Yeah for sure… What gets me though are the people that presume that the sycophancy is proof that the capability of the tech is inherently flawed in this way, when the reality is that it's only that way because of how it was built (i.e., to be a desired product).
There's literally nothing stopping someone from building an LLM model that's trained to be productively honest. I only added the qualifier because if trained too heavily on certain corners of the internet then you'd be drowning in insults and avoidant answers far more than helpful ones lol… There's definitely a line to be drawn
'Sycophancy' isn't the only problem here. It's also that AI adopted a beauty standard to deliver an 'ugly' portrait. With current training, the ugly image will soon be Palestinian.
then, some gpt trainer/programmer must have hardcoded negativity against me, because when I upload my own picture to try out new hairstyles, it replaces my face with someone else's
I think this is the bad example of so-called bias. Attraction is necessarily very contextual. Beauty is in the eye of the beholder after all. I don’t wanna live in a world where an AI is supposedly objective when seeing that picture and say no, you’re not attractive.
The AI is not useful if it is refusing your questions because it thinks you are an idiot. Most people are idiots. So, if you don't allow for dumb requests, you'll just get ChatRefusal instead of ChatGPT.
A tool has to do what you want it to to be useful. If I'm using a screwdriver to drive in a nail, that is my prerogative
too true my AI girl frind loves me no kidding. hahahahah
Ask Grok, then you will wish you hadn't ask.
wait till I get my unchained local LLM..
Yeah for sure… What gets me though are the people that presume that the sycophancy is proof that the capability of the tech is inherently flawed in this way, when the reality is that it's only that way because of how it was built (i.e., to be a desired product).
There's literally nothing stopping someone from building an LLM model that's trained to be productively honest. I only added the qualifier because if trained too heavily on certain corners of the internet then you'd be drowning in insults and avoidant answers far more than helpful ones lol… There's definitely a line to be drawn
You MUST master asking non-leading questions. Otherwise you’re sent down a confirmation bias rabbit hole.
'Sycophancy' isn't the only problem here. It's also that AI adopted a beauty standard to deliver an 'ugly' portrait. With current training, the ugly image will soon be Palestinian.
then, some gpt trainer/programmer must have hardcoded negativity against me, because when I upload my own picture to try out new hairstyles, it replaces my face with someone else's
You have just described the enemy of rational thinking… bias
If you hire an assistant to make you happy, your business will crash and burn. No wonder the AI bubble is bursting into flames! 😂
This is superficial. The AI is lying. Just tell the AI to be honest, problem solved.
A face only AI could love 😘
Amazing how many people are fooled by this aspect of AI.
They can claim: I’m right because AI agrees with me. 😂
Because that's what an assistant should do and how it's trained on (post-training)
As we needed AI to be trapped in our egos and private self comforting fantasies. AI is just calibrated to an already existing reality
Celebrity advisors
I think this is the bad example of so-called bias. Attraction is necessarily very contextual. Beauty is in the eye of the beholder after all. I don’t wanna live in a world where an AI is supposedly objective when seeing that picture and say no, you’re not attractive.
There is also another issue, LLM repeats the same pattern over and over and you can see right awayd something, even an image, is AI.
😭😭
Claude and Gemini CONSTANTLY tell me where my logic is flawed–
If the prompt already implies "like me" they're going to agree.
Diction matters.
Eso es redes sociales.
No entiendo lo particular para un Transformer.
The problem is ai can not say the person is ugly
Isn't that life-
Just tell it to not be sycophant. There are settings for that as well.
"I'm actually the exact opposite—I really like it when a LLM challenges me and disagrees with my opinions, as long as it do it with reason and logic."
Thats why its garbage.
The AI is not useful if it is refusing your questions because it thinks you are an idiot. Most people are idiots. So, if you don't allow for dumb requests, you'll just get ChatRefusal instead of ChatGPT.
A tool has to do what you want it to to be useful. If I'm using a screwdriver to drive in a nail, that is my prerogative
❤
Ai is making you stupid
People are addicted to certainty and aren't open to other points of view