[ad_1]
\n<\/p>\n
There is currently no known indefinitely scalable solution to the alignment problem. As AI progress continues, we expect to encounter a number of new alignment problems that we don\u2019t observe yet in current systems. Some of these problems we anticipate now and some of them will be entirely\u00a0new.<\/p>\n
We believe that finding an indefinitely scalable solution is likely very difficult. Instead, we aim for a more pragmatic approach: building and aligning a system that can make faster and better alignment research progress than humans\u00a0can.<\/p>\n
As we make progress on this, our AI systems can take over more and more of our alignment work and ultimately conceive, implement, study, and develop better alignment techniques than we have now. They will work together with humans to ensure that their own successors are more aligned with\u00a0humans.<\/p>\n
We believe that evaluating alignment research is substantially easier than producing it, especially when provided with evaluation assistance. Therefore human researchers will focus more and more of their effort on reviewing alignment research done by AI systems instead of generating this research by themselves. Our goal is to train models to be so aligned that we can off-load almost all of the cognitive labor required for alignment research.<\/p>\n
Importantly, we only need \u201cnarrower\u201d AI systems that have human-level capabilities in the relevant domains to do as well as humans on alignment research. We expect these AI systems are easier to align than general-purpose systems or systems much smarter than\u00a0humans.<\/p>\n
Language models are particularly well-suited for automating alignment research because they come \u201cpreloaded\u201d with a lot of knowledge and information about human values from reading the internet. Out of the box, they aren\u2019t independent agents and thus don\u2019t pursue their own goals in the world. To do alignment research they don\u2019t need unrestricted access to the internet. Yet a lot of alignment research tasks can be phrased as natural language or coding\u00a0tasks.<\/p>\n
Future versions of\u00a0WebGPT<\/a>,\u00a0InstructGPT<\/a>, and\u00a0Codex<\/a>\u00a0can provide a foundation as alignment research assistants, but they aren\u2019t sufficiently capable yet. While we don\u2019t know when our models will be capable enough to meaningfully contribute to alignment research, we think it\u2019s important to get started ahead of time. Once we train a model that could be useful, we plan to make it accessible to the external alignment research\u00a0community.
<\/p>\n<\/div>\n<\/div>\n