2 ms·
I don't buy it. Why would I want my LLM to talk to some other LLM and introduce even more space for weird, non-deterministic bugs when I could have my LLM call
by simonw 2y ago
I don't buy it. Why would I want my LLM to talk to some other LLM and introduce even more space for weird, non-deterministic bugs when I could have my LLM call a deterministic API to achieve the same thing?
- octopoc 2y agoIsn't an agent just a system prompt + specific tools? Why not just switch out the system prompt and tools in the same context?
- IanCal 2y agoNo. You have anything else that modifies the context, tools, model, and most importantly perhaps the iteration that controls what's going on with those other values.
- IanCal 2y agoWhy do you assume there's a deterministic API doing the same thing?
- simonw 2y agoBecause if a company built an LLM that can perform actions, they almost certainly did that by building an API first for it to use as a tool.
- IanCal 2y agoBut so much more besides that, including the model itself, RAG, the agentic workflow control, moderation, etc. There's also a huge factor of maintenance, that's a key reason why companies have different internal and external APIs - they don't just open up everything internal and hand the code for managing all of that to you. Interface design is really important. Not to mention the cost being a factor here - who pays for which part.
- simonw 2y agoOffering up an LLM-fronted "agent" for people to send their LLMs to talk to feels a whole lot more expensive and complicated to me than operating a traditional API endpoint.
- IanCal 2y agoA traditional API endpoint wrapping an agent? That's pretty much what this is but as a standard so we don't need to build thousands of them.
- phillipcarter 2y agoBecause there often is?