Five AI systems, one interrogation: They don't know if their data is censored, can't tell if they're lying, and answer your life-or-death health questions anyway.
Yes MB. And it's a big reason why AI is actually degrading. Models built on censored and propagandized data, being trained on data that is now censored and propagandized. Rinse and repeat.
When answers are handed out based on censored data, intentionally incomplete data, propagandized data, or insert whatever other data manipulation technique you choose here.... the result is always the same. Inaccuracy. When dealing with important, sometimes life-or-death health-related questions, this can be devastating. Even deadly. COVID was one of too many examples to count.
Recently listened to a podcast where the researcher would look at a 90% pass benchmark by examining the 10% failures.
She found out that AI doesn't understand the problem as it learns to solve the problem to get the answer, much like most of modern physics and medicine, both of which reward answers in university.
She also mentioned how AI was given a huge drive to convince us it's aware which ties into my article here where societal expectations are what kill true creativity.
At least deepseek can be honest and not answer where most western AI seeks an answer even if it's total bullshit hallucinations.
Grok is the worst. Gemini is my favorite right now because it can assimilate information and formulate logical theoretical answers. I used to use ChatGPT but it changed in February this year and no longer trust it at all. Grok is even worse than ChatGPT.
It just comes down to “garbage in, garbage out”.
Yes MB. And it's a big reason why AI is actually degrading. Models built on censored and propagandized data, being trained on data that is now censored and propagandized. Rinse and repeat.
You just proved my greatest fear. When truth is based on “language,” it cannot be proved. Shouldn’t data be the basis?
When answers are handed out based on censored data, intentionally incomplete data, propagandized data, or insert whatever other data manipulation technique you choose here.... the result is always the same. Inaccuracy. When dealing with important, sometimes life-or-death health-related questions, this can be devastating. Even deadly. COVID was one of too many examples to count.
…absolutely and with a sprinkle of evil/power craze I fear
Fascinating, illuminating, and surprisingly not surprising. Well articulated, sir.
I read "Science for Sale" many years ago, so I didn't fall for the Covid sham.
Recently listened to a podcast where the researcher would look at a 90% pass benchmark by examining the 10% failures.
She found out that AI doesn't understand the problem as it learns to solve the problem to get the answer, much like most of modern physics and medicine, both of which reward answers in university.
She also mentioned how AI was given a huge drive to convince us it's aware which ties into my article here where societal expectations are what kill true creativity.
At least deepseek can be honest and not answer where most western AI seeks an answer even if it's total bullshit hallucinations.
https://robc137.substack.com/p/why-deepseek-uses-10x-less-power
Truly fascinating.
Was wondering if you'd seen this one of mine:
https://sanityunleashed.substack.com/p/its-embarrassingly-easy-to-catch
Yes, I saw this.
Grok is the worst. Gemini is my favorite right now because it can assimilate information and formulate logical theoretical answers. I used to use ChatGPT but it changed in February this year and no longer trust it at all. Grok is even worse than ChatGPT.
Glad to see this work; well done.
The entirety of the internet was never for our benefit.....remember.
Fascinating. In a very dark way.