ChatGPT outputs too much false information to comply with the rules

OpenAI doesn’t appear to have achieved much progress since 2023, although the assessment is still in progress. The European Data Protection Board (EDPB) just published the first “GPT task force” status report, and ChatGPT isn’t doing well.

Although OpenAI made an effort to bring ChatGPT, its central AI model, into compliance with EU laws, including the General Data Protection Regulation (GDPR), the EDPB ultimately found that these efforts were insufficient.

“The measures taken to comply with the transparency principle are beneficial to avoid misinterpretation of the ChatGPT output, but they are not sufficient to comply,” states the EDPB document. The results correspond with OpenAI’s experience handling temporary stop orders from many European member states over most of 2024.

ChatGPT and OpenAI persisted in violating EU and Italian data privacy legislation according to the cointelegraph. This data given by the Italian Data Protection Office. In the interim, OpenAI hasn’t acted swiftly enough to put ChatGPT into line with EU legislation, according to the EDPB investigation.

It seems that the main grievance is that ChatGPT provides false information a lot of the time. According to the EDPB, “In actuality, the current training approach leads to a model that may also produce biased or made-up outputs due to the probabilistic nature of the system.”

OpenAI’s Challenge with ChatGPT

The report also highlights the EDPB’s concern that “ChatGPT’s outputs might be perceived as truthful by users,” regardless of their actual correctness.OpenAI will probably fail to be able to ensure ChatGPT compliance. It is not possible for humans to properly assess the accuracy of the dataset under GDPR due to the billions of data points and trillions of parameters in the GPT-4 model.OpenAI will not pleased with the EDPB’s clear statement that “technical impossibility cannot used as an excuse for not complying with these requirements.”

The results of the EDPB’s analysis point to serious flaws in the current implementation, especially about output accuracy and transparency. These are concerns that OpenAI needs to fix quickly and effectively, since technical difficulties are not an excuse for refusal. OpenAI is under pressure to come up with creative ways to comply with GDPR and win back the confidence of users and European regulators.