Tim Disney
GPT-5 Thoughts
I watched the OpenAI livestream introducing GPT-5 today and had some thoughts.
First, the presentations were all oddly stilted and awkward. Most of the presenters seemed to be mimicking Sam Altman's affect. It was weird, like I'm not expecting them to be as polished as an Apple keynote (even then Apple has a spectrum of polish) but like maybe they could have done a few more run-throughs?
Second, they had blindingly obvious errors in some of their graphs. They had a bar chart where 52 was taller than 69, which was the same height as 30. Clear case of automation bias since I'm sure they had GPT-5 check for errors and it didn't catch it so the humans didn't either.
Third, one of the demos they had was of GPT-5 explaining the Bernoulli principle in relation to how flight works. It made a web visualization showing how lift is created based on the principle and how air travels over a wing. The wild thing is that the Wikipedia page on the Bernoulli principle calls out how this is a common misconception. Now, maybe the details of the demo were actually correct but it was hard to tell from just watching (Hacker News had a good time nerdsnipping themselves over it). Strange that they went with this example because it's so fraught with misconceptions and easy to get (subtly) wrong.
Also, not a huge point, but they kept saying how GPT-5 was like having a PhD in your pocket. As a PhD haver who has worked with plenty of other PhD havers, this is just so funny. Like I get what they are trying to communicate (it's so smart!) but it's not nearly as impressive as it sounds (in the same way that Bruce Banner having 7 PhDs is supposed to read as smart but is completely absurd). People really need to understand that a PhD does not make you smart.