Artificial intelligence is one of the most discussed topics in IT. AI also plays a major role in IT security, for example when it comes to malware detection. An American student has shown that with the GPT-3 AI technology a blog article can be written automatically, which can hardly be distinguished from a human written article.
GPT-3 so easily blog articles can be created
Computer science student Liam Poor from the University of California at Berkley says it took less than an hour from the idea to the release of the first blog article. During this time, the student has also contacted a PhD student who has access to the GPT-3 technology.
GPT-3 is a technology developed by the company OpenAI. The predecessor program, GPT-2, was already highly debated in February 2019 because it was held back by the company. The reason for this restraint was too much fear of misuse of the technology. This step of the company met with a lot of criticism, among other things the company was accused of just making a big show. As a result, GPT-2 was finally released in November 2019.
With GPT-3, OpenAI has taken a different approach by not hiding the technology, but also not publishing it. Only selected researchers were allowed to test the technology so that the product could be commercialized by the end of 2020.
Student Liam Poor applied for one of these research positions, but before he even received a response from the company regarding his application, he had already discovered a PhD student who had access to GPT-3. So the process between the first hearing of GPT-3 and the publication of the first article written by GPT-3 in less than a week took place!
The success of the AI contribution
After the student has dealt more intensively with the written texts by GPT-3, he noticed that the automated articles are much more unstructured and do not follow a common thread. Literally, the student said: “He can write quite well, but he is not very logical and rational”, exactly for this reason the student chose blog topics that do not have to be logically structured in terms of content.
The topic he chose fits the category: productivity and self-improvement. With a couple of headlines, such as “Courage and creativity are more important than intelligence”, the AI could start writing the blog article.
Besides numerous comments and subscribers, the automatically generated blog actually reached number 1 on the Hacker News website. According to Poor, this proves that a KI can already pass for a human author and thus cannot be unmasked by the broad masses.
Flood of blog posts due to GPT-3
Some experts fear that such artificial intelligence could lead to a tool for mass disinformation. But also heaps of spam posts or a flood of mediocre blogs could be facing us by such technologies. Such a flood could reduce the value of online contributions significantly, but for now it remains to be seen how OpenAI will handle the experiment and when GPT-3 will be released.
Sevencast – der IT-Security Podcast
Von unterwegs, im Büro oder zu Hause hören und auf dem aktuellen Stand bleiben!