OpenAI CEO Sam Altman and Greg Brockman, co-founder and currently the president at OpenAI, have addressed concerns about AI safety following the resignation of Jan Leike. They underscored OpenAI's commitment to rigorous testing, continuous feedback, and collaboration with governments and stakeholders to ensure robust safety protocols.
Leadership Changes and Safety Concerns
Jan Leike, who co-led the superalignment team with Ilya Sutskever, resigned after voicing concerns that safety was being sidelined in favor of rapid advancements. Both Leike and Sutskever's departures have highlighted internal disagreements over the prioritization of safety versus innovation. The superalignment team, formed less than a year ago, was tasked with exploring ways to manage superintelligent AIs.
Jan Leike, in a thread on X (formerly Twitter), explained that he had disagreed with the company's leadership about its “core priorities” for “quite some time,” leading to a “breaking point.” Leike stated that OpenAI's “safety culture and processes have taken a backseat to shiny products” in recent years, and his team struggled to obtain the necessary resources for their safety work.
Altman first responded in a short answer to Leike´s post on Friday, acknowledging that OpenAI has “a lot more to do” and is “committed to doing it,” promising a longer post was forthcoming. On Saturday, Brockman posted a shared response from both himself and Altman on X, expressing gratitude for Leike's work and addressing questions following the resignation.
We're really grateful to Jan for everything he's done for OpenAI, and we know he'll continue to contribute to the mission from outside. In light of the questions his departure has raised, we wanted to explain a bit about how we think about our overall strategy.
First, we have… https://t.co/djlcqEiLLN
— Greg Brockman (@gdb) May 18, 2024
Brockman and Altman highlighted that OpenAI has “repeatedly demonstrated the incredible possibilities from scaling up deep learning and analyzed their implications,” and called for international governance of Artificial General Intelligence (AGI) before such calls were popular. They cited the company's Preparedness Framework, which predicts “catastrophic risks” and seeks to mitigate them, as a way to elevate safety work.
In their detailed statement, Altman and Brockman acknowledged the complexities of ensuring new technology is safe and outlined their approach to AI safety. They emphasized OpenAI's efforts to raise awareness about the risks and opportunities of AGI, advocate for international governance, and develop methods to assess AI systems for catastrophic risks. They also pointed to the extensive work done to safely deploy GPT-4 and the ongoing improvements based on user feedback.
Future Directions and Safety Research
Looking ahead, Altman and Brockman anticipate that future AI models will be more integrated with real-world applications and capable of performing tasks on behalf of users. They believe these systems will offer substantial benefits but will require foundational work, including scalable oversight and careful consideration of training data. They stressed the importance of maintaining a high safety standard, even if it means delaying release timelines.
Altman and Brockman concluded by stating that there is no established guide for navigating the path to AGI. They believe empirical understanding will guide the way forward and are committed to balancing the potential benefits with mitigating serious risks. They reaffirmed their commitment to safety research across different timescales and ongoing collaboration with governments and stakeholders.
