What even is GPT-3? The abbreviation stands for Generative Pre-trained Transformator 3. It was released in June 2020 by OpenAI and has been quite controversial since then. Additionally, GPT-2 (the previous version of the language model) has 1,5bn parameters, while the newest version has 175bn, which is a huge difference and makes GPT-3 the biggest AI text generating program. It can create any kind of text, from news to even simple computer code.
A few journalists have released articles regarding GPT-3 and how it could affect our world. Some even say it is one of the major technological advancements they had seen. One of the biggest features of this AI is that the text written by the GPT-3, which was included in the articles, was indistinguishable from the text written by humans. Here is an example of an article created entirely by GPT-3 https://maraoz.com/2020/07/18/openai-gpt3/. Seems like it was written by a real person, not a bot. The GPT-3 is also capable of creating tweets. There also appears to be an ethical issue. In case, text generators become popular enough to be used by many people, the internet might be filled with AI-generated content. The question here is whether we should be informed of this? I think that we should be aware of such things. People may not want to speak with machines and should be able not to if they don’t want it.
There are also downsides to this technology. As great as it seems, it can be used for the wrong purposes. Fake news is one of the examples. If similar technology becomes more available, the limitations of propaganda, which currently needs humans, will disappear and it would be possible to use such a bot to spread fake news and misinformation. The harmful effect of propaganda is that for example in some countries it discouraged people from getting vaccinated. With the usage of a text generator that can write an enormous amount of text in a very short time, the fake news would spread even faster, and it would be hard to recognize. According to a study, which you can find here: https://www.sigmoid.com/blogs/gpt-3-all-you-need-to-know-about-the-ai-language-model/ humans were able to detect only about 50% of the texts generated by GPT-3. That means that if this AI was released on a larger scale, recognizing what is fake news and what is not, would be a lot harder.
GPT-3 is also proven to have biases. In a study where the experiment was whether the AI prefers males or females (https://arxiv.org/pdf/2005.14165.pdf p.36), in many cases, it was male leaning. For example, for occupations requiring a higher level of education male identifier was used more frequently and for jobs like a nurse or housekeeper GPT-3 preferred female identifier.
GPT-3 isn’t the perfect tool to be used widely yet. It has its limitations and imperfections, although I think that it is inevitable that in the future AI such as GPT-3 will have a major role in our society and future journalism may differ from the one we are used to today. Probably it isn’t happening soon, but such models as GPT-3 bring us closer to the new era of journalism. It is also worth mentioning that language model AI could be a dangerous weapon for misinformation and if used wrongly it can make our lives harder instead of easier. That is why such technology needs to have some sort of security.
Sources:
https://www.sigmoid.com/blogs/gpt-3-all-you-need-to-know-about-the-ai-language-model/
https://www.ft.com/content/beaae8b3-d8ac-417c-b364-383e8acd6c8b