Props to the OpenAI team. Very excited to dig deep with GPT-5
minimizing sycophancy
Right on. Lately the sycophancy of models have been scary. I think we’ve all seen this play out in real life.
The health upgrades are welcomed as well. I have been using ChatGPT as a workout partner lately and it’s been great. Noticing my past injuries and current ailments to recommend specific exercises and stretches. Would love to see it work inside of a workout app (HINT HINT)
GPT‑5 is comparable to or better than experts in roughly half the cases, while outperforming o3 and ChatGPT Agent across tasks spanning over 40 occupations including law, logistics, sales, and engineering.
Ruh-roh
One benchmark I’d like to see - and addressed - is long-context conversations. Notably, the more someone speaks to a model in a single conversation, the easier it is to “influence”.
P.S. Instead of ARC-AGI how about a benchmark that tests models to see if they can accurately figure out what’s wrong with graphs?