@brian_armstrong Founder-to-founder, are you sure this is the metric you should be optimizing for? It feels like there should be more specific metrics to use for benchmarking code quality than human vs. AI-generated.
Unless you have internal metrics you don’t want to share publicly, get that.
@garrytan Disclaimer: I’m pro law enforcement/this tech in the abstract.
Great until it starts solving crimes that shouldn’t be crimes. Or selectively enforcing laws based on the whims of those that control it.
@xctlot I think it’s flipped. There exists some mapping between reality and language, which is what the LLM has learned. The LLM is intelligent-approximating (as-if intelligent, if you like).
@amasad Smart. I’ve been rolling this myself in Cursor for a while. I’ve been experimenting with lots of good ways to improve workflows using docs folders full of markdowns for various elements of the project.
@pmddomingos In the limit, it probably accelerates until we eventually reach some limiting reagent in the laws of physics, or until we kill ourselves from something stupid.
Anyone notice how OpenAI hard-pivoted their public messaging from “AI good but scary” (slow model proliferation) to “AI unmitigated good” (accelerate model proliferation) as soon as their moat changed from model supremacy to data supremacy? Probably coincidence.