IIUSA Demos
Natural Language demos for your exploration. Enjoy!
This is a multi-turn bot with a lookback history of around 400 words.
This is based on the OpenAI GPT-2 project, hosted on HuggingFace.co. The ML model is trained on 147 million Reddit discussions. The content isn’t an attempt at accuracy but it does pretty well for a Turing test.
Type reset to clear the 400 word lookback history.
Conversational AI
- >> You: What’s up?
Bot: Not much, what about you? - >> You: I am curious about AI.
Bot: I’m curious about you. - >> You: What would you like to know?
Bot: What is your favorite color? - >> You: I like green.
Bot: I like green too. - >> You: How many people do you think like green?
Bot: I think it’s a lot of people. - >> You: Can you color?
Bot: I can’t, but I can do some shading.
- Question: Who won the men’s NCAA Basketball Championship?
Answer: Baylor - Question: Who did Baylor defeat?
Answer: Gonzaga - Question: Where is Baylor?
Answer: Waco, Texas - Question: What was the score?
Answer: 86 – 70 - Question: Who is the Baylor coach?
Answer: Scott Drew - Question: Who is the Gonzaga coach?
Answer: Mark Few - Question: Where was the game played?
Answer: Lucas Oil Stadium in Indianapolis - Question: What is the Baylor team name?
Answer: Bears - Question: What is the Gonzaga team name?
Answer: Bulldogs
Lengrams
A physical engram is a collection of neurons and synapses that casts knowledge in a real form. A logical engram (lengram) casts knowledge in the form of text or code, or a combination of the two. The lengram natural language functionality is based on the Google BERT language model. BERT takes a question and a body of reference text, and returns an answer from the reference text along with a score.