GPT-5 hot take

It’s here, finally, but not everything people dreamt it would be

OpenAI just announced GPT-5. I stand by my predictions from a couple weeks ago; none of the problems I said would not be solved appear to have been solved..

Here’s my hot take:

• Took almost 3 years, many billions of dollars (over a half-trillion fieldwide).

• Good progress on many fronts.

• But still part of the pack, not a giant leap forward (e.g. Grok 4 beats it on ARC-AGI-2 results)

• OpenAI conveniently forgot to include this comparison (ARC-AGI-2) in their livestream recital of benchmark progress, which left the livestream looking like marketing rather than science.

• Fan will still find something to rejoince in, but GPT-5 is not the huge leap forward people long expected.

• Lots of questions TBD about real-world performance.

• GPT-5 is obviously not AGI.

Also:

• Pricing is good, but profits may continue to be elusive; still no clear technical moat.

• What was up with multiple graphs (at least two) in which the bars didn’t match the numbers? Did GPT-5 make them??

Never have 69.1 and 30.8 looked so different. And 69.1 is now less than 52.8?

Quote of the day: “I dont wanna read anything about exponential progress anymore”,

—X user @flowersslop

and for good measure, a deceptive (or just confused?) graph about deception:

“AGI 2027” seems more and more remote by the day.

Gary Marcus thinks that perhaps scaling is not in fact all you need.