OpenAI starts training 'GPT-5'

OpenAI has started training the model that will likely be released as GPT-5. The model's training could take months. OpenAI wants to work towards 'general artificial intelligence', a system that is smarter than humans.

The successor to GPT-4 should, among other things, form the basis for chatbots, picture generators and digital assistants, writes The New York Times . OpenAI is also introducing a new safety and security committee. The company recently closed down the previous one. The new committee must ensure that AI models are safe to use by the public and do not pose a danger to society.

Now that training has started, it remains to be seen when the release will take place. Training usually takes at least several months, after which companies often take months to test and tailor the model to the requirements. The second step, reinforced learning from human feedback, is important to ensure that an AI model does not go off course and, for example, suggest the use of force or answers questions in a way that could cause problems for OpenAI. OpenAI does not say when the successor to GPT-4 will be released. The most recent model is GPT-4o.