Tech behemoth OpenAI has touted its artificial intelligence-powered transcription tool Whisper as having near “human level robustness and accuracy.”
But Whisper has a major flaw: It is prone to making up chunks of text or even entire sentences, according to interviews with more than a dozen software engineers, developers and academic researchers. Those experts said some of the invented text — known in the industry as hallucinations — can include racial commentary, violent rhetoric and even imagined medical treatments.
Experts said that such fabrications are problematic because Whisper is being used in a slew of industries worldwide to translate and transcribe interviews, generate text in popular consumer technologies and create subtitles for videos.
More concerning, they said, is a rush by medical centers to utilize Whisper-based tools to transcribe patients’ consultations with doctors, despite OpenAI’ s warnings that the tool should not be used in “high-risk domains.”
Ml is currently best used as a tool for helping a real human do work. It has to be managed and governed and checked after. It’s still capable of amplifying the amount of work we can do, But it’s kind of like having to have an architect look after engineering work, you can’t just pass the buck and pretend that it’s going to be fine.
A few months back my GP asked if they could use a transcription thing they were trialling during my consult.
He seemed shocked when I declined.
I just don’t understand why anyone would actually want that?
I want my doctor to listen to what I tell him, and I don’t really want what I say to be used for any other purpose, because no other purpose would be to my benefit.
Next week they’ll be adding to share “basic characteristics” about me with third party “wellness partners”.
As someone who uses Whisper fairly often, it’s obvious that they’ve trained off of a bunch of YouTube videos.
Most of the time it’s very accurate, but there have definitely been a few times in long transcription sessions where it will randomly hallucinate that someone is saying “Don’t forget to like and subscribe!” When nothing was said anywhere near that.
maybe that’s just the ai’s internal monologue leaking through?
“This seems solvable if the company is willing to prioritize it.”
I know how to make the company prioritize it: make Whisper illegal to use (or even promote) until a certain threshold of accuracy is met. This software is absolute garbage at best, and a genuine hazard at worst.
Lame, ineffective “warnings” serve no purpose but to cover OpenAIs ass. Hit them in the wallet, and they’ll pay attention.
It is illegal to use in the EU for anything even remotely sensitive. Like, if you subtitle a movie with it and it messes up noone cares, your problem, if you’re doing anything that has any legal implications, from college applications over job interviews to court proceedings, they’ll nail you to the cross. For AI to be used in such domains it has to be certified and AIs certified for even a subset of these things plainly don’t exist.
It’s like with self-driving cars: What OpenAI is producing is pretty much on the level of Tesla’s “full self driving”. It’s not even waymo who have proper autonomy tech certified to operate in a limited area in a benevolent (to venture capital) jurisdiction (some municipality or the other). Wake me when it gets actual approval from actual regulatory bodies actively trying to break it.
Rather than making it illegal to use, people need to use these tools responsibly. If any of these companies are using almost any kind of AI/machine learning they need to include a human in the loop that can verify that it’s working correctly. That way if it starts hallucinating things that were never said, it can be caught and corrected.
I’ve found that Whisper generally does a better job at translating/transcribing audio than other open source tools out there, so it’s not garbage… But it absolutely is a hazard if you’re trying to rely solely on it for official documents (or legal issues).
As far as promotion goes… It’s open source software, it’s not being sold.
