Anti-LLM Sentiment Considered Harmful

Indeed, there are several things you can do to reduce the harms. You can avoid paying for new development. Using models like Kimi 2.6 from e.g. openrouter.ai means you basically only pay for the inference compute and not any training or scraping, running models locally is even better as you don’t provide any economic incentives at all. Another thing you can pay attention to is the transparency of their training data. For example comma is trained on a corpus of permissively licenced text, but these kinds of models are currently very underpowered. Of course, you can always also discuss Haskell with real people here or over on the #haskell IRC or Haskell Matrix chat channels.