Camden

I've really gotten into local LLMs lately. I'd love more articles on them. I use average models like gemma3:4b or minstral 7b, ones that fit in 8gb VRAM. I feel like they don't work too well with follow up questions sometimes and admittedly get simple stuff blatantly wrong, on a level where even free AI online does magnitudes better. I would love more articles about tweaking the system prompt and helpful tools for making slightly dumber local models work well.

A free standard experience followed by a $12/mo annually billed subscription is a no go. Why can't we have a $2/mo double usage subscription? I hate when companies do this. Also, modern phones are capable enough to run AI for voice transcription and light summarization without using the cloud. I would much rather see an app that taps into that. Samsung can easily transcribe and summarize on device, not sure how hard it is to target that as a developer.