The creators of ChatGPT have launched a tool now used by 40,000 doctors, but they are warning about potential errors it could make.

OpenAI has announced that its Whisper model should be used with caution due to possible errors it may generate.

Artificial intelligence is currently dominating the tech landscape, drawing significant attention. Its impact is such that recently, Adobe announced a change in how it produces AI-generated art, while Apple is offering up to one million euros to anyone who can identify vulnerabilities in its AI technology. However, the news about Whisper, OpenAI’s transcription tool, is not as encouraging.

According to the portal The Verge, OpenAI has warned about the risks of using Whisper in “high-risk settings” like healthcare, as incorrect transcriptions can have serious consequences. Despite these warnings, various medical institutions have adopted Whisper through a tool from Nabla, reaching over 30,000 doctors and 40 healthcare systems. Given the scale of these figures, the outlook is concerning.

Privacy issues and high error frequency

Following its privacy policy, Nabla deletes the original audio files, preventing any comparison between the AI-generated transcriptions and the original recordings. This has raised security concerns, as some parents have refused to share audio from their children’s medical consultations with tech companies. Moreover, researchers at the University of Michigan found errors in 80% of the transcriptions of public meetings, further intensifying concerns about the accuracy of these tools in healthcare.

By: Nestor Castillo, ForAllTeach Director