I’ve noticed Sam mentioning user feedback multiple times on Twitter, so I wanted to share my thoughts here.
GPT-4.5 has been the closest I’ve seen to a truly “human” model, and it’s been immensely helpful. While I mainly use LLMs for coding at work—where most models are interchangeable with certain quirks—4.5 + memory tackles a different axis. I’ve been using as a life planner and a kind of “mind debugger.”
Previously, I found Claude 3.5 to be the most human-like AI, but with 3.7, it seems to have shifted toward being an overly verbose coding tool. I’d love to see OpenAI step in and fill that gap.
I understand inference costs are substantial, and I hope the team can find ways to scale effectively. However, limiting GPT-4.5 to only 50 requests per week feels too restrictive. o3-mini and mini-high limits were upped, and I hope 4.5 can likewise be upped.