The secret to making #AIChatbots sound #smart and #spew less #toxic nonsense is to use a technique called reinforcement learning from #HumanFeedback, which uses input from people to improve the model’s answers. It relies on a small army of #human #data #annotators who evaluate whether a string of text makes sense and sounds fluent and natural. They decide whether a response should be kept in the AI model’s database or removed. https://www.technologyreview.com/2023/06/13/1074560/we-are-all-ais-free-data-workers

In the intro to his keynote on Reasoning with Realistically Imperfect Knowledge, Alexander Gray is comparing gpt-3 rlhf to Shub-Niggurath, a mythical goddess from the Lovecraftian monster universe
#eswc2023 #lovecraft #reinforcementlearning #humanfeedback