We Gave GPT-4.5 a Myers-Briggs Test. It’s an Extrovert.

Hello, and happy Sunday!  I’ve been writing for Context Window for a a couple of months, and I'm happy to report I am now at the helm. (I also help lead Every’s consulting business.) I’m obsessed with how AI tools are developing and, more importantly, what they can do for you. Every week you'll get my analysis of the latest news and trends in AI, along with everything we published last week and updates from Every Studio.—Alex Duffy

Was this newsletter forwarded to you? Sign up to get it in your inbox.

---

Release notes

Allow us to re-introduce GPT-4.5

When GPT-4.5 arrived last week, expectations were stratospheric. For two years, people had been imagining a leap forward from GPT-4—an AI so brilliant it would feel superhuman in all regards. It was an impossible standard, and unsurprisingly, GPT-4.5 didn’t meet it. As we wrote last week, GPT-4.5 might not totally blow your mind, but might befriend you instead.

Having spent more time with the model, we can confidently say it’s lived up to that promise: We at Every are officially friends with GPT-4.5.

Beyond benchmarks: Subtlety and nuance

Initially, GPT-4.5’s benchmark scores didn't amaze us, aside from its standout improvement on SimpleQA, a benchmark meant to test factual accuracy—on which it nearly doubled GPT-4’s performance and dramatically reduced hallucinations.