Talking counterfactual #2: Microsoft’s DialoGPT

In episode two of our new series, we take on Microsoft’s DialoGPT. It is a GPT-2 transformer-style architecture trained on 147 million Reddit discussion threads released in November 2019 [1,2]. Now, Reddit might not be the best spot to train your AI (so we heard 😉), but it is one of the largest data sources for human discussions. Let’s see if DialoGPT can cope with crass.ai’s flavour of counterfactual conditionals.

Talking counterfactual #2: Microsoft’s DialoGPT Read More »