Anti-LLM Sentiment Considered Harmful

Perhaps good to mention here, is that local models are slowly getting more powerful as well. The models that I can run on my laptop are not as powerful as the biggest downloadable models, and those still are not as powerful as OpenAI/Claude/Grok/etc., but their both quality and efficiency has been rapidly improving over time. They might eventually provide a free and genuinely good alternative for the big companies.

Right now they don’t actually do agentic stuff (in my experience), but they do okay with reading code, and suggesting improvements. See the screenshot below, which was generated by running gemma4(Google’s free model) on ollama (local LLM runner tool), entirely on my own laptop. The question btw was about GHC’s .gitlab/rel_eng/mk-ghcup-metadata/mk_ghcup_metadata. py file