> Models have gotten noticeably smarter and see more use across a variety of fields and contexts.

Is that really true? The papers I've read seem to indicate the hallucination rate is getting higher.

Models from a few years ago are comparatively dumb. Basically useless when it comes to performing tasks you'd give to o3 or Gemini 2.5 Pro. Even smaller reasoning models can do things that would've been impossible in 2023.