Yesterday, OpenAI announced their new Safety and Security Committee which will be led by Sam Altman and OpenAI Board Directors Bret Taylor, Adam D’Angelo, and Nicole Seligman. The committee will be accountable for making recommendations to the full Board on safety and security decisions for OpenAI projects.
The committee will also include technical and policy experts from OpenAI, including, Aleksander Madry (Head of Preparedness), Lilian Weng (Head of Safety Systems), John Schulman (Head of Alignment Science), Matt Knight (Head of Security), and Jakub Pachocki (Chief Scientist) who recently took over Ilya Sutskever’s position. Former cybersecurity officials, Rob Joyce and Jon Carlin will also be consulted.
Interestingly, the announcement comes at a time when the next superior model has been confirmed to be in works. Altman even hinted at the same at the recent Microsoft Build 2024 event.
90-Days Window
The company blog says,“OpenAI has recently begun training its next frontier model and we anticipate the resulting systems to bring us to the next level of capabilities on our path to AGI.” This probably being a direct hint at GPT-5. Satya Nadella had also indirectly confirmed the same in a recent interview.
Satya Nadella, when asked if he tried GPT-5, initially responded, "Oh yes, of course." When prompted to correct himself, he did not deny the statement. pic.twitter.com/g2b2HXskXX
— Ale𝕏 Fazio (@alxfazio) May 21, 2024
The initial task of the Safety and Security Committee will focus on evaluating and developing OpenAI’s processes and safeguards over the next 90 days. Post which, the committee will share the recommendations to the complete Board for review and then public announcement.
Interestingly, GPT-5 was rumoured to release during the summer which coincides with the 90 days window for evaluation. However, with the US elections slated to take place in November this year, it is also presumed that GPT-5 may be released after that.
The upcoming GPT-5 model is said to have superior qualities and even probably act as an AI agent. It is said to have multimodal capabilities supporting text, image, audio and video from the start. The latest text-to-video generation model Sora will also be embedded into the system.
While the company is gearing up for its next model, the Safety and Security committee announcement is likely a preemptive step from the company’s side, considering how the safety alignment team was recently abandoned after Ilya and Jan Leike’s exit. Further, users were quick to react to the announcement with many being critical of the same.
Source: X
The post GPT-5 is Arriving Sooner Than Ever appeared first on AIM.
