Submission + - AskSlashdot: What would it require for you to trust an AI? (win.tue.nl) 2
shanen writes: Do you trust AI in general? What about DeepSeek in particular? Can anyone even explain what "trust" means these days? For example, I bet most of you trust Amazon and Amazon's secret AIs more than you should...
Recently I've been dabblling with a number of the GAI websites. Of course if I was a serious scientist then I'd have my own hardware and be running local tests. Or at least I'd be rooting around arXiv to see what the serious scientists are saying... All I have (as usual) are too many questions and too few jokes. Especially funny jokes. I wouldn't recognize a funny AI joke until after the AI had pwned me to pieces...
Actually the funniest joke I can think of on this topic involves DeepSeek. One of my experimental conversations on that website involved the topic of trust. DeepSeek turned out to be extremely good at explaining why I should not trust it. Every computer security problem I ever thought of or heard about and some more besides. "So what's that got to do with the price of tea in China?"
It's like the accountant who gets asked what 2 plus 2 is. After locking the doors and shading all the windows, the accountant whispers in your ear: "What do you want it to be?" And the price of tea in China is whatever Xi wants it to be! I bet you assumed the accountant was male, right? So the next questions are whether DeepSeek can do accounting or windows?
So let me start with some questions about DeepSeek in particular:
Have you run it locally and compared the responses with the website's responses? My hypothesis is that your mileage should differ...
It's well established that DeepSeek doesn't want to talk about many "political" topics. Is that based on a distorted model of the world? Or is the censorship implemented in the query interface after the model was trained? My hypothesis is that it must have been trained with lots of data because the cost of removing all of the bad stuff would have been prohibitive. Defining "bad" doesn't matter because I bet everyone agrees the Internet is chock full of bad data these years. Unless perhaps another AI filtered the data first?
What does trust mean? What sort of responses am I hoping to see? How many people still using today's Slashdot have even heard of "Reflections on Trusting Trust"? How many of the identities on today's Slashdot are just (AI-driven?) sock puppets? (Speculations on a mutual timeline building tool to verify childhood friendships?)
In closing, if you asked an AI to analyze all of the conversations on Slashdot, what sort of changes would it show over the years? My hypothesis on this question is that the interactions based on books will trend down. Maybe that's my selective memory, but I think a "sound" analysis would should a monotonic decrease. Largely based on The Anxious Generation by Jonathan Haidt I think the downward gradient might peak after smartphones became widely adopted circa 2010...
Recently I've been dabblling with a number of the GAI websites. Of course if I was a serious scientist then I'd have my own hardware and be running local tests. Or at least I'd be rooting around arXiv to see what the serious scientists are saying... All I have (as usual) are too many questions and too few jokes. Especially funny jokes. I wouldn't recognize a funny AI joke until after the AI had pwned me to pieces...
Actually the funniest joke I can think of on this topic involves DeepSeek. One of my experimental conversations on that website involved the topic of trust. DeepSeek turned out to be extremely good at explaining why I should not trust it. Every computer security problem I ever thought of or heard about and some more besides. "So what's that got to do with the price of tea in China?"
It's like the accountant who gets asked what 2 plus 2 is. After locking the doors and shading all the windows, the accountant whispers in your ear: "What do you want it to be?" And the price of tea in China is whatever Xi wants it to be! I bet you assumed the accountant was male, right? So the next questions are whether DeepSeek can do accounting or windows?
So let me start with some questions about DeepSeek in particular:
Have you run it locally and compared the responses with the website's responses? My hypothesis is that your mileage should differ...
It's well established that DeepSeek doesn't want to talk about many "political" topics. Is that based on a distorted model of the world? Or is the censorship implemented in the query interface after the model was trained? My hypothesis is that it must have been trained with lots of data because the cost of removing all of the bad stuff would have been prohibitive. Defining "bad" doesn't matter because I bet everyone agrees the Internet is chock full of bad data these years. Unless perhaps another AI filtered the data first?
What does trust mean? What sort of responses am I hoping to see? How many people still using today's Slashdot have even heard of "Reflections on Trusting Trust"? How many of the identities on today's Slashdot are just (AI-driven?) sock puppets? (Speculations on a mutual timeline building tool to verify childhood friendships?)
In closing, if you asked an AI to analyze all of the conversations on Slashdot, what sort of changes would it show over the years? My hypothesis on this question is that the interactions based on books will trend down. Maybe that's my selective memory, but I think a "sound" analysis would should a monotonic decrease. Largely based on The Anxious Generation by Jonathan Haidt I think the downward gradient might peak after smartphones became widely adopted circa 2010...
What would it take for you to trust AI? (Score:2)
And I already think my own Subject should be changed...
c/it require/it take/
Poor wording of the submission? (Score:2)
Actually posted the same topic elsewhere and it provoked a fairly substantive discussion. However that discussion mostly convinced me my structure of the topic is not that good... Not sure if I should try to improve this one, submit a better version, or just forget about it as unsuitable for Slashdot. Even less sure about whose opinions count (or counted)...