AI-powered
podcast player
Listen to all your favourite podcasts with AI-powered features
The Challenges of Model Evaluation
There's lots of different ways that people evaluate these models and hugging faces leaderboard. My kind of dream in the long term for machine learning is to turn into something that's more like a taxonomic science. Instead of studying an evaluation of one model at a particular point, we start to build family trees of these models. We try to generalize our findings about those things to families of models. So you can imagine a whole field of science based around measuring properties of machine learning models and then trying to see how they diffuse through the family trees of this.