OpenAI CEO Sam Altman has revealed that the expected next-generation AI model, GPT-5, is unlikely to be released this year. The company is prioritizing the enhancement of existing models that focus on reasoning and problem-solving due to increasing model complexity and limited computational resources.
OpenAI is working to ensure sufficient computational infrastructure for its generative AI models. Reports indicate that OpenAI has been collaborating with Broadcom to develop an AI chip aimed at powering its models, potentially available by 2026. However, many within OpenAI believe GPT-4o is not ready for public release.
Short-term constraints in computational capacity mean that ChatGPT's advanced voice mode will not integrate the visual features demonstrated earlier. This demonstration was reportedly intended to offset attention from Google's I/O Developer Conference.
Additionally, the release of an audio-only version of the advanced voice mode has been delayed by several months. Altman also noted that there is no set timeline for the next major version of OpenAI's image generator, DALL-E.
In the realm of video generation, OpenAI's tool Sora is facing challenges related to model refinement, safety, and computational scalability, lagging behind competitors. The original system featured in February took over 10 minutes to process a one-minute video clip. Notably, Sora's co-lead Tim Brooks departed OpenAI for Google in October.
Despite these hurdles, Altman remains optimistic, emphasizing the improvement and future iterations of the o1 series of "reasoning" models. OpenAI's DevDay event in London previewed several upcoming features of the o1 series, including image understanding. Altman assures that there will be notable releases later this year, although not under the name GPT-5.
