OpenAI has not yet started training GPT-5, the successor to its GPT-4 language model, months after the company pledged to not work on the model “for some time” due to concerns about the potential dangers of large language models.
In an interview at a conference hosted by Indian newspaper Economic Times, OpenAI CEO Sam Altman said that the company is still working on new ideas for GPT-5, but that it is not yet ready to start training the model.
“We have a lot of work to do before we start that model,” Altman said. “We’re working on the new ideas that we think we need for it, but we are certainly not close to it to start.”
Altman’s comments come amid growing concerns about the potential dangers of large language models. In recent months, there have been a number of high-profile incidents involving large language models generating harmful or offensive text.
In one incident, a large language model was used to generate realistic-looking fake news articles that were designed to spread misinformation. In another incident, a large language model was used to generate hate speech.
These incidents have raised concerns that large language models could be used to create harmful content that could have a negative impact on society.
OpenAI has taken steps to address these concerns. The company has developed a set of safety guidelines for large language models, and it has also limited the access that researchers have to its language models.
However, Altman acknowledged that there is still more work to be done to ensure that large language models are used safely.
“We’re still in the early days of this,” Altman said. “We’re still learning how to build these models safely, and we’re still learning how to use them safely.”