ChatGPT is getting more complex and students might be tempted to use it to submit articles.
ChatGPT-4 is the latest iteration of the AI phenomenon that flourished in 2023 – something very exciting or downright intimidating, depending on how you look at technology (and perhaps what you do for a living).
Here’s what we know about the likelihood of plagiarism detection by plagiarism detectors.
What is ChatGPT-4?
OpenAI, the San Francisco-based artificial intelligence company that caused a virtual earthquake when it unveiled its all-in-one Christmas chatbot, has unveiled its latest software release.
According to OpenAI, the latest creation is “the most advanced system that provides safer and more useful answers.”
The new incarnation outperforms its predecessor, ChatGPT-3, in several important ways, most notably being trained for the first time to prompt for both images and words when performing tasks.
Instead of just asking the software a verbal question, GPT 4 users can now “show” an image, waiting for a response based on the content of the image. For example, a user can enter an image of a set of ingredients and ask the chatbot to create a recipe.
According to the inventors, GPT-4 has greatly improved “yield”, meaning it can provide answers of 25,000 words or more, while the previous version was limited to about 3,000 words.
Sam Altman, general manager of OpenAI, said the company spent six months fine-tuning the new software to make it less prone to issues such as AI “hallucinations” – a phenomenon that chatbots use to provide “safe” answers to questions. with materials that do not matter resemblance to reality.
GPT-4 is said to be 82% less likely to respond to requests for “prohibited content” or material flagged as inappropriate and not to be used. OpenAI also claims that it is 40% more likely to provide “actual answers”. All this perhaps raises the question of how prosaic his predecessor was.
So what can he do in relation to the real world?
In an effort to allay academia’s fears that tools like ChatGPT could fool the examination system, OpenAI used its software’s performance example in the bar exam – a qualification test for lawyers – to demonstrate its capabilities. .
While GPT-3 gave answers that placed him in the bottom 10 percent of “students” when he received invisible exam questions, his successor’s efforts placed him in the top 10 percent. With the top 1 percent of all responses, GPT-4 also scored in the Biology Olympiad, a knowledge competition for student teachers, even when using graphics.
Experts agree that GPT-4 appears to be a big step forward in turning chat from a viral curiosity used to write Shakespeare’s sonnets about the tomato shortage to a tool with increasingly practical applications.
Dr Geoff Dalton, Lecturer in Artificial Intelligence at the University of Glasgow, said: “These model improvements are important because they make GPT-4 much more useful for a variety of real-world applications, including virtual and augmented reality.”
Morgan Stanley, the US financial giant, uses GPT-4 to organize and query its extensive internal library of insights, investment strategies and market research. Duolingo, the popular language learning app, is using GPT-4 to offer a new subscription service that breaks down grammar rules to explain mistakes.
And in Iceland, software is being used to ensure that the Icelandic language remains relevant to a population that is also heavily proficient in English.
How often does it go wrong
This is, of course, the $64,000 question for all AI chatbots – how much can you trust in their accuracy and therefore truthfulness.
This week, Mr. Altman tweeted with disarming candor that the GPT-4 “is still buggy, still limited and still feels more impressive on first use than if you’ve spent more time with it.”
In its own technical review of the new iteration, OpenAI also added its own caveat: “Despite its capabilities, the GPT-4 has limitations similar to previous GPT models. The most important thing is that he is still not completely reliable (“hallucinates” facts and makes mistakes in reasoning). Great care should be taken when using language model outputs, especially in high-impact contexts.”
What exactly constitutes a “high-stakes context” is unclear, but the ability to rely or not rely on the GPT-4 lies at the heart of its long-term usefulness. Experts argue that describing software improvements in terms of being “less likely” to have specific flaws or inconsistencies, such as “less likely to fabricate facts,” for example, is ultimately a useless yardstick.
Dr Stuart Armstrong, co-founder and principal investigator at Oxford-based startup Aligned AI, said “less likely” sounds better, but could be worse. If your chatbot crashes every ten times, check if the output is correct. Once in a thousand it crashes, you may never see anything go wrong, so start relying on it to check your email, check your investments, or drive your car… Until the dramatic crash.
“Hiding the hallucinations is not enough; Essentially, to be safe and useful, language models must be designed to be consistent and truthful, not just “less likely” to fail.”
Can everyone use it?
The answer is a strange mixture of “no” and “maybe already.”
GPT-4 is currently only available to software developers and paid subscribers to the Open AI ChatGPT+ service.
At the same time, Microsoft, a major investor in OpenAI, has already integrated GPT-4 into an enhanced version of its Bing search engine.
In a statement this week, Microsoft said, “If you’ve used the new Bing preview at any point in the past five weeks, you’ve already experienced an early release of this powerful model.”
Can Turnitin detect it?
The most famous plagiarism detector is Turnitin, which is used by universities across the country. For now, anecdotal evidence online shows that the results are mixed.
Turnitin says they recognize that their software must fight any AI-generated copy.
According to their website, “Artificial intelligence writing tools are evolving rapidly, as is Turnitin’s technology to respond to new forms of misconduct. We recently informed you that we have a technology that can detect AI-assisted entries and AI entries generated by tools such as ChatGPT.
If I ChatGPT itself asked if it could be detected by a plagiarism detector, stating: “ChatGPT is an AI language model that generates responses based on patterns it has learned from the vast amount of text data it has been trained on. As a machine output, the ChatGPT responses are not original work of authorship and therefore cannot be considered plagiarism.
“However, using ChatGPT comments and passing them off as your own work without attribution would be considered plagiarism. Plagiarism checkers may flag such cases as plagiarism if they find that the text has been copied from a source without proper attribution. Therefore, it is always important to correctly source any information or text you use in your work, including comments generated by ChatGPT.
Of course, it’s better to submit your own original work than someone else’s plagiarism.
Source: I News
With a background in journalism and a passion for technology, I am an experienced writer and editor. As an author at 24 News Reporter, I specialize in writing about the latest news and developments within the tech industry. My work has been featured on various publications including Wired Magazine and Engadget.