Post #727189
2026-03-20 02:48 UTC
Replies (3)
-
@mawhrin@circumstances.run 2026-03-20 08:13
@anttipeltola@mastodon.world there's a typo in the article. the actual slowdown is 20,000 times.
-
@kumarvibe@mastodon.social 2026-03-20 09:18
@anttipeltola@mastodon.world according to the linked article they only ran a benchmark *after* the LLM rewrote SQLite. Yeeeaaah, that’s not gonna work. Here’s what happens when you tell the LLM to run a benchmark continuously after each change and discard any patch that makes things slower: https://awesomeagents.ai/news/shopify-ceo-ai-agent-liquid-engine-53-faster/ Feedback loops matter. They’ve always mattered for hand written software but they matter more for AI because LLMs are good at reacting to corrective instructions.
-
@gregalotl@c.im 2026-03-22 09:12
@anttipeltola@mastodon.world OK. Here's a conundrum. I am and always have been a crappy coder but I work in software as a UX designer and I'm good at that. I have loads of ideas about useful projects that I can't code or would take ages, inevitably getting bogged down in the swamp of self righteousness that is Stack Overflow. Competent coders are busy building their own visions and won't help. AI is the solution for me. In the last few weeks, I (and I mean AI) have knocked three long-term, 'meaning to get it done for ages', projects off my list. Not to say AI is good all the time. But once the structure's there I've been able to steer, cajole & constrain, through careful prompt construction. It puts me in the game. At one level, AI is building prototypes (sorry Figma) and if I commercialized any of these, I'd go with meat based coders to refine or start again. But, I'd be confident that I'd specified the back end and frontend user experience I wanted