Introducing Superalignment
openai.com
We need scientific and technical breakthroughs to steer and control AI systems much smarter than us. To solve this problem within four years, we’re starting a new team, co-led by Ilya Sutskever and Jan Leike, and dedicating 20% of the compute we’ve secured to date to this effort. We’re looking for excellent ML researchers and engineers to join us.

OpenAI’s Superalignment team will address the core technical challenges of controlling superintelligent AI systems and ensuring their alignment with human values and goals.

To accomplish this, they are developing a ‘human-level automated alignment researcher,’ which itself is an AI. This automated researcher will utilize human feedback and assist in evaluating other AI systems, playing a critical role in advancing alignment research. The ultimate aim is to build AI systems that can conceive, implement, and develop improved alignment techniques.

OpenAI’s hypothesis is that AI systems can make faster and more effective progress in alignment research compared to humans. Through collaboration between AI systems and human researchers, continuous improvements will be made to ensure AI alignment with human values.

So, using AI to control other AI; what do you think?

  • mackwinstonEnglish
    3·
    1 year ago

    Given that human values can be pretty screwed up, what could possibly go wrong?

    • TattersEnglish
      2·
      1 year ago

      A lot of politics seems to be a clash of diametrically opposed value systems, so I wonder which ones the AI will have, and will they by any chance be close to the researchers own?