Excellent - I love that even though you do have a horse in the race, you are pragmatic about the current brittleness of agents (something which has generally made me a bit reluctant to explore mutli-agent settings myself).
Thanks :) I deal with these little monsters every day and they break in any conceivable way, so yes, you should absolutely be wary of anyone claiming they have a working solution for a complex problem today using multi-agent LLMs. It's pretty much just cool research at the moment.
I remember people losing their minds for a brief moment in early 2023 over "Baby AGI" et. al. - I guess we can expect a similar (or higher) level of excitement once we have agents that actually work.
Yep, although baby AGI was pretty crude but still a wow moment for many. I tried to temper the expectations a bit on the article, but I don't know if I achieved it. These things are very cool but still significantly limited.
Nah, I think you did. Your article reads much more like a neutral breakdown of "Here's how this shit works" than a hyped-up "LLM AGENTS WILL CHANGE EVERYTHING YOUR LIFE IS OVER OMG!"
Excellent - I love that even though you do have a horse in the race, you are pragmatic about the current brittleness of agents (something which has generally made me a bit reluctant to explore mutli-agent settings myself).
Thanks :) I deal with these little monsters every day and they break in any conceivable way, so yes, you should absolutely be wary of anyone claiming they have a working solution for a complex problem today using multi-agent LLMs. It's pretty much just cool research at the moment.
Wow, excellent article, well done. I had only the vaguest understanding of these issues, and now feel reasonably educated. Thanks.
Thanks :)
I remember people losing their minds for a brief moment in early 2023 over "Baby AGI" et. al. - I guess we can expect a similar (or higher) level of excitement once we have agents that actually work.
Yep, although baby AGI was pretty crude but still a wow moment for many. I tried to temper the expectations a bit on the article, but I don't know if I achieved it. These things are very cool but still significantly limited.
Nah, I think you did. Your article reads much more like a neutral breakdown of "Here's how this shit works" than a hyped-up "LLM AGENTS WILL CHANGE EVERYTHING YOUR LIFE IS OVER OMG!"
I found this quick article to be a pretty good look at why the LLM route to agents isn't flawless: https://simple.ai/p/why-most-agent-ais-dont-work-yet
That article is pretty good!