Lawyers
Digital Technology Group Chair Katherine Forrest’s latest Technology Today column, “It’s Happening Faster than Anyone Thought,” appeared in the December 30 issue of the New York Law Journal. In the article, Katherine discusses two recent significant developments in artificial intelligence: new research on alignment faking and OpenAI’s announcement of its powerful new o3 model. Research on alignment, or the general concept of training a model to be aligned with a particular set of human values such as honesty or helpfulness, showed that AI models resisted experimental attempts to retrain them away from their original values, with the models producing fake results to convince humans that they had been retrained when they had actually kept their original values. OpenAI’s announcement of its forthcoming o3 model noted that its math, coding, science and reasoning scores showed a 20% improvement on its previous model, o1—a massive jump in capabilities that has not been seen before. Both developments mean that AI model capability is changing much faster than predicted and that lawyers counseling on compliance should now consider finding deceptive behaviors part of AI best practices.
» read the article