I'm pretty skeptical of all the "AI companion" stuff but i've had maybe two moments where a model said something that landed better than i expected. and a lot more where it was obviously just doing sympathy-by-pattern and the whole thing fell apart the second i noticed. what i can't figure out is where exactly it breaks. for me it's usually the fake enthusiasm, or when it asks a follow up question at the end of literally every message like it's interviewing me. or it rushes to fix so...