On May 28, OpenAI announced that the board of directors had established a security advisory committee to control the direction of AI development.
The Safety and Security Committee’s first priority will be to evaluate and further evolve OpenAI’s processes and safeguards over the next 90 days. At the end of the 90 days, the Safety and Security Committee will share their recommendations with the full Board. Following full Board review, OpenAI will publicly share updates on the adoption of the recommendations in a manner consistent with safety and security.
OpenAI announced that it has started training the next generation of cutting-edge models. The AGI vision has become more pragmatic. OpenAI said in the announcement that it recently started training its next generation of cutting-edge models in an effort to surpass the current GPT-4 large model. We expect these systems to enable us to reach higher levels of capability on the path to artificial general intelligence (AGI). While we are proud to build and release models that are industry-leading in both capability and safety, we welcome a vigorous debate at this important moment.
As a background, OpenAI released GPT-4 in March 2023. With just the sketches drawn by President Brockman in his notebook, GPT-4 can build a real website. I believe that many stock investors began to believe that "AI may really change something" after seeing this scene.
Since then, OpenAI has gradually launched a series of text, image, audio and video generation applications around GPT-4. Just two weeks ago, the company released its latest version of GPT-4o. By significantly increasing the interaction rate of the model, "real-time interaction" like real-person conversation is achieved. In the final analysis, GPT-4o, which is slightly stronger in all aspects, still does not escape the capabilities of GPT-4.
Because of this, the outside world has always had high expectations for "GPT-5". However, according to OpenAI's latest statement, the next generation of large models may not be available to the public until next year. AI models usually take months or even years to train, and the development team needs to make several months of fine-tuning before releasing them to the public.
As OpenAI builds AI that gets closer to the level of human intelligence, the company has become more cautious about its statements about general artificial intelligence.
Anna Makanju, OpenAI's vice president of global affairs, said OpenAI's mission is to build general intelligence that can perform tasks at the "current human cognitive level."
Sam Altman said a few years ago that OpenAI's ultimate goal is to build a "super intelligence" that is more advanced than humans. He also said that he spends half of his time thinking about how to build "super artificial intelligence."
