OpenAI has been a proponent of the open source spirit of the Internet for a very long time, and the GPT AI model has been open source for a very long time, at best, after a period of launch, and then open source.
That changed with ChatGPT’s adoption of GPT-3.5, which, along with the latest GPT-4, was not open source.
Recently, in an interview with the media, Ilya Sutskever, the founder of OpenAI, answered the question, saying that it is irrational to open source artificial intelligence.
“These models are very strong, and they’re going to get stronger,” he said. “In a way, it’s easy to do a lot of harm with these models if someone wants to.”
At the same time, Sutskever says, “As the capabilities of AI continue to improve, there’s a reason not to want to reveal them.”
Sam Altman, CEO of OpenAI, has previously made a similar point.
Altman has said in interviews that AI tools can help people become more productive, healthier, and smarter, but added that the world may not be that far away from scary AI tools, so it’s “critical” to regulate them.
According to the official response, the risk and serious consequences of AI abuse may be one of the main reasons why OpenAI chose not to open source its models.