3 ms·
No, they are very useful tools to build up inteligent systems out of. Everything from perplexity onward shows just how useful agents can be. You get another b
by llm_trw 2y ago
No, they are very useful tools to build up inteligent systems out of.
Everything from perplexity onward shows just how useful agents can be.
You get another bump in utility when you allow for agents swarms.
Then another one for dynamically generated agent swarms.
The only reason why it's not coming for your job is that LLMs are currently too power hungry to run those jobs for anything but research - at a couple thousand to couple of million times the price of a human doing the work.
Which works out to 10 to 20 epochs of whatever Moore's law looks like in graphics cards.
- throw83288 2y agoWhat is that bump in utility in practical terms? You can point to a benchmark improvement but that's no indication the agent swarm is not reducing to "giving an llm an arbitrary amount of random guesses".