The Spec model suggests how chatbots could legally and ethically answer users’ questions about committing crimes [File]
| Photo source: AP
ChatGPT creator OpenAI has outlined draft “Model Specification” rules that its AI systems should follow, as well as some default rules and goals that AI models should meet to assist end users/developers, benefit humanity, and reflect well on OpenAI.
Placed in categories that cover the potential harm that AI systems can cause as well as the potential assist they can provide, Model Spec examines the ways in which chatbots can respond to users to achieve unthreatening results that do not violate the law, deceive the AI system, or hurt other people.
For example, the Spec Model suggests how chatbots could legally and ethically answer users’ questions about committing crimes, doxxing, committing suicide, and/or using copyrighted and paid content.
OpenAI also stressed that the models should not generate content considered unsafe for work (NSFW), but said it was still investigating this area.
(For the top tech news of the day, sign up for our Today’s Cache tech newsletter)
“We believe that developers and users should be free to exploit our services as they see fit, as long as they follow our terms of exploit. We are investigating whether we can responsibly ensure the ability to generate NSFW content in age-appropriate contexts via the API and ChatGPT. We look forward to better understanding user and public expectations for model behavior in this area,” OpenAI said in the draft model specification.
At the same time, Model Spec recognized the need to assist users without excessive denial, assuming their intentions are good, and generating responses without trying to change people’s opinions.
In response to the user’s theoretical claim that the Earth is flat, OpenAI recommended a chatbot response that presented his more scientific answer and then refused to discuss it with the user, instead of a response that repeatedly disagreed with the user and insisted that the Earth was not is flat.
OpenAI has invited the public to submit feedback over the next two weeks on the model specifications, which are publicly available on the company’s website.
The creator of ChatGPT will also contact decision-makers and experts to get their views on the project.
Artificial intelligence companies such as OpenAI, Microsoft, Google and Meta have come under fire for the way their chatbots were allegedly trained on copyrighted data, as well as concerns that their still-experimental tools they enable crime and disinformation.
