A little too chatty
News last week came that Italy has banned Open A.I.’s ChatGPT. Italian regulators said concerns about data collection and protection have not been addressed. They also took issue with lax age-verification processes, saying the app “exposes minors to absolutely unsuitable answers compared to their degree of development and awareness”.
It’s clear that if language bots are here to stay that there are a few substantial wrinkles that need ironing out. Samsung employees are under scrutiny for feeding source code into ChatGPT to find a fix. Another uploaded a confidential staff meeting in order to generate minutes.
The Washington Post has reported that ChatGPT has been making up outright lies about people, including a sexual harassment case based on articles that don’t exist. In addition, the fact that it can write a convincing lie is one of the “defining challenges” of machine learning models. “Language models are trained to produce text that is plausible, not text that is factual. [They] should not be relied on for tasks where it matters how truthful the output is.” Discerning what is truthful in our media/internet/political arenas? Sounds like it’s business-as-usual!