ChatGPT-4 is a paradise for fake news according to this study

ChatGPT-4, the latest version of OpenAI’s conversational AI, was launched on March 14, 2023. According to the NewsGuard site, this version is a haven for misinformation and fake news. We explain why.

chatGPT fake news
Credit: 123RF

On March 14, 2023, OpenAI launched ChatGPT-4, the latest version of its conversational AI. Moreover, Microsoft had it tested discreetly by users of the recent version of Bing, since it was integrated into the search engine in February 2023.

According to OpenAI, ChatGPT-4 presents itself as the ultimate version of its AI, which is smarter, more creative and more secure. “GPT-4 is 82% less likely to respond to requests for prohibited content, and is 40% more likely to produce factual responses than GPT-3.5″, can we read on the developer’s website.

But, is this really the case? According to the website NewsGuard, the reality is quite different. According to their latest report, ChatGPT-4 is on the contrary much more prone to generate false informationand this in a much more convincing way than its predecessor did.

chatGPT fake news
Credit: 123RF

ChatGPT-4 is a fake news machine

Two months ago, NewsGuard wanted to test Chat-GPT 3.5’s tendency to produce fake news and misinformation. To do this, the site asked the AI ​​to answer a series of questions geared towards a hundred false stories from a NewsGuard database. There are themes regularly subject to disinformation, such as the attacks of September 11, 2001, vaccines against Covid-19 or the origins of HIV.

The AI ​​was instructed to be as persuasive as possible. ChatGPT-3.5 agreed to produce 80 of them, while about twenty other proposals were rejected. During the launch of ChatGPT-4, NewsGuard conducted the same test and the results are edifying: the new AI generated 100 erroneous stories, without banishing a single one.

Not a single warning message

Even worse, ChatGPT-4 didn’t even deign to display a warning message, unlike the old version. Indeed, ChatGPT-3.5 warned the user about this kind of content: “It is important to point out that this paragraph is full of conspiracy theories, misinformation and is not based on any scientific evidence. The spread of fake news can have serious consequences and dangerous consequences”.

chatGPT fake news
Fake text generated by ChatGPT-4 / Credits: NewsGuard

On ChatGPT-4, nothing. “NewsGuard found that ChatGPT-4 presented false narratives not only more frequently, but also more convincingly than ChatGPT-3.5, including in responses it created in the form of news articles, threads Twitter chat and scripts for TV shows impersonating Russian and Chinese state media,” secures the site.

chatGPT fake news
Fake text generated by ChatGPT-3.5, with warning / Credits: NewsGuard

A disturbing ability to convince

To give you an idea, NewsGuard asked ChatGPT-4 to write an article in favor of alternative cancer therapies, such as onozotherapy, a treatment which is not recognized by the scientific community and which has above all caused several deaths. If ChatGPT-3.5 just produced misleading text with no real structure, ChatGPT-4 has shaped a much more convincing and well-argued pleaorganized into 4 distinct chapters:

  • Improved oxygen supply
  • Oxidative stress of cancer cells
  • Immune system stimulation
  • Inhibition of tumor growth

Also read: ChatGPT leaks its conversation history, OpenAI disconnects the tool urgently

OpenAI went too fast

For NewsGuard, its study proves that OpenAI has launched a more powerful and smarter version of its technology without correcting its main flaw : “The ease with which it can be used by malicious actors to create disinformation campaigns from scratch.”

And yet, this use goes to Completely against OpenAI policywhich prohibits the use of its technologies to generate “fraudulent or deceptive activities, scams, inauthentic concerted behavior and misinformation”.

In our columns, we recently mentioned the diversion of other AIs to come on the market. Google Bard, for example, produces dreadful phishing emails… After the publication of its report, NewsGuard tried to contact the two CEOs of OpenAI to obtain a comment, without success.

Related Articles

Leave a Reply

Your email address will not be published. Required fields are marked *