I'm sorry, but you still have to think(itsallaboutthebit.com)
435 points by drogus 2 days ago | 181 comments
tl;dr: DHH's AI-generated Rust, Elixir, and Go rewrites of Campfire produce misleading comparisons because the LLM made different implicit trade-offs (backwards compatibility, Redis vs in-process queues, sync vs async locks) across languages, and the benchmarks only measured throughput. Deeper testing reveals the Rust version drops 99% of notifications under load due to a small broadcast channel capacity, while Elixir's unbounded mailboxes trade memory and latency for delivery—neither is strictly better. The takeaway: language choice doesn't automatically solve concurrency problems, and you still need to understand code, failure modes, and trade-offs.
HN Discussion:
  • ↑AI-generated low-quality work is cheap to produce but expensive to debunk
  • ↑LLM translation worked here only because humans already did the hard thinking
  • ↑Developers should still read code and understand systems rather than abdicate to AI
  • ↓Worrying about understanding code echoes outdated concerns like distrusting compilers
  • ~Shipping half-understood code is fine for low-value features to save energy