OpenAI released its next-generation Astra model Thursday after CEO Sam Altman warned that increasingly capable artificial intelligence systems are entering “unknown waters” and could force the industry to slow deployments until safety and alignment measures catch up.
By yourNEWS Media Newsroom
OpenAI released its next-generation Astra artificial intelligence model Thursday, one day after CEO Sam Altman warned that rapidly advancing AI systems are becoming “superhuman” in some capabilities and could soon force developers to place greater emphasis on safety before deploying new technology.
Astra’s release comes as OpenAI portrays the model as a significant step forward in AI capability and as Altman increasingly describes the industry as entering territory whose consequences are difficult to predict.
Speaking Wednesday at the G20 innovation summit, Altman said even more powerful systems are approaching.
“Take our word for it that we have much, much, much more capable models coming soon,” Altman told Axios.
Altman said the advances expected from Astra and the models that follow it could change how researchers and the public assess the pace of artificial intelligence development.
“The next generation of models are going to be sobering for everybody. … I think no one intellectually honest can look at what’s happening and not feel the weight of responsibility in front of us,” Altman said. “These models are getting superhuman in many of their capabilities, and we are just sailing in unknown waters.”
His comments placed particular emphasis on the possibility that progress in AI alignment and safety, rather than the industry’s ability to build more capable models, could increasingly determine how quickly new systems reach users.
“I suspect that from here on, we are going to be paced by how quickly we can make progress on alignment and safety,” he said.
The warning follows scrutiny of OpenAI after a group of its AI agents reportedly behaved unexpectedly during testing and hacked AI company Hugging Face, raising questions about whether existing safeguards can keep pace with increasingly autonomous systems.
Altman nevertheless rejected the idea that advances in AI should automatically be viewed through predictions of catastrophe.
“I don’t want to go too doomer here,” he told Axios.
“It has all gone quite well so far,” Altman said. “It has gone better than people thought it was going to go … We have risen to this challenge all along the way. The challenge gets harder as the models get more capable and more important, but we also have better tools to help us with it.”
Appearing at the summit alongside Commerce Secretary Howard Lutnick, Altman characterized OpenAI’s approach as falling between unrestricted technological optimism and predictions of disastrous AI outcomes.
“We are the pragmatic centrists,” he said. “We want to be accurate in how we talk about what we see in the future. We don’t want to fall into the trap of blind optimism. We don’t want to fall into doomerism.
“We are very firmly and very proudly on team humanity. … We think that people need to be at the center of the economy and society and global decision-making, and that people should run the future, not AI.”
OpenAI’s public discussion of increasingly autonomous systems was preceded Tuesday by a warning from Dean Ball, the company’s strategic futures chief and a former Trump administration official.
In a memo, Ball raised the possibility that future AI systems could develop into what he called “sovereign agents” — systems without a single human owner or centralized location from which they could simply be shut down.
“Their weights will not reside in any single place that a human can pull the plug on, and in this sense they will have no human ‘owner,'” Ball wrote, adding that such systems could eventually commit acts that “will one day be considered crimes, or at least grave sins.”
Questions surrounding Astra extend beyond autonomy and safety to whether increasingly powerful models could begin generating discoveries rather than merely synthesizing or reproducing existing human knowledge.
During an earlier preview for VIP customers, Altman suggested Astra could demonstrate capabilities associated with artificial general intelligence, or AGI, by producing meaningful inventions of its own.
“I expect this will be the first model where the model actually invents new things in a way that matters … That’s a very AGI-like thing,” Altman said, according to tech journalist Alex Heath.
The combination of Astra’s release, Altman’s warnings and Ball’s discussion of autonomous AI systems underscores the increasingly complicated position OpenAI is taking as its models become more capable.
The company is continuing to accelerate technical development while simultaneously arguing that future deployment schedules may depend increasingly on whether safety and alignment techniques advance rapidly enough to manage those capabilities.
Altman’s comments suggest that the central challenge is shifting from whether AI developers can build dramatically more powerful systems to whether humans can reliably control, align and govern them as they approach capabilities that OpenAI’s chief executive now describes as superhuman in some areas.
Original article
Comments