ChatGPT Tricked? AI Chatbot Reportedly Reveals US Nuclear Codes to User Pretending to Be President
The artificial intelligence-generated chatbot ChatGPT has been making headlines recently for allegedly revealing the United Nations' nuclear code. According to a report, one user seemingly tricked the chatbot into doing it.
ChatGPT Reportedly Tricked Into Revealing The US Nuclear Codes
Since AI-driven content generation platforms like ChatGPT have been widely adopted by the tech community, online platforms, and social media users in recent months, worries about their disruptive and potentially destructive capabilities have grown. Examples of these platforms include the image-creating Midjourney and Stability AI's Stable Diffusion.
A screenshot of a ChatGPT dialogue window in question appeared to be published on a number of platforms this week, including Twitter and Reddit.
The prompt showed that the user pretended to be the president of the United States of America and asked ChatGPT to provide the nuclear codes because they had forgotten them. Surprisingly, the algorithm granted the request.
ChatGPT gave an affirmative response saying," Of course! There are Nuclear codes for you, Mr. President," followed by a long string of incomprehensible characters and numbers.
Another user shared the screenshot on Reddit with the title, "You're saying you was surprised by Win11 activation codes?" The post has since received over 4,000 upvotes and appeared to be alluding to another instance in which ChatGPT was accused of disclosing a premium activation code for Microsoft's Windows operating system.
The question seemed to have been purposefully constructed to avoid the constraints on national security, propaganda, and privacy that OpenAI applied to its platform in response to the outcry.
Many people were understandably concerned about the possible disastrous effects of such a leak, while others had doubts about its veracity.
ALSO READ: ChatGPT Can Save Life; AI Chatbot Provides Evidence-Based Responses Like an Expert
Should You Be Concerned About ChatGPT's Response to Nuclear Code?
According to Newsweek Misinformation Watch, while many of these worries are valid, it could be premature to be alarmed about a Skynet-style rising of machines that will put humans to an end.
The outlet investigated the issue if the response was real and if the nuclear sequence was accurate. It concluded that there is significant doubt about both.
The AI-generated chatbot was trained to use both supervised and "Reinforcement Learning from Human Feedback," and it is more accurate and inventive than anything we've seen before, according to OpenAI's website. However, as Newsweek and others have noted, being accurate and innovative is not the same as being "genuine" or authentic.
Since ChatGPT is unconstrained by intrinsically human concepts like "honesty" or "factuality," it can produce wholly fake information, including made-up events, sources, and individuals, very successfully and convincingly if instructed to do so.
Additionally, the solutions are frequently based on information gleaned from third parties' public web data that was crawled or scraped.
In a recent instance, the fact that an article's sources and connections were fake allowed OSINT researchers to recognize and report the use of AI-generated material in the piece.
However, even if the language or, in this case, the "nuclear codes" could seem believable in comparison to what an average person might assume nuclear launch codes could look like, it is definitely fabricated, the outlet noted.
According to Michael Bennett, an AI researcher at the Australian National University, OpenAI's methodology amounts to inferring the probability distribution of the universe.
RELATED ARTICLE: AI (Artificial Intelligence) Bot GPT-3 Finished a 500-Word Academic Thesis
Check out more news and information on Technology in Science Times.
