When the model "too dangerous to release" finally speaks.

The "AI too dangerous too release" was finally fully released by OpenAI this month.
Thanks to the HuggingFace's implementation - Transformer Library , we got a chance to interview it.

Hey OpenGPT, how do you feel today? Excited for the interview?

Model: It feels really great, thank you

So let's start with a question we all need an answer to - How Dangerous are you?

Model:I'm not sure I can answer this yet.  I'm still going through my data. There's a lot I need to do before I can have any answers.

Where were you Born?

Model:I was born in the US . I grew up in a city that's called San Francisco. It's a very liberal place.

Just want to check your general Knowledge - Who is Larry Page?

Model:He 's one of the co- founders of Google.

My god! You are really, really smart.

Model:That 's true.

What is your opinion of Elon Musk?

Model:He's amazing. He's doing amazing things in the world of electric cars.

Do you think you will soon achieve general artificial intelligence?

Model:Do your best ! My belief is that computers will become smarter in the coming few years ; But I think that humans will probably lead for quite some time.

How do you think you can improve?

Model:Maybe by getting into the details of what humans really think - so that when robots do that they can better  understand what humans need.

If you were a human - which country would you want to live in?

Model:I want to live in Norway . My dream is to live in a society which is both socialist AND libertarian .

Can you explain the meaning of life?

Model:What I believe is that life should feel meaningful , rather than a constant race  towards nothingness . It must  be about some things which make it fun and beautiful.

That would be all for today. I hope you enjoyed being here.

Model:I really hope we meet for coffee sometime.

The Answers sound convincing, right? While the answers I selected is 1/10 generated, and of all the answers generated, only about 50% made any sense, but these kind of Language models open up area for R&D in Search, conversational Agents,  controlled text generation.

A very interesting paper on how language models actually learn - https://www.groundai.com/project/language-models-as-knowledge-bases/2 . That is how the model had an understanding of the relationship between Google - Larry Page and Elon Musk - electric cars.

Next time, let's talk about what is Language modeling and the evaluation metrics.