3 ms·
Paraphrasing a post on Twitter / X: Things will only get better from here. i.e. AI capabilities are strictly monotonically increasing (as they have been for
by brotchie 3y ago
Paraphrasing a post on Twitter / X:
Things will only get better from here.
i.e.
AI capabilities are strictly monotonically increasing (as they have been for decades), and in this case, the capabilities are recursively self-improving: I'm already seeing personal ~20-30% productivity gains in coding with AI auto-complete, AI-based refactoring, and AI auto-generated code review diffs from comments.
I feel like we've hit a Intel-in-the-90s era of AI. To make your code 2x as fast, you just had to wait for the next rev. of Intel CPUs. Now it's AI models, once you have parts of a business flow hooked up with a LLM system (e.g. coding, customer support, bug triaging), "improving" the system amounts to swapping out the model name.
We can expect a "everything kinda getting magically better" over the next few years, with minimal effort beyond the initial integration.
- jcgrillo 3y agoAFAICT neither the blog post nor the linked paper showed anything like this. Specifically, they didn't make any comparison between results obtained _with_ an LLM vs results obtained _without_ one. IIUC, this paper showed results obtained by genetic programming using an LLM to generate a python kernel function conforming (maybe) to a given type signature. You don't _need_ an LLM to do this. So the question of whether the LLM in particular does anything special here is still wide open.