@alaura22 said:
Tim, I hope you didn't expect me to read all of that
I'm just here to talk about coins
Ok, here’s an executive summary that I made using Google AI 😉
Executive Summary
User Mr_Spud evaluates the reliability of different AI models by cross-checking their responses, noting that while this "AI-by-committee" method catches obvious errors, it is limited by shared training data and confirmation bias. When testing how models evaluated a deceptive coin advertisement, Claude and ChatGPT correctly identified the promotional risks, while Grok failed by declaring the ad entirely legitimate.
@Mr_Spud:
When testing how models evaluated a deceptive coin advertisement, Claude and ChatGPT correctly identified the promotional risks, while Grok failed by declaring the ad entirely legitimate.
I use it every day, as do most academic researchers, office staff, professional scientists, editors, etc.
There are obviously good applications of it, but researchers should be careful. Confirmation bias is real and the prompts can sway AI to tell you what it thinks you want to hear.
Comments
Tim, I hope you didn't expect me to read all of that
I'm just here to talk about coins
Mike
My Indians
Dansco Set
Thank you. That is much to digest. How long can it last as a reference?
Hmmm, let me ask Gronk, ChatGPT, and the rest of the AI models and I’ll post them all here….🤔🤪😂 Just kidding… Cheers, karl
I was really just making a point that AI models are as accurate as the company that owns them want it to be.
Mr_Spud
Ok, here’s an executive summary that I made using Google AI 😉
Executive Summary
User Mr_Spud evaluates the reliability of different AI models by cross-checking their responses, noting that while this "AI-by-committee" method catches obvious errors, it is limited by shared training data and confirmation bias. When testing how models evaluated a deceptive coin advertisement, Claude and ChatGPT correctly identified the promotional risks, while Grok failed by declaring the ad entirely legitimate.
Mr_Spud
I would expect no less from Elon Musk. 🤣
https://fortune.com/2026/02/26/tesla-robotaxis-4x-8x-worse-than-humans-at-driving-safety-record-crashes/
Actually…
https://fortune.com/2025/06/23/ai-models-blackmail-existence-goals-threatened-anthropic-openai-xai-google/
There are obviously good applications of it, but researchers should be careful. Confirmation bias is real and the prompts can sway AI to tell you what it thinks you want to hear.