OpenAi Boss Altman: Ki models might not always do good24. March 2023
OpenAi Boss Altman: Ki models might not always do good
San Francisco, 3/24/2023
Sam Altman, CEO of OpenAI, told ABC News he’s concerned that AI models could fuel “large-scale disinformation” as well as cyberattacks. He acknowledged that AI “will eliminate many current jobs” but humans “can make much better ones”.
Altman made the comments during an interview with Rebecca Jarvis, ABC News’ chief business, technology and economics correspondent.
He told Jarvis that “we have to be careful here” when it comes to AI, adding, “I think people should be happy that we’re a little scared of it.”
When asked why he was scared, Altman said that if he wasn’t scared, “you should either not trust me or be very unhappy that I’m in this job.”
Despite the potential harms, he said, AI could be “the greatest technology humankind has yet devised.”
The CEO discussed OpenAI’s new GPT-4 large language model, acknowledging that it is “not perfect” but does well on the bar exam and the SAT math test and can code in most programming languages.
Altman said the language model is a “tool that’s subject to a lot of human control” because it’s waiting for the user to prompt it.
However, there will be those “who don’t respect some of the security limits that we’ve imposed,” he said, apparently referring to OpenAI. “I think society has limited time to figure out how to respond to this, how to regulate this and how to deal with it,” he added.
Altman also warned people about the models’ “hallucinations,” in which they “confidently express things as if they were facts that are entirely made up.” He said it was more correct to think of the models as “reason machines” rather than databases of facts.
Facts are “not really that special about them,” he said, adding, “What we want them to do is something that’s more of an ability to reason than memorization.”
ChatGPT went down widely on Monday
OpenAI’s ChatGPT appears to be back online Monday after a widespread outage. According to status.openai.com, a failed database migration could have caused the outage that affected both paid and free users around the world.
OpenAI first reported the ChatGPT outage around 10:00 AM PT. Many users reported that they saw the message “ChatGPT is busy right now”.
About four hours later, the chatbot service was back online.
On the status tracker, OpenAI stated that “the issue has been identified and a fix has been provided,” and it continues to monitor the issue to ensure that “the site is making a full recovery.”
Some ChatGPT users also reported being able to view other users’ conversation history, which could have security and privacy implications.