OpenAI recently launched its new model, the o3-mini. To celebrate this event, several company executives, including CEO Sam Altman, answered questions from Reddit users in an AMA session. One statement made during the session was particularly noteworthy.
Open source. In response to a question from one of the participants, Altman admitted that OpenAI has lost some leadership in the AI field following the successful emergence of DeepSeek. He also acknowledged that OpenAI might have been “on the wrong side of history” when it comes to open sourcing its products.
OpenAI isn’t that “open.” OpenAI initially embraced its name by offering its first models, GPT-1 and GPT-2, with notable transparency and inviting community collaboration. However, this approach changed with GPT-3 and subsequent models. With these later versions, OpenAI provided minimal technical details to prevent competitors from gaining an advantage.
The possibility of OpenAI releasing an open source AI is still uncertain. Altman also mentioned that OpenAI is discussing potentially publishing the weights used to train its models, which would be valuable information for the community. However, he emphasized that “not everyone at OpenAI shares this view” and noted that any potential change in strategy in this regard is “not our current highest priority.”
Older models may become open source. Kevin Weil, OpenAI’s chief product officer, indicated that the company is considering releasing older models that are no longer state-of-the-art as open source. “We’ll definitely think about doing more of this,” he said, although he didn’t provide further details or estimated timelines.
DeepSeek has changed the landscape. Altman also emphasized that OpenAI plans to develop better models incrementally. However, he admitted that their leadership in the AI field will no longer be as clear-cut as it once was, largely due to the emergence of DeepSeek.
The o3-mini model will show how it “reasons.” The introduction of DeepSeek R1 has led OpenAI to decide that o3-mini should reveal its “thought process.” The company previously kept this information concealed to prevent competitors from gaining insights to compete effectively.
No GPT-5 on sight and limited DALL-E 4 updates. During the AMA session, participants asked the OpenAI team about the highly anticipated GPT-5 model. Altman said the company hasn’t set a release date yet. Meanwhile, Weil mentioned that the team is currently working on the successor to DALL-E 3, an imaging model launched in late 2023 that has since been surpassed by several alternatives. “I think it’s going to be worth the wait,” Weil said.
Image | TechCrunch
View 0 comments