About a year ago, OpenAI developed a watermarking system for text generated by ChatGPT and a tool that could detect it. However, there are serious discussions within the company about whether this system should be offered to everyone. On the one hand, this watermarking system is seen as a step that creates liability as it enables the detection of texts produced by artificial intelligence (AI), on the other hand, there is concern that this step may negatively affect the company’s profits.
Working principle of OpenAI watermarking system
The watermarking system creates a detectable pattern by adjusting the way the model predicts previous words and phrases. This technique enables the recognition of texts created by artificial intelligence without reducing the text quality. In a global survey, a majority of respondents expressed support for the AI text detection tool. However, nearly 30 percent of users said they would use ChatGPT less if watermarking was implemented.
OpenAI announced that the system is 99.9 percent effective and resistant to manipulation. Despite this, it was stated that malicious users can easily bypass this system by rearranging words with a different model. This raises questions about the effectiveness of the watermarking system.
The watermarking system is thought to be of great importance, especially for the education sector. Teachers have difficulty detecting the assignments prepared by students with artificial intelligence-supported tools. This system can help teachers evaluate students’ actual performance. In addition, it is seen as an important step towards academic honesty and originality.
However, OpenAI’s concerns include the negative perception of the usefulness of AI tools for non-native English users. It is stated that the watermarking system may cause stigma among these users and reduce the use of AI tools.
While some employees within OpenAI believe the watermarking system is effective, they recommend trying less controversial and as yet unproven alternative methods, considering user sensitivities. In a blog update published today, the company stated that they are in the early stages of embedding metadata. It is stated that there will be no false positives since this technique is cryptographically signed, but it is still unclear how successful it will be.
It remains unclear whether the watermarking system will eventually be made available and what role this system will play in the future of artificial intelligence-supported vehicles. How OpenAI will follow on this issue and the reactions of users will directly affect the decisions the company will make. Balancing technological innovations in terms of ethics and user experience is among the most important elements in this process.
The steps taken by OpenAI in the future may greatly shape the usage areas of artificial intelligence technologies and the trust in these technologies. This process will be critical not only for OpenAI, but also for the development of the artificial intelligence industry in general.
Source link: https://www.teknoblog.com/openai-filigranlama-sistemi-ic-tartismalara-yol-aciyor/