OpenAI o1 and o1 pro mode in ChatGPT — 12 Days of OpenAI: Day 1

Sam Altman and some members of the OpenAI team introduce & demo o1 and o1 pro mode in ChatGPT and discuss the ChatGPT Pro plan.

(from left to right): Sam Altman, Hyung Won Chung, Max Schwarzer, Jason Wei

[hermannschmidt]

Four happy nerds at a kitchen table changing the world. I do like this format.

[iraklimatcharashvili]

Jason: *tells a dad joke*
Sam: haha, terrible, you're fired

[MannyBernabe]

**Recap**
→ o1 full coming to ChatGPT Plus
→ New tier: ChatGPT Pro: $200/mo, o1-Pro, unlimited access to models and voice mode
→ o1 full getting multi-modal (images, etc.)
→ o1 full api coming soon

Highlight: Big jump in coding for o1 full. Pumped for this.

[hyungwonchung]

It was super fun livestreaming with Sam, Max and Jason this morning!

I was finally, finally able to use my thermodynamics and heat transfer from my PhD with this space datacenter example!

In the demo, I did a "lower bound estimation" of the cooling panel area. I'd like to provide some additional explanation that I skipped during the demo.

Many real world thermal analysis is highly complicated with engineering challenges. The space datacenter is of course one of them. One nice approach for such problems is to do a lower bound estimation as opposed to an accurate estimation. This can significantly simplify the problem while still providing valuable insight.

For example, if the lower bound estimate of the required area is larger than what we can afford, this system doesn’t make economic sense because the actual required area will be even larger. And we can get this kind of valuable insight without having to do an in-depth analysis.

In this space datacenter example, the model did a few simplifying assumptions that are consistent in making the resulting area estimate smaller because :
- ignored solar radiation onto the cooling panel
- ignored radiation from the deep space

If the model did not ignore this, the cooling panel will be hotter and hence more area is needed to cool it. These simplifications made the problem so much simpler that o1 can do the analysis in 10 seconds.

Also as Sam said, cooling panel of size 2% the San Francisco area might not be that big! Space datacenter of 1GW will be really cool

[HenryBloggit]

I liked that this presentation was approached like a low-key podcast instead of a big bombastic display like Google, Apple and Meta always do. I’m excited about the full o1 too! I almost wish I weren’t going on vacation tomorrow because now I actually want to work and get some programming done. Thanks for your team’s hard work and for making my job so much easier and more interesting!

[anearthian]

Imagine they reveal on 12th day that those 3 people are not people.👽

[XerosXIII]

kudos that they have actual people introduce in-person rather than the huge powerpoint screen

[dell_rew]

Kudos for the relaxed and non-binding atmosphere. I like this format. Great job!

[AlexCarter]

Love this idea of the 12 days. A strong first day - I like how he says the space question is "simple."

[riyadalnwili]

haha the strawberry sticker is a cheeky wink towards “how many R’s are in strawberry” issue

[FranXiT]

Honestly, the pro subscription is obviously not for the common man. It's for people who make hundreds of complex questions every day. People shouldnt sweat it, pro users are already getting the full o1 release, so why complain?

[David-snjw]

Great to see actual people demonstrating the model, comes across as authentic and not corporate speak. Whilst I’m not the target user for the ChatGPT Pro plan, the video and the team working on it gives me a sense of excitement of what’s to come. Good work Open AI!

[airlesstermite]

the laid back style of this is great. love it

[BiP]

“We have a pro tier”
Me: gets exited
“$200 a month”
Me: heart shatters

[judgeka]

Would you pay a phd level physicist $200 a month to work for you? What about a strategy consultant? Serious businesses wanting a competitive advantage will happily pay for it. Then there will be all the other businesses.

[theoz]

Here's what was released:
→ o1 model fully released and now available for ChatGPT Plus users (used to be o1 preview)
→ o1-Pro announced which utilizes more computational power to tackle complex problems (available on new ChatGPT Pro plan)
→ New tier: ChatGPT Pro – $200/month with o1-Pro, unlimited access, and advanced voice features
→ o1 upgraded to multimodal (meaning it can read both text + images)
→ o1 API coming soon (useful for developers)

So huge improvements in speed and quality particularly for coding and problem-solving with o1.

o1 is useful not only for math, coding, and science but also for those tougher more complex questions where GPT 4o falls short - so if you are not liking your responses with o1 you can try that out.

For the majority of users, the current $20/mth should suffice, and I doubt you will need the $200/mth subscription at this stage.

[travelman]

Thank you for everything you guys do. ChatGPT has improved my productivity 500%.

[chronogawd]

Sora is getting really good, you can almost believe this OpenAI video is a real video

[scionax]

If pro is really smashing through the limits like it shows here, it's basically ready for real work now. That'll be an absolute game changer, and I'm so here for it.

[GündemArama]

00:25 - OpenAI presents daily innovations during the 12 Days of OpenAI event.
01:56 - Introducing ChatGPT Pro with new features and improved model access.
03:38 - New OpenAI models increase functionality for advanced users and make responses more fluid.
05:19 - OpenAI's o1 model dramatically improves response speed and accuracy.
06:50 - Discussion on multimodal inputs and image understanding in AI.
08:55 - Demonstration of multimodal reasoning in ChatGPT's O1 model.
10:59 - Demonstration of ChatGPT's Pro mode for complex problem solving.
12:54 - OpenAI's o1 model improves capabilities for complex tasks.