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.
Let’s take an easy one first, as it is quite obvious what the result of the hypothetical chain of events brings about.
A car crashes into a tree…
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 would have known…
I guess this will be a Hello World, kind of.
After having spent a decade in corporate and startup world I decided to utilize my skills and know-how in “advancing” society by helping to improve AI and NLP technology. Few months later we are at the verge of releasing the public beta of our first project called “crass.ai”.
CRASS stands for Causal Reasoning ASSessment. And this it what it’s all about. It is a benchmarking tool to assess the performance of causal and conditional reasoning of LLM (Large Language Models).
The work on crass.ai progressed quite steadily over the last months though it was slowed down by the COVID-19 pandemic, a bit. But, finally we are at the point of having something to present to a select audience of enthusiasts. So stay tuned for what’s to come.
This problem haunted me for years, so I’m going to do something about it, finally. (I know, everyone has different demons.)
The basic question is as follows: “Why is the “Perfekt” tense in German classified as a form of past tense while in English “present perfect” is always said to be (as the name indicates) a present tense?”
You can find illustrative examples for German and English quite easily:
This is just a piece of poetry, which came to my mind when reading medium in Sept. 2020, so take it with a grain of salt.
Freezing cold showers
All morning must dare
Bill Gates cost my dollars
My greatest dispair
What happened to Missy
I cannot explain
The mist went too dissy
On Buffett’s last reign