In response to this, OpenAI has created a "foundations team" to explore alternatives such as synthetic training data. The Orion model, originally thought to be the code name for GPT-5, is now expected to arrive in 2025. However, concerns about the power infrastructure's ability to support the model's operation have been raised.
Key takeaways:
- OpenAI's next-generation Orion model of ChatGPT may not be as advanced as expected, showing only a small improvement over its predecessor, GPT-4.
- The AI industry is facing a bottleneck due to a dwindling supply of high-quality text and other data for training new models.
- OpenAI has created a 'foundations team' to find solutions to the lack of appropriate training data, potentially using synthetic training data.
- Orion, originally thought to be the code name for OpenAI's GPT-5, is expected to arrive in 2025, but there are concerns about whether there will be enough power to support it without overloading electrical grids.