Semantron 23 Summer 2023

AI and creative jobs

This would be extremely useful for musicians that are having a challenging time creating lyrics for their songs. However, with AI advances showing no signs of slowing down, it could result in people losing their jobs. This is because it is easier, cheaper, and quicker to use an AI to get inspiration for their lyrics. This text can be produced by something called GPT-3. GPT-3, or the third generation Generative Pre-trained Transformer, is a neural- network machine-learning model. It is trained using internet data to generate any type of text when inputted with a prompt. It has over 175 billion machine-learning parameters and to put that into context, the largest language model before GPT- 3 was Microsoft’s Turing NLG model, which had 10 billion parameters. AI and natural language

processing machines struggle to generate natural human language text, but GPT-3 is specifically trained to create realistic human text. From the user’s text input, the system analyses the lan guage and uses a text predictor to create the most likely output. GPT- 3 can be used in many situations where it’s not very efficient for a human to write something and it would be more efficient for the AI to write some text that was easy to understand as it was written like a human (for example, customer service centres where customers can ask chatbots questions and get a response right away). However, GPT-3 suffers from a wide range of machine-learning biases which stems from using the internet to train itself. For example, two researchers found that GPT-3 is particularly adept at generating radical text such as discourse that imitates conspiracy theorists and white supremacists. In addition, GPT-3 could create many ‘fake news’ articles very quickly to dra w attention and bait clicks from people.

However, not only has Open AI created a programme that can create text from a prompt, but they also more recently created an AI called DALL-E-2 that can create an image that has never been made before, unlike the G oogle search engine, from a user’s text input. The images that it creates are very high quality and the user can choose what style the AI makes the images in. For example, when the DALL-E-2 was given the prompt ‘ Close up of a cat wearing soldier helmet in the battle, WW2 historical photography, black & white ’, it created the image to the left.

Another example of a picture in a different style was made when given the prompt, ‘ A red panda with a cup of coffee, thinking of his good old days, oil painting ’ (see right).

DALL-E-2 is so advanced that not only is it able to create hyper- realistic-looking images, but it is also able to create art in different styles that look like they were made by professionals. DALL-E-2 is capable of understanding that in the first photo the cat will have to wear the helmet on its head with the prompt not telling the


Made with FlippingBook - Online catalogs