In a recent Reddit AMA, OpenAI’s CEO, Sam Altman, candidly revealed that limitations in computing capacity are a major hurdle preventing the company from releasing new products more frequently. Addressing a question about the lengthy development process for OpenAI’s forthcoming AI models, Altman explained, “All of these models have gotten quite complex. We also face a lot of limitations and hard decisions about how we allocate our compute towards many great ideas.”
Reports indicate that OpenAI has been encountering difficulties in securing the necessary computing infrastructure to run and train its advanced generative models. Notably, it has been working with Broadcom to develop a specialized AI chip expected to be ready by 2026.
Due to these capacity constraints, features like the much-anticipated Advanced Voice Mode for ChatGPT won’t be incorporating the vision capabilities that were initially hinted at back in April anytime soon. During a demonstration in April, OpenAI showcased the new ChatGPT app functioning on a smartphone, adeptly responding to visual inputs such as clothing seen through the phone’s camera. However, this demo was reportedly expedited to gain attention away from a major event held by Google, and many within OpenAI questioned if the tech was ready.
Altman also stated that OpenAI’s next iteration of its image generator, DALL-E, does not yet have a set release date. Similarly, Sora, the company’s video creation tool, is still under development as they aim to enhance the model’s performance, safety, and scalability. Sora has faced competitive challenges and technical issues, taking upwards of ten minutes to generate just a one-minute video clip.
Adding to the company’s transitions, one of Sora’s project co-leads, Tim Brooks, departed for Google.
The AMA touched on other topics as well. Altman mentioned that OpenAI might eventually allow “NSFW” content in ChatGPT, believing in the maturity of adult users. Moreover, OpenAI is focusing on enhancing its o1 series of “reasoning” models. New features, including image understanding, were previewed at the DevDay conference in London.
While Altman teased promising new releases expected later this year, he clarified, “Nothing that we are going to call GPT-5, though.”
