Post #1757869
2026-04-25 17:05 UTC
@aredridel I’m a little confused; you are making this sound like it should be a pro-LLM argument but it appears to be not-pro-LLM?
Most claims I see in my workplace seem to indicate productivity of late 2025 / early 2026 models, although it seems like that has plateaued. I feel like this study would largely be dismissed due to time period and model versions.
Replies (1)
-
@aredridel@kolektiva.social 2026-04-25 21:26
@Fishrock It's a _bad_ anti-LLM argument; early-2025-era coding tools, on weirdly selected tasks, in ways that open source maintainers don’t really work, go slower than doing it by hand.