OpenAI AMA: AGI Fast Takeoff and DeepSeek Competition

Recently, shortly after the launch of DeepSeek, the OpenAI team—including CEO Sam Altman, Chief Research Officer Mark Chen, and Chief Product Officer Kevin Weil—participated in a Reddit AMA (Ask Me Anything) to engage with the community's questions.

One of the most notable takeaways was Sam Altman’s admission that he regretted not adopting an open-source approach, which appeared to be a response to DeepSeek’s market impact. He also shifted his stance from previously accusing DeepSeek of infringing on OpenAI’s intellectual property to instead praising it, acknowledging that the performance gap between their models might narrow. Other responses hinted at DeepSeek’s influence, such as OpenAI accelerating the release of its o3 model and planning to lower the cost of its Plus plan.

From the extensive Q&A, we’ve distilled five key takeaways, including how model iterations are progressing toward AGI. Notably, while the event was framed as an "Ask Me Anything," the participants naturally chose which questions to address. For instance, Sam Altman did not respond to a pointed question from a user who asked, "Why are you working with the US government Nuclear team to work with AI? Did you guys NOT see Terminator yet?"

Six core members participated, including Sam Altman, Mark Chen, Kevin Weil, VP of Engineering Srinivas Narayanan, API Research Director Michelle Pokrass, and Research Director Hongyu Ren.

1. New Voice and Image Generation Models "Coming Soon"

Since the launch of advanced voice features last September, many have been impressed by ChatGPT’s conversational abilities, likening the experience to the movie Her. The current voice feature is powered by Whisper, an automatic speech recognition model. One user asked whether Whisper is being updated and if new voice models are on the horizon.

Sam Altman confirmed that new voice models are in development, while API Research Director Michelle Pokrass noted that they had open-sourced Whisper v3-turbo at a developer conference. However, specific details about upcoming updates remain undisclosed.

As for image generation, it's been a year since the release of DALL·E 3, and its iteration speed has lagged behind competitors. Chief Product Officer Kevin Weil stated that they are working on a new model leveraging GPT-4o’s image generation capabilities. However, he did not provide a clear timeline for its release, indicating that users will need to wait a bit longer.

2. Full Version of o3 Expected "Within a Few Months"

OpenAI introduced its latest flagship model, o3, last December, followed by the release of o3-mini on February 3rd, likely in response to DeepSeek. During the AMA, Sam Altman stated that the full version of o3 would be available sometime between “more than a few weeks” and “less than a few months.”

The currently available o3-mini is optimized for speed and ease of use, excelling in math problem-solving, code generation, and answering scientific queries. Meanwhile, o3-mini-high, available to paying users, offers even stronger reasoning and coding capabilities.

3. Plans to Gradually Lower Plus Plan Prices

OpenAI has long envisioned making AI accessible to everyone to enhance human life. However, with limited free-tier usage and a $20 monthly fee for the most affordable paid plan, that goal remains a work in progress.

When asked whether OpenAI would raise the Plus plan price, Sam Altman instead indicated that they intend to lower it gradually over time. It appears that competition from DeepSeek is prompting OpenAI to refocus on its original mission.

4. Where Are OpenAI's Robots and AGI Headed?

On February 5th, Figure announced the end of its collaboration with OpenAI, citing breakthroughs in AI development. This comes as no surprise, given that OpenAI is actively working on its own robotics initiatives. Sam Altman and Kevin Weil emphasized that their primary focus in robotics remains on "learning" and striving to create a truly capable robot.

Regarding AGI, a user posed a hypothetical scenario: "Let's say it's 2030 and you've just created a system most would call AGI. It aces every benchmark you throw at it, and it beats your best engineers and researchers in both speed and performance. What now? Is there a plan beyond 'offer it on the website'?" Altman responded that the primary goal of AGI remains accelerating scientific discovery.

VP of Engineering Srinivas Narayanan shared his vision for AI’s future, stating, "The interface through which we interact with AI will change pretty fundamentally. Things will be more agentic. AI will continuously work on our behalf, on complex tasks, and on our goals in the background. They will check in with us whenever it is useful. Robotics should also advance enough for them to do useful tasks in the real world for us."

In 2021, OpenAI disbanded its robotics team. It's now making a return to the field. (Source: OpenAI)

Additionally, in another response about Recursive Self-Improvement, Sam Altman stated, "i personally think a fast takeoff is more plausible than i thought a couple of years ago. probably time to write something about this..." This suggests that OpenAI's internal research on AGI may indicate a trajectory toward exponential improvement—where AI could rapidly enhance itself, surpass human intelligence, and potentially achieve superintelligence.

5. Facing DeepSeek: The Extent of Open-Sourcing Remains Uncertain

Although Sam Altman expressed regret over OpenAI’s lack of open-sourcing, he also acknowledged internal disagreements on the issue. While OpenAI has open-sourced models like Whisper v3-turbo, when asked whether they would consider open-sourcing older models that are no longer cutting-edge—such as GPT-2 and GPT-3—Kevin Weil indicated that they were open to the idea but did not make any commitments.

Another user, seemingly influenced by DeepSeek’s ability to display its thought process, asked whether OpenAI would provide full visibility into all thinking tokens. Altman responded, "yeah we are gonna show a much more helpful and detailed version of this, soon. credit to r1 for updating us."

In this AMA, Sam Altman faced DeepSeek with a mix of courtesy and strategy.

However, Weil also pointed out that revealing an AI model’s entire thought process could make it easier for competitors to replicate. As a result, OpenAI is trying to strike a balance between maintaining a competitive edge and enhancing the user experience.