ChatGPT Still Spreads Falsehoods, Says EU Data Watchdog
The post ChatGPT Still Spreads Falsehoods, Says EU Data Watchdog appeared on BitcoinEthereumNews.com.
ChatGPT, the viral chatbot from OpenAI, still falls short of the European Union’s data accuracy standards, according to a new report by the EU’s privacy watchdog. Also read: Europe’s AI Act Gets Final Approval With Up To $38M Fines The European Union Data Protection Board’s (EDPB) report is not a legal document. However, it will inform a common approach for regulating ChatGPT across the 27 member states of the EU. ChatGPT Fails to Comply With Data Standards National watchdogs formed the ‘ChatGPT task force’ last year under the stimulus of ongoing cases in several European countries against the AI model. The task force falls under the EDPB. In a report published May 24, the task force said: “Although the measures taken to comply with the transparency principle are beneficial to avoid misinterpretation of the output of ChatGPT, they are not sufficient to comply with the data accuracy principle.” Large language models, or LLMs, are notorious for “hallucinating” – tech speak for when AI chatbots spew out falsehoods, often with confidence. Chatbots like ChatGPT or Google’s Gemini are powered by LLMs. Under Europe’s tough General Data Protection Regulation (GDPR), users can sue for being mispresented or denied the opportunity to correct inaccurate information about themselves. According to the task force’s report, “Due to the probabilistic nature of the (AI) system, the current training approach leads to a model which may also produce biased or made-up outputs.” “The outputs provided by ChatGPT are likely to be taken as factually accurate by end users, including information relating to individuals, regardless of their actual accuracy,” it added. European Union flag. Source: Pixabay EU Doubles Down on Compliance OpenAI has previously cited technical complexity for its failure to correct misinformation, but the European Union watchdog is doubling down. “In particular, technical impossibility cannot be invoked…
Filed under: News - @ May 25, 2024 8:14 pm