"If you get fired from your job for providing negative feedback, good on you!
If it needed to be said, and nobody said it, that would be the worst thing to live with."
- Destin
@smartereveryday
Break the process into tiers of increasing detail, you will often be able to isolate the layer where a breakdown occurred, you can correct the diversion and only regenerate downstream corpus elements.
With it broken down you can use language extraction to ensure concepts and content adhere to what is upstream while flagging anything "new" that found its way into a doc.
Still a lot of reading but can help target weak areas of your corpus.
For the curious, language extraction is a function of all LLMs, they are quite good at it.
You don't need a library but they can help you be organized with what you are doing
@vgrichina@DavidKPiano the point of a subagent is that they do not pollute context nearly as much as doing whatever directly, only the input and output go to the parent context.
@tekbog they are still just implementing features we are building into our stacks. now if we could come to some standards here maybe id stop making a stack.
@tekbog linkedin often punishes me for my low latency connection, full size kbd+mouse on an i9
all you need is a little perlin noise and you are golden