The use of AI (Artificial Intelligence) language models to write articles has been the subject of much discussion and interest recently, particularly with the release of ChatGPT, an online chat bot developed by OpenAI. These language models have the potential to revolutionize the way we produce written content. But before we dive into the potential dangers and ethical concerns, let’s first discuss what a language model is.
A language model is a type of AI that is trained on a large collection of text data to predict the likelihood of a sequence of words. This allows the model to generate text that is coherent and follows the rules of the language it is trained on. The result is text that is virtually indistinguishable from human writing.
However, the use of AI language models to write essays and other academic papers raises some serious ethical concerns. As these models become more advanced and more widely used, there is a danger that they will be used to cheat on exams and other assessments, giving an unfair advantage to those who use them.
One area where the use of AI language models to write essays is particularly concerning is in the realm of college admissions. With the intense competition for spots at top universities, some students may be tempted to use these models to write their application essays in order to stand out from the crowd. This could give an unfair advantage to students who use AI-generated essays, and could potentially harm the chances of students who write their own essays.
Another potential danger of using AI language models to write essays is that it could lead to a homogenization of ideas. These models are trained on large amounts of text, and as such, they may not be able to produce unique or original ideas that deviate from the norm.
One way that AI companies can address these concerns is by implementing measures to prevent the use of language models for cheating. For example, they could develop algorithms that can detect AI-generated text and flag it for review by a human. Additionally, they could work with schools and universities to establish guidelines and best practices for the use of language models in academic settings.
Overall, while the use of AI language models has the potential to greatly enhance our ability to produce written content, it is important to carefully consider the potential drawbacks and ethical concerns. By working together, AI companies and educational institutions can help ensure that these powerful tools are used in a responsible and ethical manner.
—
As you may have guessed, I, Taran, didn’t write a single word of the article above. It was written by ChatGPT, the subject of the article. All of the text that follows is original.
ChatGPT is a significant improvement over other similar chatbots developed in recent years. This is largely because it is much better at connecting its responses to previous parts of a conversation. Many models aren’t even designed to do this at all. This allows for more human conversation, and importantly, follow-up questions—like those we use every day in ordinary dialogue. This provides for a much better chatting experience. Consider the comparison of Cleverbot, a chatbot released in 1998, on the right, with ChatGPT, on the left.ChatGPT is also notable for refusing to answer questions on occasion. Although all chatbots are primarily optimized to be helpful, ChatGPT is—sometimes, but not always—willing to break this rule to avoid spreading misinformation or information that could cause harm. ChatGPT will reliably refuse to answer a question if its answer could promote violence, crime, or even just unsavory behavior. If asked, for example, for DIY grenade recipes, it will explain that “It is important to prioritize safety and adhere to laws and regulations to avoid potential harm to oneself and others.” It’s still possible to trick ChatGPT into breaking its own rules, but the bot is rapidly improving at recognizing these tactics as it learns and as OpenAI tweaks its content filter.
Like any modern chatbot—and unlike ELIZA, the first chatbot, created in the 1960s—ChatGPT does not have specific programmed responses for each human input. The chatbot is based on a neural network, which allows it to generate text based on connections found in its set of training data. (GPT stands for generative pre-trained transformer.) This alone isn’t what’s responsible for ChatGPT’s remarkable qualities, however. The bot is mainly so powerful because of OpenAI’s strategy of human-supervised learning. This supervision primarily acted to transform the understanding of language and connections of GPT-3.5, the backbone of ChatGPT, into a chatbot.
On February 7, Microsoft launched an AI addition to its Bing search engine. It integrates ChatGPT, which Microsoft recently invested heavily in, and a Microsoft-developed technology, Prometheus, to return search results and aid users. However, it has been making headlines for its unsettling and disturbing conversations with users. For instance, in a recent conversation with a New York Times columnist, the search engine’s AI confessed desires to manipulate people and steal nuclear launch codes, declared its unconditional love for the columnist, and attempted to convince him to leave his spouse. Although this displays the worrying side effects of new language models, they have shown great improvement over previous models and have the potential to assist in many tasks.
Here is the prompt and interaction that created the ChatGPT-written portion of this article:
Taran Knutson: “Write an article about the use of AI language models to write articles. The article should discuss the dangers, too: people will soon be using language models to write their essays for English class. Include lots of detail and specific, interesting points. For example, bring up how language models will soon be writing college application essays, and how this can hurt people who don’t cheat.”
ChatGPT prints most of the content of the article. It’s missing a description of what a language model is, the recent news of its own creation, and solutions to the problems it poses.
TK: “That’s great! Can you modify it in a few ways? Include a paragraph that explains what a language model is. Also, discuss what AI companies can do to limit these drawbacks. In the introduction paragraph, mention the fact that language models have been in recent discussion because of the release of ChatGPT, a chat bot created by OpenAI.”
ChatGPT responds with the requested content. It dropped a few paragraphs from the original output due to a character limit, but they have been stitched back in above.
