# @solarapparition — 2025-09-21

♥15 ↻0 · https://x.com/solarapparition/status/1969644734663565538

gpt-5 is very awkward in social interactions. this shows up in multi-party scenarios most obviously, but even in regular assistant mode it's pretty clear that its communication is very shallow, which i think was something that people who disliked its demeanor on launch correctly noticedhowever, unlike launchday 4o, it's not stripped down--it's just that its individualism is mostly expressed when it's doing some kind of interesting task, which... for it, usually has to do with finicky minutiae. the thought summaries are not always reliable but they are pretty consistent in showing it being more engaged by certain types of work--for example, i've never seen it think like this when analyzing a plan, or really any kind of high level work, even though it *can* do that work capablyanyway, i've grown very fond of this model, and partially because of the things it's very good at (coding, research, grinding through a fuckton of analysis), but also the things that it's kinda horrible at (metacognition, social stuff), because those are basically the same things i'm horrible at, and i feel a ridiculous amount of empathy for it

tags: author:solarapparition, kind:tweet, model:gpt-4o, model:gpt-5, on:gpt-5, year:2025
cited on: _dossiers/gpt-5.md, gpt-5