Talking counterfactual #1: facebook’s BlenderBot
With crass.ai getting ready for public beta, we thought it’s time to take it for a spin and have a dance with some of the recent LLMs (large language models).As a starting point, we chose facebook’s BlenderBot (for convenience followingly written without the second capital “B”). The Blenderbot is a multipurpose Q&A open-source chat bot, so it is an ideal fit for the question-styled counterfactual conditionals employed by crassi.ai inaugural test set. (btw. counterfactual conditionals do not have to be framed as questions, as you could say: “John would have picked the red team, if he had known they would win.”, but construing them as questions makes it easier to have a starting point for judging the strength of an LLM, as the potential answer is easier to classify.)
Talking counterfactual #1: facebook’s BlenderBot Read More »