At the 2023 Zhiyuan Artificial Intelligence Conference held today, OpenAI co-founder and CEO, Sam Altman delivered a speech via video connection, which is also Sam Altman speaks to a Chinese audience for the first time.
He said that the world’s attention is mainly focused on solving today’s artificial intelligence problems, and the solutions to many problems are very urgent. AI capabilities are growing rapidly and now need to be applied responsibly to the world.
He believes that in the next ten years, AGI will exceed the professional level of humans in the early 1990s, and may eventually exceed the overall productivity of the largest human company. The potential benefits are huge. He also emphasized that advancing AGI safety is one of the most important areas for the industry to find common ground.
Image source Pexels
For GPT-4 from the completion of pre-training to deployment, Sam Altman said that the OpenAI team spent eight months studying how to pre-judge risks and give countermeasures. The team is on the right path, GPT-4 Alignment exceeds all current code. But for more advanced systems, alignment is still an unresolved problem, which requires new technical methods and strengthened governance supervision. After all, the future AGI may be a system of 100,000 lines of binary code, which is difficult for human supervisors to judge. Whether the scaled model is doing something detrimental.
As for whether OpenAI will open the GPT source code, Sam Altman said: At present, OpenAI has open sourced some models, and as time goes by, more models can be expected to be open source. In the future, as the model becomes larger and larger, ensuring correctness will become expensive, and open sourcing everything may not be the optimal path.
