Two Minute Technology: GPT-3

Two Minute Technology: GPT-3

What is GPT-3?

Third generation Generative Pre-trained Transformer, or GPT-3, is a neural network machine learning model trained using internet data to generate any type of human-like text. Developed by OpenAI, it requires a small amount of input text to produce large volumes of relevant and sophisticated machine-generated text.

GPT-3's deep learning neural network is a model with over 175 billion machine learning parameters, making it the largest neural network ever produced. To put things into scale, Microsoft's Turing NLG model had 10 billion parameters. As a result, GPT-3 is better than any prior language-prediction model, since it is able to generate text that is convincing enough to seem like a human could have written it. Below are two example conversations using GPT-3 technology. The first video is a conversation on existential themes between two AI's. The second video is a conversation between a person and AI.


What can GPT-3 do?

The focus of GPT-3 is on generating natural human language text. However, generating understandable content is quite the challenge for machines that don't really understand the complexities and nuances of language. That is why GPT-3 uses text found on the Internet to train itself on how to produce realistic human text.

GPT-3 has been used to create articles, poetry, stories, news reports, websites and dialogue. It is also being used for automated conversational tasks, such as responding to any text that a person types into the computer or phone with a new piece of text appropriate to the context. We call these virtual assistants. GPT-3 is also used in customer service centers, often called chatbots that help answer customer questions. There are researchers and developers working on expanding the usage of GPT-3. For example, researchers are working on creating AI therapists using GPT-3 models. This video shows an example of how GPT-3 could be used to create AI therapists. A potential improvement of this model could be slowing down the output response. Slowing down GPT-3 response can provide a more human touch to the interaction. What do you think could improve this AI therapist?

Risks and Limitations

The biggest limitation is that GPT-3 is not constantly learning. Since it is pre-trained, it doesn't have an ongoing long term memory that learns after each interaction. Like all neural networks, GPT-3 lacks the ability to explain and interpret why certain inputs result in specific outputs. Furthermore, it has a limited input size. A user cannot provide a lot of text input for output. GPT-3 can only handle input text that is a few sentences long.

A primary concern with GPT-3 is the risk of machine learning bias. Since the model was trained on Internet text, many of the biases that humans exhibit on the Internet is exposed to GPT-3. For example, researchers at Middlebury Institute of International Studies found that GPT-3 is particularly adept at generating radical text that mimics conspiracy theorists and white supremacists. The quality of these generated text is high enough that people are worried that GPT-3 will be used to create "fake news" articles. For more information on GPT-3, feel free to check out OpenAI's website at the following link!

GPT-3 Powers the Next Generation of Apps

Message from the Blogger

Hello! Thank you for reading this blog post. I hope you enjoyed it and learned something new today. See you next time!


Comments

Post a Comment

Popular posts from this blog

Two Minute Technology: Machine Learning

Image
Two Minute Technology: Machine Learning Any learning entity - be it an animal, human or a machine begins with an initial set of observations - raw data . This kind of data can stem from interactions, transactions, information exchange, examples, instructions or experiences. A brain - whether it belongs to a human or animal, tries to look for patterns inside that initial data and then uses it to perform actions or make decisions. This same phenomenon can be applied to machines, using machine learning. Machine learning is a subfield of Artificial Intelligence. According to Arthur Samuel, a computer scientist, machine learning gives computers the ability to learn without being explicitly programmed. Another computer scientist, Tom Mitchell, took the same details of machine learning, but broke it down further in his definition. I find Mitchell's definition more fitting to describe exactly what machine learning is. "Machine learning is a well-posed learning problem: a computer...
Read more

How apple has dominated the phone company

Image
  In the growth of smartphones in 2007  apple made basically the first touch screen held cell phone in which it was different from all of the others that was around. This tech guru Steve jobs knew at the time that he had something special. At this time he was competing with companies that were making flip phones, blackberries or even the slide up phones at the time. This was an outside the box move in which no one was doing at the time or even thought about doing.  Cell phones have come a long way back in the day before that they were carrying like bricks with them and you can tell it was a bad time. This mix of new styles of phones which basically turned the whole industry into touch screens and smartphones. This company lost their lead Steve jobs who was the man with the plan and stated this trend but the company has been able to continue to grow.  Many people don't know that the company made a bad deal with AT&T at the time. This deal was made to get all of t...
Read more

Ryan's Top Ten Albums - In Rainbows - Radiohead

Image
  In Rainbows - Radiohead Released 2007 - Genre: Art Rock, Experimental Rock While the albums I talked about in this blog are not ranked in any particular order, I can confidently say that this is my favorite album. Radiohead has always been one of my favorite bands, my original idea for this blog was to do an entry on each of their nine studio albums. In Rainbows came at a very unique point in the band’s career, as they had already seemingly reinvented the wheel twice, with the release of OK Computer in 1997, and Kid A in 2000, and their few albums since had been highly regarded but never to that same level. In Rainbows was their attempt at reinventing their sound again and doing something groundbreaking, and they delivered. The buildup to the album's release was very unique, as the band promoted the album almost solely on their own website, where they boasted a pay-what-you-want scheme for the album, where fans could pay any amount (or nothing at all) to download the album. I...
Read more