The CEO of OpenAI, Sam Altman, points out that the shortage of computing capacity is delaying the company's products.
The CEO of OpenAI, Sam Altman, acknowledged that the lack of processing capacity is one of the reasons preventing the company from launching products as frequently as it would like.
Sam Altman, CEO of OpenAI, acknowledged in a Q&A session on Reddit that the scarcity of processing capacity is a significant hurdle preventing the company from launching products more frequently. In response to a question about why OpenAI's new artificial intelligence models have taken so long to develop, Altman explained that “all of these models have become quite complex” and mentioned the limitations and difficult decisions related to allocating computing resources to various promising ideas.
Several reports indicate that OpenAI has faced challenges in securing sufficient computing infrastructure to train and run its generative models. Recently, a media outlet reported that OpenAI has been collaborating with Broadcom to develop an AI chip intended to run models, with expectations that this chip will be available by 2026 at the latest.
Due to this limited capacity, Altman announced that the new advanced conversational feature of ChatGPT, known as Advanced Voice Mode, will not incorporate the visual capabilities that had been anticipated in April. At an event held that month, OpenAI showcased a version of the ChatGPT app running on a smartphone and responding to visual cues, such as the clothing a person was wearing in the camera’s field of view. However, it was revealed that this demonstration was rushed to distract from Google's I/O developer event happening the same week.
Within OpenAI, some members felt that the GPT-4o model was not ready for release, which was evident in the delay of the voice-only release of Advanced Voice Mode. During the same session, Altman mentioned that there is no updated timeline for the release of DALL-E, their image generator, and that they are still refining Sora, their video generation tool. Kevin Weil, product director at OpenAI, indicated that Sora has faced technical setbacks that have made it challenging to compete with rival systems like those from Luma and Runway.
It was reported that the original version of Sora, presented in February, took over 10 minutes to process a one-minute video clip. In October, one of Sora's co-leaders, Tim Brooks, left the company to join Google. Altman also commented that OpenAI is still considering the possibility of allowing “NSFW” content in ChatGPT “someday” and emphasized that the company's main priority is to improve its series of reasoning models o1 and their successors. This week, OpenAI previewed several features that will come to o1, including image understanding, announcing that there are “very good releases” expected by the end of this year, while clarifying that none of this will be called GPT-5.
