Large language models need red teaming from people unlike their builders—artists, musicians, people from different backgrounds—to catch problems before launch. As companies build thousands of AI agents, nobody knows who's responsible for keeping them safe and accurate.
Chapters
One thing that's important about artificial intelligence and building models, especially large language models, is that they are hinged on inclusive input prior to production, right? There's an entire concept called AI red teaming, of similar to cybersecurity but very different in the AI world, where we need people to just test a model to make sure it doesn't do anything inappropriate. Yeah. But we don't need people who built the model to do that. We need completely divergent thinkers. We need artists and musicians. We need people who were raised poor, people who went to university. Like, we need all of these different perspectives because the models serve lots of different people, right? And so now companies are going to be looking for these fragmented roles to say, hey, can you come and help me train this model? Can you come and help me teach this model? In a world where we're building thousands of agents, thousands of models in a company, who's going to actually make sure that those models are accurate?
