hutmagier

The pace of development in the field of large language models has been staggering throughout the last months. Just when I settled on a new candidate for “Talking counterfactual” a new model with supposedly superior capabilities was released, be it LLaMA (now already LLaMA 2), MPT, Falcon, Cohere’s models, Claude, Palm 2 aka Bard etc […]

It has been quite some time, but here we are again back at the counterfactual game and today I have a premiere for you. We have a guest who will be the second time on the show presenting an improved version of itself. I’m talking of no other than BlenderBot 3 (for convenience followingly written

Throughout the year 2021 many new interesting models were released (for instance GPT-J [1] or Megatron-Turing NLG [2]) others got updated (ERNIE 3.0 [3], PLATO-XL [4] and blenderbot 2.0 [5]), so it is quite difficult to pick the next target for our talking counterfactual series. Though we have been looking mostly into chatbased systems as

I assume GPT-3 from OpenAI does not need an introduction. Released in summer 2020, it is a 175 billion parameter transformer based behemoth delivering state of the art performance in many NLP tasks. So, it’s an obvious choice to try out and see how it handles counterfactual conditionals. Other than DialoGPT and Blenderbot, GPT-3 is