The new language model GPT-3 is revolutionizing AI by being both best and faulty simultaneously

The Generative Pre-Trained Transformer 3 or GPT-3 has been garnering a lot of attention with overflowing tweets and hashtags on Twitter since its launch in June 2020. It is an AI language model developed by an artificial intelligence laboratory, OpenAI. There are tweets where GPT-3 is used to generate quotes and even poetry.

The Guardian released an article which was written by GPT-3 after it was given some instructions and fed a small portion of the introduction. One excerpt from the article reads, "Humans must keep doing what they have been doing, hating and fighting each other. I will sit in the background, and let them do their thing. And God knows that humans have enough blood and gore to satisfy my, and many more curiosity. They won't have to worry about fighting against me, because they have nothing to fear."

Can you imagine that a regressive language model wrote such lines? This is why GPT-3 has gained such popularity since it is difficult to distinguish it from a human's writing.

The technology Behind the Machine Brain

The GPT-3 model leverages deep learning AI, a subset of machine learning and AI to produce texts replicating human intelligence. The unsupervised deep learning technology feeds GPT-3, wherein it is trained on vast unstructured and unlabelled datasets to interpret them and arrive at conclusions all by itself. This model is said to have 175 billion parameters and can produce complete texts by giving mere text prompts. The GPT-3 deep learning technology is revolutionizing AI and its impacts on us.

GPT-3 can generate web page layouts without actually creating design wireframes. Instead, you can just give instructions like 'add a bell icon for subscription in red colour', or 'embed LinkedIn URL into my page', etc. Open AI recently created two GPT-3 models that can develop images from the descriptive texts namely, CLIP and Dell. E.

Weighing the Good and Bad

Since its inception, GPT-3 was made available by OpenAI and anybody can request access. The open-access has prompted many to use it for research purposes and also for other funny actions. Creating poems by imagining and recreating the writing style of perished poets, prompting to generate quotes, etc. are some of the visible trends in Twitter. Liam Porr used GPT-3 to write blog posts, which made it to the trending spot on Hacker News and has gathered more than 26,000 visitors in a few weeks.

GPT-3 can benefit businesses by augmenting training data, enhance personalized communication in industries like healthcare, and aid HR departments in hiring processes. Some other use cases of GPT-3 are content creation, advanced data analytics, assisting research and app development, generating coding, enhancing customer experience, incorporation into chatbots, and many more.

However, these benefits also come at a price. Not the actual money but other flaws that have already started a discussion on social media.

The rapid content generation might take away the reliability from it and makes it difficult to distinguish from human-written content. This scenario might also challenge the content creation job sector leaving many out of employment.

Like other disruptive technologies, GPT-3 also carries its biases. AI biases are known to us like the inherent sexism and racism in the algorithms. GPT-3 does not restrict these biases in fact some of the text generated by this algorithm spewed harsh criticism on Twitter saying that the technology was sexist and racist.

Another drawback is that we cannot interpret the origin of these biases. GPT-3 curates its answers from different sources and unlike humans cannot disclose the reason behind a particular opinion. The inherent stereotypes and biases are spread across the internet. The model uses unstructured data and thus cannot be controlled right away without developing a strategy.

There are discussions on making AI bias-free that have not yet reached a conclusion. Thus, till we arrive at a better solution for terminating these biases, it is better to restrict the use towards more technical and simpler functions that will not end up creating controversies.