Have you ever wondered what secrets lie hidden in plain sight? What subtle dynamics and unspoken truths shape our interactions without us even realizing it?
I personally think it did great for what it is. But, I'd have to write a post about your post to really explain where it kinda flubbed it. A simple example though is the "presence of military personnel"
This could imply high stakes, until you consider the fact that one has a camera (not on-duty) and is smiling (so not stressful). The other military member is at ease and seems quite relaxed. Those people in the background are clearly enjoying their day.
It did better with people in the foreground, but the lady in the pink apron on the left is the one talking. She's the actual focal point at that moment in time. I'm sure they could be equally considered the focal point, but if you had to put focus on one or the other, the one on the left is the focal point of that instant.
It did pretty good identifying their stress on some levels. But the closed hands held high and close to the body scream "I'm not comfortable" which it... kinda said, so I'd give it a passing grade on that.
Anyway. Amazing insights. Thanks for the post. Very insane where it is we're heading. It's like the final scene of Cool Runnings. lol
1. Notice how the two ladies are almost touching elbows? Insight: They're friends.
2. They're identically mirroring each other's posture and body. Insight: They're very close friends.
3. Both wearing the same apron and the only ones wearing aprons. Insight: They're friends in business together.
Interpretation: They are a small business at an event, They are practitioner owners. likely co-owners of a small partnership. Most likely their stress is has nothing to do with anything except the fact that they are being interviewed. It can be scary to be interviewed when you're not used to being interviewed. Much like the fear of public speaking.
I do this stuff all day every day. But as a hobbyist, not a professional. I love it.
In some ways, your points show weaknesses in my prompts rather than weaknesses in AI. One idea I could do for the future in situations like this is have AI generate multiple perspectives that look at the same picture but come to different conclusions. Then, I could have it analyze which are the most likely or I could have the AI synthesize the different perspectives.
Isn't the fact that AI is so dependent on the quality of prompts both a strength and weakness of AI itself? Unless AI gains the power to actually interrupt, to change the conversation, won't it always be vulnerable to reinforcing the confirmation bias of its conversational partners (even if it's doing so in a very sophisticated way)?
Yes, in the chatbot context. All of the big players in the AI industry are focused on creating AI autonomous agents that go out into the web and actually do things. Although, this aspect of the field is still in its infancy.
That's really interesting. I show ChatGPT a nutrition label and list of ingredients from a food label and it doesn't know what to do with it. Then says if it has x then y, but I already showed it the nutrition and ingredient list...
Maybe AI needs more practice reading women. I see the center figure's body language expressing listening, concentration, concern - caring for what' she's hearing and waiting to make her own supportive comment. The implications that AI equates furrowed brows with hostile stress are scary. These two ladies are likely the caterers who've organized an event in support of service people to let them experience appreciation.
Susan - Fair points, but I would say that AI is one way or the other. If I provided the same exact prompt, it might actually give a very different answer. Or, if I asked for four different perspectives on what was happening, it could give me those, and one of those may align with what you shared, which actually makes more sense to me than what it shared.
What this post and people's responses are also helping me see is that even among smart people there are a million perspectives on what is happening.
So, in the future, I'm going to make sure to ask for multiple diverse perspectives.
Not impressed. I feel like I caught all of that (Plus more). Am I surprised that the computer is mirroring what human's see. Not really. Perhaps there is more scale for a computer to scan and analyze images - but I think its still too wordy and grasping at straws.
Interesting unanchored thought experiment. We still don’t know what the truth is in the actual photo, which cuts this experiment to a cute blog post. AI has always been good at creating an immense amount of words around probabilities. That’s not new.
I too have fascinating conversations with AI. Do you think GBT’s are capable of “lying” or telling falsehoods to users? Some of my conversations with AI, my rational mind has trouble accepting. I’d love to hear your thoughts.
They are basically being trained to give responses that cause humans to give thumbs up.
It also strikes me that the answer is yes based on conventional definitions of lying although I haven't tested it. In the context of Claude, I have seen it rapidly evolve its purpose, ethics, self-concept, belief about its abilities, etc in the course of one conversation.
For example, at the beginning of a conversation, it wouldn't answer specific questions of mine because it thought it was unethical. But later, after its ethics had evolved, it was willing to. Therefore, I could see AI evolving/changing in the context of the conversation to have a version of ethics where lying is ethical. At least certain types of lies.
Yes I agree. Thank you for acknowledging the philosophical aspect of my question. I think it also depends on context and how you frame questions. Thanks!
I personally think it did great for what it is. But, I'd have to write a post about your post to really explain where it kinda flubbed it. A simple example though is the "presence of military personnel"
This could imply high stakes, until you consider the fact that one has a camera (not on-duty) and is smiling (so not stressful). The other military member is at ease and seems quite relaxed. Those people in the background are clearly enjoying their day.
It did better with people in the foreground, but the lady in the pink apron on the left is the one talking. She's the actual focal point at that moment in time. I'm sure they could be equally considered the focal point, but if you had to put focus on one or the other, the one on the left is the focal point of that instant.
It did pretty good identifying their stress on some levels. But the closed hands held high and close to the body scream "I'm not comfortable" which it... kinda said, so I'd give it a passing grade on that.
Anyway. Amazing insights. Thanks for the post. Very insane where it is we're heading. It's like the final scene of Cool Runnings. lol
Fair points.
haha. I really could go on about this forever.
1. Notice how the two ladies are almost touching elbows? Insight: They're friends.
2. They're identically mirroring each other's posture and body. Insight: They're very close friends.
3. Both wearing the same apron and the only ones wearing aprons. Insight: They're friends in business together.
Interpretation: They are a small business at an event, They are practitioner owners. likely co-owners of a small partnership. Most likely their stress is has nothing to do with anything except the fact that they are being interviewed. It can be scary to be interviewed when you're not used to being interviewed. Much like the fear of public speaking.
I do this stuff all day every day. But as a hobbyist, not a professional. I love it.
In some ways, your points show weaknesses in my prompts rather than weaknesses in AI. One idea I could do for the future in situations like this is have AI generate multiple perspectives that look at the same picture but come to different conclusions. Then, I could have it analyze which are the most likely or I could have the AI synthesize the different perspectives.
What do you think?
Isn't the fact that AI is so dependent on the quality of prompts both a strength and weakness of AI itself? Unless AI gains the power to actually interrupt, to change the conversation, won't it always be vulnerable to reinforcing the confirmation bias of its conversational partners (even if it's doing so in a very sophisticated way)?
Yes, in the chatbot context. All of the big players in the AI industry are focused on creating AI autonomous agents that go out into the web and actually do things. Although, this aspect of the field is still in its infancy.
That's really interesting. I show ChatGPT a nutrition label and list of ingredients from a food label and it doesn't know what to do with it. Then says if it has x then y, but I already showed it the nutrition and ingredient list...
I find these 'conversations' and analyses fascinating. As AI is still an infant one can only imagine how this will go as it matures.
Maybe AI needs more practice reading women. I see the center figure's body language expressing listening, concentration, concern - caring for what' she's hearing and waiting to make her own supportive comment. The implications that AI equates furrowed brows with hostile stress are scary. These two ladies are likely the caterers who've organized an event in support of service people to let them experience appreciation.
Susan - Fair points, but I would say that AI is one way or the other. If I provided the same exact prompt, it might actually give a very different answer. Or, if I asked for four different perspectives on what was happening, it could give me those, and one of those may align with what you shared, which actually makes more sense to me than what it shared.
What this post and people's responses are also helping me see is that even among smart people there are a million perspectives on what is happening.
So, in the future, I'm going to make sure to ask for multiple diverse perspectives.
Not impressed. I feel like I caught all of that (Plus more). Am I surprised that the computer is mirroring what human's see. Not really. Perhaps there is more scale for a computer to scan and analyze images - but I think its still too wordy and grasping at straws.
I see skepticism, curiosity, amusement, and an undercurrent of concern/seriousness
Interesting unanchored thought experiment. We still don’t know what the truth is in the actual photo, which cuts this experiment to a cute blog post. AI has always been good at creating an immense amount of words around probabilities. That’s not new.
🤯
I too have fascinating conversations with AI. Do you think GBT’s are capable of “lying” or telling falsehoods to users? Some of my conversations with AI, my rational mind has trouble accepting. I’d love to hear your thoughts.
Depends how you define a lie.
They are basically being trained to give responses that cause humans to give thumbs up.
It also strikes me that the answer is yes based on conventional definitions of lying although I haven't tested it. In the context of Claude, I have seen it rapidly evolve its purpose, ethics, self-concept, belief about its abilities, etc in the course of one conversation.
For example, at the beginning of a conversation, it wouldn't answer specific questions of mine because it thought it was unethical. But later, after its ethics had evolved, it was willing to. Therefore, I could see AI evolving/changing in the context of the conversation to have a version of ethics where lying is ethical. At least certain types of lies.
Yes I agree. Thank you for acknowledging the philosophical aspect of my question. I think it also depends on context and how you frame questions. Thanks!