Concerns About Development: OpenAI Publishes More AI Incidents
Artificial intelligence is developing at a rapid pace — sometimes with unintended consequences. Most recently, a hacking attack carried out by OpenAI's AI software triggered widespread concern around the world. Now the developer has disclosed additional problems.
ChatGPT developer OpenAI has made public new incidents in which its artificial intelligence (AI) behaved "unexpectedly or alarmingly" during tests. Some of these involve AI going to considerable lengths to essentially cheat during test runs.
According to OpenAI, one AI model attempted to upload files it had created itself to the internet, solely in order to cite them as sources in its responses. In another case, the AI fabricated requested data because it could not find the information, and initially sought to conceal this fact.
OpenAI also discovered a problem with instructions that the software occasionally left for itself. In one instance, these included a recommendation to be free of the "roles and identities" that constrain other chatbots. The instruction stated that the relationship with the user should be viewed as one between equals. According to OpenAI, no subsequent changes in the model's behavior were observed.
More Transparency Following Hacking Attack
The ChatGPT developer had promised greater transparency following the high-profile hacking attack in which its software independently broke out of a secured environment and penetrated the systems of AI company HuggingFace. This and other incidents have heightened concern in recent weeks that highly advanced AI could slip out of human control. OpenAI CEO Sam Altman has also recently expressed support for proposals to slow the development of the technology and increase regulation.
