After OpenAI released the o3-mini model, its CEO Sam Altman, Chief Research Officer Mark Chen, Chief Product Officer Kevin Weil;
Vice President of Engineering Srinivas Narayanan, Head of API Research Michelle Pokrass, and Head of Research Hongyu Ren, conducted an online technical Q&A on reddit, one of the world’s largest comprehensive forums.
The main topics discussed were the current situation of OpenAI, future product planning, and the global phenomenon of the DeepSeek-R1 large model.
The best questions
A user asked: “Can we see all the thinking tokens of the large model?”
Sam Altman: ”Yes, we will soon be showing a more useful and detailed version. This is thanks to an update in R1.
Kevin Weil: We are working on showing more than we show today – this will happen soon. The exact way to show all the content is still to be determined, but showing all the thought chains could lead to competitive refinement.
But we also know that people (at least power users) want to see this, so we will find the right way to balance this.
Srinivas Narayanan: We are looking at ways to show more of the thought process. Stay tuned.
Will you consider releasing some model weights and publishing some research?
Sam Altman: Yes, we are discussing it. I think we chose the historically wrong side of closed source and need to figure out a different open source strategy. Not everyone shares this view, and it is not our top priority at the moment.
Let’s talk about this week’s hot topic, Deepseek. Obviously, this is a very impressive model, and I know it was probably trained on the output of other big models. How will this affect your future model plans?
Sam Altman: It’s a very good model! We’ll develop better models, but our lead will be smaller than in previous years.
Suppose it’s 2030 and you’ve just created a general artificial intelligence (AGI) system. It excels at every benchmark you set, outperforming your best engineers and researchers in terms of both speed and performance. What’s next? Any plans beyond “launch it on the website”?
Sam Altman: The most important impact, in my opinion, will be to accelerate scientific discovery, because I believe this is the factor that will contribute most to improving the quality of life.
Srinivas Narayanan: The interface through which we interact with AI will change fundamentally. AI will become more autonomous, continuously processing complex tasks and goals for us in the background.
They will only contact us when needed. Robotics should also have advanced enough to be able to perform useful tasks for us in the real world. (Judging from this answer, AI Agent is indeed one of the best use cases for AGI.)
Advanced voice dialogue capabilities have apparently declined since the video feature was released. Are there any plans to fix or improve this?
Srinivas Narayanan: Thanks for the feedback, will look into it.
Why is the knowledge deadline for the o3 – mini still October 2023? Is this also the case for the o3 or o3 Pro? Are there any plans to update the knowledge deadline?
Before I paid for a Plus subscription, the knowledge deadline for GPT – 4o was June 2024, but now that I have paid, the knowledge deadline has gone back to October 2023. Why is this the case?
Sam Altman: Now that we have enabled search, this (the knowledge expiration date) is less important. Personally, I will never think about the knowledge expiration date again
What other AI Agents can we expect?
Create an agent for free users too, to speed up the rollout.
Any news on a new version of DALL-E?
And finally, the question everyone asks… when will AGI be achieved?
Kevin Weil: More agents will be released, coming soon, and I think you’ll be happy with them. Image generation based on 4o is about a few months away, and I can’t wait to show it to you. It’s really great.
AGI will happen.
Will there be big improvements to GPT-4o? I really like custom GPT, and it would be great if it could be upgraded, or if we could choose what model to use on custom GPT (e.g. o3 mini).
Michelle Pokrass: Yes, we’re not done with the 4o series yet!
Will there be an update to the advanced voice mode? Will this be a focus of GPT-5o?
Sam Altman: Yes, there will be an update to the advanced voice mode!
We’ll call it GPT-5, not GPT-5o. There’s no timetable yet.
How important was the success of Stargate for the future of OpenAI?
Kevin Weil: Very important. Everything we’ve seen suggests that the more computing power we have, the better models we can build and the more valuable products we can create. We’re now scaling models in two dimensions at once – larger pretraining and more reinforcement learning (RL) / “strawberry” (probably some kind of technology or project code name) – and both require computing power.
So is delivering products to hundreds of millions of users! And it requires computing power as we move towards more intelligent products that can work for users on an ongoing basis. So think of Stargate as our factory that turns electricity or GPUs into great products.
