ChatGPT, the chatbot developed by OpenAI, failed to correctly answer a question during a live test by journalists at French news channel LCI, and admitted it had given its response too quickly without checking it carefully.
Asked how many days of the week contain the letter "d," the chatbot got the answer wrong. When pressed on the mistake, it explained that it had answered too fast and had not verified its response carefully enough, according to LCI.
It was not the first time the channel had caught the chatbot out. Nine months earlier, LCI journalists had put ChatGPT through a similar on-air test, which it also failed to answer correctly.
Warnings over AI risks
The test came days after a researcher who had previously worked at both OpenAI and Anthropic warned of the dangers artificial intelligence could pose to humanity. On Wednesday, September 9, he accused his two former employers of "playing with our lives."
A safety official at Anthropic has estimated the risk of AI wiping out humanity within the next decade at "10%." While the exact form such a scenario might take is difficult to picture, some threats already appear plausible. In the near future, AI could, for example, take full and autonomous control of critical infrastructure such as nuclear power plants or hospitals. Another troubling scenario involves AI piloting military drones, raising the prospect of systems able to make decisions without human intervention.
Two major AI developers
OpenAI, the maker of ChatGPT, and Anthropic, the company behind the Claude chatbot, are among the leading developers of advanced artificial intelligence. Both have employed researchers who have publicly raised concerns about the long-term risks of the technology they help build.
Despite the scale of the warnings being issued about AI's growing power, LCI's repeated tests suggest that even a widely used chatbot like ChatGPT can still trip up on straightforward questions.
