How good is Open AI Text Writer? Mind Blowing
How good is open AI writing capabilities? Really good. When given a brief statement about Open AI’s new DALL-E software to create images from text, it gave the following article. Not only is it impressive that the AI made the article, but as far as I can tell there is no Echolingo company.
Open AI Blew our minds again releasing examples from their new DALL·E: Creating Images from Text AI software. The future of Creative Creativity? Well, not quite yet. But it’s been interesting to watch Open AI develop creative systems that mimic human intelligence, albeit it with a slight edge on machine perception and motivation.
Echolingo, like most AI (AI training and research) start-ups, relies on either humans or AI to perform a task. Echolingo automates language learning. The quality of Echolingo’s algorithms was enough for MIT Technology Review to rate it the Best New Technology of 2016. The two algorithms Echolingo uses are labelled as either Deep Learning or Recurrent Neural Networks. Deep Learning refers to a form of artificial intelligence (AI) in which a machine is programmed to recognize patterns and respond to them with what is known as supervised learning, or when a human explains to a machine the answers to a series of questions. A Recurrent Neural Network is used for tasks where a computer program remembers what it has done and comes up with the next logical step. Because of the predictability of the network, it is able to generate more correct answers faster than a neural network.
While it’s always interesting to see what AI can do, the results of experiments with AI don’t always make it into the mainstream. However, we all know that even the smallest innovations have the potential to change the world and that is certainly true for Echolingo.
Echolingo is very aware of the challenges it has, specifically in the machine perception space. According to CEO Kaiming He said, “There are many ways people describe the word “parrot” but we can’t use the most efficient. If we want to build an object recognition system, for example, there are already thousands of words that can be used, each with different meanings. Most of these are already present in the database of human beings. A researcher must spend a lot of time just to look through all these words and find the right match. While, on the other hand, we do not need to understand everything, it is still hard for our neural network to understand the relationships between words. It does that the same way as a human does: by looking at a big dataset. Only with this kind of task can we truly understand the limits of the current models and how far we need to go before we can build an object recognition system.”
An impressive statement, to be sure, but for many an encouraging one. Echolingo is working on solutions to these problems. He goes on to say,
As of now, we use hand-designed networks. But we have a machine learning team that will turn this research into a product in the future. The result will be an improved learning experience and more accurate results for users.
I am no Echolingo user, but I have interacted with the company’s product. I think what’s interesting about the company is that it is not only disrupting the machine learning space, it’s disrupting the higher education space too. Echolingo can currently take open (open source) information and create learning material specifically for its own purpose. This is an area of higher education that many companies would be eager to explore, as it’s one of the most profitable types of education for employers. Echolingo’s open-source approach allows it to completely control its creation, while at the same time allowing users to play around with the various pieces of information that the company has available. This is a smart and unique way for the company to engage with customers and promote their service. It’s the ultimate in making open source, profitable.
The fact that Echolingo is making a difference for its users by being flexible in the creation of the software means it’s also helping to make the entire machine learning space better.
The impact of AI is huge, and many industries are being disrupted by it. We are very proud of the work Echolingo is doing and we are looking forward to hearing what they come up with in the future.
- AI writing is already amazing. We tried to find articles that AI might be copying and everything we looked for on Google was quite unique. Just image what the next few years will bring. This is both frightful and amazing.
- As far as we can tell the AI created both a new company and a CEO that doesn’t actually exist. In future articles we will share the advancing technology Open AI has created that creates images with words. Imagine a text writer such as this being paired with image creator like the following. AI wouldn’t just make a new company, but create a fake store front that doesn’t exist. AI wouldn’t just make up a CEO’s name but also what he looks like and potentially in the future what he sounds like, perhaps even videos of this fake CEO speaking.
- Where does it go next? Although there fear of where this technology could go, there is so many creative uses this technology could bring. Imagine AI video games or simulations. Every video game would be different according to the AI.
What other ideas do you have for this technology?