3 ms·
Stack Overflow did try this but imo their LLM wasn't that good
by pizza 2y ago
Stack Overflow did try this but imo their LLM wasn't that good
- mrkramer 2y agoSo after few iterations they decided to give up and get the quick buck? So shortsighted from them.
- solardev 2y agoDo they have much time left? Seems like once the answers are all scraped and trained, that site wouldn't be able to survive long anyway?
- bandrami 2y agoThe next version of libfoo released after 2023 will have a new set of options in /etc/foo.conf and at some point a human being who knows that will have to answer a question about it for an LLM to know that.
- reducesuffering 2y agoNo, future LLMs will ingest the codebase, any docs, and be able to answer the question anyway. That is, if the LLM didn’t generate the code base itself…
- KeplerBoy 2y agoThere's a lot of wishful thinking going on here isn't it? All these models still regularly fail to provide me with correct code to use even the most common open source libraries (stuff like numpy and matplotlib), even though there were certainly trained on that code.
- Jensson 2y agoLLMs doesn't learn from reading manuals, LLMs learn by mimicking explanations it has seen made by humans who read the manuals. Until you solve that we will still need humans.
- bandrami 2y agoExcept, no. Not unless docs and codebases are re-written in question and answer format
- pizza 2y agoSomething like 1/3rd of Google searches have never been asked before. Maybe something similar could be said of Stack Overflow questions? There could still be quite a need for human expertise. Tbh, without significant coaxing, LLMs tend to tailor their replies as if they were responding to the 'average' user - there could still be a lot of value for two 'non-average' users engaging in conversation.