This post is also available in:
עברית (Hebrew)
Leading AI technology giant OpenAI revealed it has been setting up a safety and security committee and begun training a new AI model to supplant its GPT-4 system for the ChatGPT chatbot. The company said in a blog post that the said committee will be tasked with advising the full board on “critical safety and security decisions” for its projects and operations.
The news about the new safety committee comes amid the swirling debates regarding AI safety in the industry in general and at the company in particular, after it was thrust into the spotlight by researcher Jan Leike who resigned and shared his criticism at OpenAI for letting safety “take a backseat to shiny products.”
Leike also announced just last week that he is joining the rival AI company Anthropic, which was founded by ex-OpenAI leaders, to “continue the superalignment mission” there (which was focused on managing AI risks).
OpenAI, on the other hand, said it has “recently begun training its next frontier model” and that its AI models lead the industry on capability and safety, though it made no mention of the recent controversy. “We welcome a robust debate at this important moment,” the company said.
The people on the safety committee include OpenAI CEO Sam Altman and Chairman Bret Taylor, four OpenAI technical and policy experts, Quora CEO Adam D’Angelo and former Sony general counsel Nicole Seligman. The committee’s first job will reportedly be to evaluate and further develop OpenAI’s processes and safeguards and present its recommendations to the board in 90 days. The company released a statement claiming that it will then publicly release the recommendations it will be adopting “in a manner that is consistent with safety and security.”
‘AI models’ are prediction systems that are trained on massive datasets to generate on-demand text, images, video and human-like conversation, while ‘frontier models’ are the most powerful, cutting edge AI systems currently on the market.
This information was provided by Techxplore.
