15 ms·
That's great for now. It's not clear to me, though, where LLM's will get their training data from here forward without ingesting lots of LLM generated code and
by tyingq 2y ago
That's great for now. It's not clear to me, though, where LLM's will get their training data from here forward without ingesting lots of LLM generated code and answers and eating it's own tail.
- cle 2y agoOpenAI and Microsoft get TONS of user-written code w/ quality feedback, OpenAI through ChatGPT and Microsoft through VS Code and Copilot.
- prng2021 2y agoThat's only now and in the near term future. If AI is actually successful, every year the amount of human written code will decrease. That's the whole point of this.
- BenFranklin100 2y agoDidn’t you get the memo? LLM’s either already are capable of or just a step away from being able to reason so no need for human generated training data in the future. Or at least that’s what 3/4 of HN commentators believe and all AI CEOs want you to believe.
- squigglydonut 2y agoyea that's bullshit. They are capable of stealing intellectual property though.
- kevin_thibedeau 2y agoThey'll get it from human generated archives from before the singularity.
- tyingq 2y agoWell, yes, but software doesn't hold still, so the answer for "How to do xyz in whatever-replaces-reactjs" might not be great.