\n<\/p>\n
Training AI models with other AI models has become a very popular goal for neolabs \u2014 and now, a researcher in Anthropic\u2019s fellows program has given us an early look at what it might look like in practice.<\/p>\n
On Friday, Anthropic published a new paper titled \u201cAutomated Researchers Can Reliably Mitigate Alignment Failures,\u201d <\/a>detailing how AI systems could reliably improve a model\u2019s performance on a set of alignment benchmarks. When given 10 benchmarks for specific misaligned behaviors, the automated systems were able to improve performance on every single one without degrading overall performance.<\/p>\n Led by Anthropic Fellow Chen Yueh-Han, the system replicates much of the traditional approach to research. Each automated system searches the available literature, proposes a method, and trains the model using that method for 30 minutes, gradually increasing the benchmark over several iterations. Effective methods are preserved while ineffective ones are discarded, allowing the system to operate quickly and at a great scale.<\/p>\n \u201cOverall, these results provide early evidence that automated alignment post-training could become practical in the near term,\u201d the paper reads.<\/p>\n The paper is a step toward recursive self-improvement<\/a>, which many see as the next significant step in AI progress. If models can improve their own alignment training, it\u2019s plausible they could improve training practices more broadly \u2014 at which point, human AI researchers might soon become obsolete.<\/p>\n The paper isn\u2019t shy about addressing this idea, explicitly comparing the Automated Alignment Researcher (AAR) to its human equivalent. \u201cThe best AAR method beats what experienced humans propose, on average within six hours,\u201d the paper reads. \u201cHuman guided research directions do not lead to stronger performance.\u201d<\/p>\n There\u2019s even a cost comparison, in case anyone wasn\u2019t convinced. \u201cAn AAR costs roughly $4 per hour in API inference against the $150 per hour we pay our human researchers.\u201d<\/p>\n In fairness, the paper also points out a few limitations to this approach. The automated system only works insofar as the benchmarks reflect the actual alignment goals, and even then there\u2019s significant work to be done in establishing and maintaining those benchmarks \u2014 not to mention maintaining and expanding on the literature the automated researchers are draw from.<\/p>\n<\/div>\n When you purchase through links in our articles, we may earn a small commission<\/a>. This doesn\u2019t affect our editorial independence.<\/em><\/p>\n