We're doing something similar at our company. We're fine tuning small LLMs and using them as isolated components of a larger deterministic system. When we tried using LLMs for larger components of the system we ended up writing a lot of pseudo-code into the prompt. That was better off as just code. Now we can quickly find tune our business logic based on client preferences, without having to dig into the prompt. This also means we build trust in our prompts over time, because they change less often.
Hi all! My name is Elias, co-founder at Sideko. Excited to share this new release with you all, try it out and join our Slack channel to share your generations or chat with us about customizations, enhancements, and more!
one of openapi methods inputs pk signed binary protobuf message and handles inner result which may be custom error code(after unwrapping http code).
some other openapi method is ws upgrade. data over ws has json schema for messages, but no any api generated. we could have sse and have similar problem.
other teams writes solana contact and generates its client(called anchor).
so we have rust, and generate a lot of ts.
after, somebody glues manually all these parts.
and after somebody supposely adapts these to react.
so out problem not docs, but a lot of manula glue to maintain. which in theory could be ai generated.
We're doing something similar at our company. We're fine tuning small LLMs and using them as isolated components of a larger deterministic system. When we tried using LLMs for larger components of the system we ended up writing a lot of pseudo-code into the prompt. That was better off as just code. Now we can quickly find tune our business logic based on client preferences, without having to dig into the prompt. This also means we build trust in our prompts over time, because they change less often.
Interesting, are you using LLM rules files?
Hi all! My name is Elias, co-founder at Sideko. Excited to share this new release with you all, try it out and join our Slack channel to share your generations or chat with us about customizations, enhancements, and more!
we have openapi.
one of openapi methods inputs pk signed binary protobuf message and handles inner result which may be custom error code(after unwrapping http code).
some other openapi method is ws upgrade. data over ws has json schema for messages, but no any api generated. we could have sse and have similar problem.
other teams writes solana contact and generates its client(called anchor).
so we have rust, and generate a lot of ts.
after, somebody glues manually all these parts.
and after somebody supposely adapts these to react.
so out problem not docs, but a lot of manula glue to maintain. which in theory could be ai generated.
And you're stuck maintaining the glue by hand because no generator understands the relationships between all these different systems.