and accord.
(All this is based on running models locally if that wasnt clear, i refuse to do this with a public pay as you go model)