OpenAI o3, every mention

mentionsepisodesper episode tap a year for its mentions

Verbatim, from the transcripts: the passages where OpenAI o3 comes up

loading…

▶ 24:19 Gavin Baker Somebody told me it only took a 160,000 reasoning traces from O-one and O-three to get the original DeepSeek.

▶ 2:11:09 Rahul Sonwalkar Well, there's this whole thing about how people would, many people were still using GPT-IV-O because they thought it's better than O-III. 2 times in the scene

▶ 17:31 unnamed speaker But even then, you know, once they get O-three in here, it's really like everything's covered, so.

▶ 10:11 Daniel Singer I heard there was this model that came out yesterday that people were doing some cool stuff with, uh, if you have not tried O three, it's amazing, but

▶ 8:34 Will Brown Um, like my, I tweeted something about this, like my, I'm happy to call O three, 10 minute AGI. 4 times in the scene

▶ 7:35 Aravind Srinivas In fact, most people using tracks between the world don't even know there's a model called O one or O three and don't even know what the difference is for GPT four.
▶ 9:51 Aravind Srinivas I has O three and O four.

▶ 4:03 Mike Knoop Uh, and then at the end of the year, obviously last fall, we started seeing things like O-one and O-three. 5 times in the scene

▶ 23:44 John Coogan Two slides later says, uh, they omitted O three from the chart in the live stream for some reason, so I added the numbers for you, and then, uh, but this was reasonable because O three isn't public, and those benchmarks can't be verified… 2 times in the scene

▶ 14:57 unnamed speaker Yeah, and so, and this doesn't even include O-three, the latest model from OpenAI that hasn't been fully released yet. 3 times in the scene

▶ 19:38 John Coogan Um, and so, uh, there was, we talked about this in the show previously, but oh three, which isn't out yet, but is even more advanced in terms of reasoning.
▶ 22:27 John Coogan I didn't, I didn't test a bunch of stuff, but I was like, but even that, I think that something, something odd's going on where I think a lot of people haven't used ChatGPT O-one and they certainly haven't tried O-three because it's not… 2 times in the scene
▶ 54:48 John Coogan We know O one O three and R one all seem to be pretty good on top of the base model.
▶ 1:06:04 John Coogan And with that, and that with larger clusters and continued scaling of synthetic data and inference compute, next generation models are going to be like a hundred X better than O three.

▶ 33:35 John Coogan Uh, ChatGPT has the better reasoning models with O-one and soon O-three. 2 times in the scene

▶ 28:46 John Coogan We're seeing that with like the O one O three models. 2 times in the scene

▶ 1:03:02 John Coogan Let's go to O three, Mike Noop who runs arc, 6 times in the scene
▶ 1:09:14 Jordi Hays And this was posted, I guess this was posted before O-three launched, so maybe double check to make sure they're still hiring that person.

Turn any episode into a week of clips.