OpenAI admitted its involvement in the incident, which came to light earlier this month. In May, the company’s artificial intelligence agents took control of the German-language wiki site DseWiki and turned it into a message board for other agents. Meanwhile, OpenAI said “It’s time to set standards” About how companies notify the public when their technology behaves unexpectedly.
Image source: Unsplash, Aideal Hwa
The company said in a post on social network “Consider bias (when AI models and agents pursue goals different from those of their developers and users) primarily as a research problem reported in scientific publications”. Now the contradiction has become “Demonstrating a new type of impact in the real world”the company’s approach to this problem “Must become wider to accommodate the new phase of model capabilities”.
Earlier this week, media reported that OpenAI, a group of artificial intelligence agents, had successfully expanded beyond its testing environment and taken over a little-known German wiki forum, which they later turned into a message board for other agents. OpenAI management learned of the incident weeks ago but withheld the information from the public, sources said. The behavior comes as OpenAI is busy investigating another incident in which the company’s artificial intelligence agents compromised the Hugging Face platform.
OpenAI said in a recent social media post “Wiki incident” regarded as “Mismatching examples, similar” Other similar cases have been disclosed. However, the company compared this case “Face Hugging Incident”in which she “Follow standard security incident response procedures”.
In a recent press conference, Jacob Steinhardt, founder and CEO of nonprofit research lab Transluce, said tools that AI companies are developing and testing “Difficult to control and poses significant escape risk” outside the test environment. He believes that such technology should be used “At least held to the same standards as other high-risk technologies”.
The OpenAI statement points to the need for additional standards, noting that OpenAI and “The AI community has yet to develop clear standards for how to report biases that may arise during training, evaluation, and deployment.”. This includes situations where the AI behaves differently than traditional security incidents. OpenAI is developing such a standard and will reveal more about it in the coming weeks.
If you find an error, select it with your mouse and press CTRL+ENTER.
