In a recent Reddit AMA (Ask Me Anything), Sam Altman, CEO of OpenAI, admitted that insufficient computational power is one of the main reasons the company struggles to release new products frequently. He noted that AI models are becoming increasingly complex, and OpenAI faces numerous constraints and tough decisions in allocating computational resources to realize many great ideas.
According to various reports, OpenAI has encountered difficulties in securing adequate computational infrastructure. This week, Reuters cited sources claiming that OpenAI is collaborating with Broadcom to develop AI chips for running its models, with an expected release before 2026.
Due to the scarcity of computational resources, Altman revealed that the visual capabilities originally planned for ChatGPT's Advanced Voice Mode will not be introduced in the short term. At a launch event in April, OpenAI demonstrated its ChatGPT application running on smartphones and identifying visual information such as clothing worn by people within the camera's field of view. However, reports suggest that this demonstration was aimed at stealing the spotlight from Google's I/O Developer Conference, and many OpenAI insiders believe GPT-4o was not ready for public release.
During the AMA, Altman mentioned that the next major update for OpenAI's image generator DALL-E does not yet have a release date. Meanwhile, OpenAI's video generation tool Sora has been delayed due to the need to refine the model, ensure safety, and expand computational capabilities. Sora has also faced technical setbacks, putting it at a disadvantage compared to competitors like Luma and Runway.
Additionally, in October, Tim Brooks, co-head of Sora, left for Google. Altman mentioned in the AMA that OpenAI is still considering whether to allow "adult content" in ChatGPT and stated that the company's current priority is to enhance the performance of the o1 series "reasoning" models and their subsequent versions. At this week's DevDay conference, OpenAI showcased some upcoming new features, including image understanding.
Altman concluded: "We will release some very nice new versions later this year, but we won't call it GPT-5."
Key Points:
🌐 OpenAI CEO cites insufficient computational power as the main reason for delayed product releases.
💻 The company is collaborating with Broadcom to develop AI chips, expected to launch before 2026.
📅 Updates for products like DALL-E and Sora do not have a definite release schedule.
