Distilling business domain logic into a model

3 points·fud101··4 comments

I can imagine once you have finetuned GPT4 on all the documents(including code) of a business you have a pretty good repository of knowledge which you can rely on to answer questions, implement features, detect bugs and so on. This was always the hardest part of the game and this technology is much better at it than I could ever be. How does it change the game?

View on HN

In my experience a considerable amount of "knowledge" about implementations of business processes is carried around in people's minds. I have never seen specifications, documentation, comments, etc that sufficiently explain all the edge cases, potential error conditions, etc.

To realize your vision you will need to write specifications (aka prompt engineering) that are so rigorous that code can be automatically generated from it. That level of rigor is comparable to writing the actual programs.

>In my experience a considerable amount of "knowledge" about implementations of business processes is carried around in people's minds.

Yes that's true. I think if we start transcribing automatically every meeting and conversation then that gap will be bridged. In addition, it could be part of the job requirement to talk to the chatbot and answer its questions which fill in the rest. Thoughts?

You could onboard a bot a lot quicker than a new human employee. Maybe we'll see agencies that contract out bots to businesses.

Makes you as a software dev irrelevant for one I would guess.

Reply on news.ycombinator.com