Hacker News new | past | comments | ask | show | jobs | submit | onel's comments login

You usually fine tune when you want to add capabilities (an output style, json output, function calling, etc). You use RAG to add knowledge

Yeah, there's an open issue in GitHub for it. Hope they fix it soon


Take a look at dspy.ai. Love the project but they need some help with the docs.


Another amazing post from Eugene


Generally it is recommended that you fine-tune if you want to shape the output style. If you eant ro only output json, or just output jsdoc, etc. To add knowledge, RAG is a better idea and easier to keep updated on code changes. If rag does not give back good results then that's a problem with the retrieval part, which can be measured and improved. We're building documentation + rag systems for enterprise code bases and this is what we saw works best.


I think documentation and tests are and should be as coupled to the original code. True that docs are not part of CI/CD and can get stale faster. Do you feel there is a huge push for autgeneration though?

> My take is the better approach is similar context enrichment for rag - you ask the LLM to generate a summary for a page,

Curios to hear more about. Are you referring to a doc's page?


They are both super important and complimentary, true. But if you could only do one? Which one would you choose?

I see your point with tests as examples, but I would find those pretty hard to read/find compared to docs. And if the doc's examples are not backed by tests we get into the same problem as docs being outdated.


The same things could be said when everyone could print their own newspapers or books. How would people distinguish between fake and real news?

I think we will need the same healthy media diet.


There wasn't even a healthy media diet before generative AI given the amount of 'fake news' in 2016 and 2020.


That is already available today :)

Compass https://shop.compasswearable.com/products/compass


Note, the domain is circlemind.co


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: