1·
23 hours agoI noticed that there are sometimes huge differences between models and providers. ChatGPT, which most people use, is a useless sycophant because it’s made to value user satisfaction higher than factual correctness (so it makes shit up und basically never pushes back if you challenge it). When using Mistral AI I had the AI tell me outright that I’m wrong (though it still generated very friendly text to do so, and it only properly does it when in “thinking” mode). No clue how Anthropic or Google do in this regard, just that ChatGPT is especially awful. In the end all models are risky.
Not for complex tasks, really. Gemma4b or similar might be just enough as the backend for a voice assistant replacing Alexa (which I honestly hope, I need to get family members off of that fucking spy devices…). Small web searches might just fit. Or image recognition for Nextcloud Memories, stuff like that.