OpenAI's progress hampered by lack of compute power: Sam Altman
OpenAI CEO Sam Altman has admitted that the company's product development is being hindered by a shortage of compute capacity. The admission came during a Reddit AMA session, where Altman detailed the complexity of their models and the challenges they face in allocating resources. "All of these models have gotten quite complex," he said, adding, "We also face a lot of limitations and hard decisions about [how] we allocated our compute toward many great ideas."
OpenAI collaborates with Broadcom on AI chip
OpenAI is said to be working with Broadcom to create an AI chip optimized for running models, as reported by Reuters reported this earlier this week. The new chip could be launched as early as 2026. The strategic partnership would help OpenAI tackle its persistent challenge of obtaining enough compute infrastructure for its generative models.
OpenAI's advanced voice mode for ChatGPT faces delays
The limited compute capacity has also affected the development of OpenAI's Advanced Voice Mode for ChatGPT. Altman confirmed that the feature, which was first introduced in April with vision capabilities, won't be released anytime soon. This delay comes after internal concerns about the readiness of GPT-4o and the subsequent postponement of the voice-only version of Advanced Voice Mode by several months.
Uncertainty surrounds future releases of OpenAI's image generator
The future release of DALL-E, OpenAI's image generator, also remains uncertain amid these challenges. "We don't have a release plan yet," Altman said during the AMA session. Meanwhile, the launch of Sora, OpenAI's video-generating tool, has also been delayed due to the need for model refinement and scaling compute resources, according to Kevin Weil, OpenAI's Chief Product Officer.
OpenAI's Sora faces competition and technical challenges
Sora has been plagued by technical issues, making it lag behind competitors like Luma and Runway. The Information reported that the first system, which was unveiled in February, took more than 10 minutes of processing time to generate a one-minute video clip. The situation worsened with the exit of Tim Brooks, one of the co-leads on Sora, who joined Google in October.
OpenAI considers NSFW content, prioritizes reasoning models
Altman also revealed that OpenAI is considering the inclusion of "NSFW" content in ChatGPT in the future. He revealed that the company's top priority is enhancing its o1 series of "reasoning" models and their successors. At its recent DevDay conference in London, OpenAI previewed several features coming to o1 later this year including image understanding capabilities.
