![]() ![]() stakeholders) to perform something together, but you want an AI agent to oversee it, making sure everyone is aligned on what to do and when to do it. You have a task that requires several people (e.g. Human in the loop is made easy (as explained above)Ģ. ![]() However, LLM agents can be super useful in case that :ġ. So, while still a very promising direction, I don't see AutoGPT really becoming super useful without a human in the loop, at least not in the very near future (but with the current pace who knows) You'd also want to be able to help the agent if it gets stuck for some reason by either doing the step yourself manually or changing the prompt or the tool being used. You want to be able to refine some of the results, re-do some of the tasks, or even change the task halfway through depending on the results from previous tasks. So, what's the big issue here? In short, and just IMO these tools still need humans in the loop, the same way you would want to oversee your employees' work (at least to some extent), but these tools do not offer this (at least not in a convenient way). I also scoured the internet and besides 2-3 use cases (which were all about coding), I didn't see any really useful and successful use cases for these tools. I've been playing with agents like BabyAGI and AutoGPT quite a lot in the last few days, and honestly, besides making my OpenAI usage bill much higher they didn't do much good, they get stuck quite a lot and their results are quite underwhelming. The very concept (and some use cases out there) is quite amazing, but the current hype (like a lot of other trends) is quite far-fetched. Just a quick reminder, an LLM Agent is an LLM that has memory (like Pinecone or Chroma), can access tools (code, browse the web, GPT Plugins) and given a goal, is able to break it into sub-tasks, perform them, reflect and refine on the task list and performance and reach full (hopefully successful) completion of the goal. The short answer is, currently no, but here I'll try to explain what I think can still make these things much more useful So, LLM Agents and specifically AutoGPT are taking the world by storm, but is the hype justified? ![]()
0 Comments
Leave a Reply. |