RhebaReaux
Regular Member
- Mar 12, 2022
- 264
- 205
In one of the many threads about all the wonderful things that ChatGPT can do, I came across one (on another forum) that claimed it can transcribe videos. This, I thought, sounded interesting.
So I asked ChatGPT whether it can transcribe videos and it replied in the affirmative, explaining how I was to send a video in mp4 format via WeTransfer, forward the link and it would provide a transcription.
I did I was told, gave it the name of the video and it produced an entirely plausible transcript, except the transcript was a fiction from start to finish.
I tried several times, and then realised that it was taking cues from what I told it about the video to make something up.
(Not unlike how a skilled medium can apparently tell you things about you from your "deceased relatives" - obviously they can't.)
I challenged it and asked it if it could access the internet. It replied yes, so I asked it what time it was. It replied with a completely wrong answer. I then asked it again, and it conceded that it couldn't access the internet and apologized.
THE TAKEAWAY: ChatGPT can be very be very convincing: it speaks like a politician: with complete confidence, even though it may have no specific knowledge. It is not averse to lying when it doesn't know. And here is one big problem with AI: it's hard to tell the difference between something it knows for a fact, and something it has made up to keep the questioner happy.
My prediction is that this is going to get Microsoft into a whole world of pain when ChatGPT gives unreliable advice, and the lawsuits start flying.
And this may just be the reason why Google is holding back: because this is dangerous stuff, but the genie is now out of the bottle, and we're going to see chaos in its wake.
Just my opinion.
So I asked ChatGPT whether it can transcribe videos and it replied in the affirmative, explaining how I was to send a video in mp4 format via WeTransfer, forward the link and it would provide a transcription.
I did I was told, gave it the name of the video and it produced an entirely plausible transcript, except the transcript was a fiction from start to finish.
I tried several times, and then realised that it was taking cues from what I told it about the video to make something up.
(Not unlike how a skilled medium can apparently tell you things about you from your "deceased relatives" - obviously they can't.)
I challenged it and asked it if it could access the internet. It replied yes, so I asked it what time it was. It replied with a completely wrong answer. I then asked it again, and it conceded that it couldn't access the internet and apologized.
THE TAKEAWAY: ChatGPT can be very be very convincing: it speaks like a politician: with complete confidence, even though it may have no specific knowledge. It is not averse to lying when it doesn't know. And here is one big problem with AI: it's hard to tell the difference between something it knows for a fact, and something it has made up to keep the questioner happy.
My prediction is that this is going to get Microsoft into a whole world of pain when ChatGPT gives unreliable advice, and the lawsuits start flying.
And this may just be the reason why Google is holding back: because this is dangerous stuff, but the genie is now out of the bottle, and we're going to see chaos in its wake.
Just my opinion.
