What Humly is doing together with OpenAI

OpenAI was founded in 2015 by Elon Musk, and has a stated goal of promoting Friendly Artificial Intelligence, FAI, that benefits humanity as a whole.

After the summer OpenAI released their new GPT-3, Generative Pre-Trained Transformer 3 - an autoregressive language model that can produce human-like text.

We’ve always kept an eye on what OpenAI is doing. They seem to have good values and focus on making sure it is used for the greater good.

When they released their beta program it was for a select few and not widely available. In fact, they had their doubts about releasing it at all since it was so good that it can really act like a human and write texts all on it’s own. It seems we’re approaching a disruption in text generation.

I reached out to the co-founder and CTO, Greck Brockman and described our mission statement to “make a workplace where human interactions are made better by technology” and we were immediately accepted. I think our organizations have a lot in common.

GPT-3 is an amazing AI model, it is the biggest in the world right now with it’s 175 billion parameters - 10 times bigger than number 2. Those of us that have had the privilege to test it are blown away by its capabilities.

We would describe it as a model that knows English better than any human on earth and at the same time knows almost everything. It could almost pass the Turing test, you know the test that if a human can not tell if it is a computer you are speaking to it will pass the test.
The Turing test was introduced in 1950 and up until now, no computer model has passed it.

We’ve had a few chat sessions with it and we think GPT-3 is definitely a potential candidate for success... If something would give it away it is possibly that it knows TOO MUCH. You can ask it what atoms are in propane, what is beyond a black hole, how to fix a certificate error in Chrome on a MacBook and it even knows the link to the blog entry where the certificate error has a workaround. No human would know that, so it fails because it is too good.

I’ve already started using GPT-3 in my daily life. For example I’ve asked it to explain complicated legal passages in contracts with customers where it can output a two line, super clear explanation for what is really meant.

We definitely see a great potential in AI and are already working with a few ideas, we’re also looking at Google's Tensorflow and Microsoft AI. Handled with care and respect I believe that AI will be a natural part of the future.

Just like when the Internet changed the world AI will be a force of disruption, and change our lives - forever. 

Leave a comment below to share your experience or ask questions that you might have.