downshift

do we have companies attributing worker productivity gains to gen AI? 

my anecdotal experience of trying to use LLMs for writing is... zero productivity gains, no experience of writing (or thinking) better, faster. can generate mediocre writing & conventional ideas endlessly, but turning that mid text into anything I'd want to publish feels harder than just being thinking & writing about something myself from the jump

i would guess that LLMs more or less double the speed with which i can implement software systems.

i spend more time on spec upfront, which helps me create detailed prompts for the LLM. coding is way accelerated moving forward from there since the LLM now has a highly constrained solution space. 

TL;DR: LLMs are bad at design, but good at coding

working on /microsub; creator of @degentipbot.eth; producer/dj; host of /campfire 🎧 + /downshift ⚙️

thanks for the thoughtful answer!

do you think that LLMs make you *better* or just *faster*?

faster, mostly. the art is in the architecture (LLMs are *very* bad here)

better only in the sense of mundane things like not making typos and syntax errors, and getting to the bottom of bugs with less research and headache

You‘re using what? Copilot, local LLM, web Ui? 

To let the ai code with / for you?

Developer 🤓 build, develop and operate Farcaster solutions @ dtech.vision also love tech, decentralization, fitness and oldschool rap 😎

just ChatGPT in the browser, with a ~slow copy/paste workflow (works well enough for now, though)

very high on my list. once @microsub is live, i’ll be taking a pass

Have you tried claude? Saw folks really loving it and haven’t tried yet