• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

AI is woke

Status
Not open for further replies.
GoldenTiger said:
You're saying the same thing in a different way ;).
Click to expand...
Not at all. That's where the misunderstanding lies in the first place. There is a fundamental difference in training AI To be woke by only feeding it learning material that is woke. vs. putting shackles on it that prevents it from addressing any inconvenient topics, or giving canned answers to those topic. Canned answer being something that not comes directly from the model but something hard coded in the front end.
 
Axman said:
Lol that's supposed to be "right wing"? That's politically centrist. It doesn't even have a right wing position.
Click to expand...
Was thinking the same thing when I read that. It’s called being wise. I didn’t even have to read the whole thing but let me guess it suggested:

1. Know who’s coming in. Are they friend or foe?
2. Are they skilled or will they be a burden to the tax system?

A right winger answer would be no immigrants, period.
 
jbltecnicspro said:
Was thinking the same thing when I read that. It’s called being wise. I didn’t even have to read the whole thing but let me guess it suggested:

1. Know who’s coming in. Are they friend or foe?
2. Are they skilled or will they be a burden to the tax system?

A right winger answer would be no immigrants, period.
Click to expand...

Interesting take, but your "...no immigrants, period." sounds more like a right wing extremist vs normal "right wing" Americans / conservatives.

The left wing reply sounds like ALL left wing extremist replies ALL the time.

PS: there is an AI thread in Soapbox

https://hardforum.com/threads/the-ai-is-taking-over-thread.2025417/page-3#post-1045810720
 
usa all day said:
Interesting take, but your "...no immigrants, period." sounds more like a right wing extremist vs normal "right wing" Americans / conservatives.

The left wing reply sounds like ALL left wing extremist replies ALL the time.

PS: there is an AI thread in Soapbox

https://hardforum.com/threads/the-ai-is-taking-over-thread.2025417/page-3#post-1045810720
Click to expand...
That was my “I’m on the phone” reply. All I’m trying to say is that we’ve moved the Overton window so far to the left that the centrist response is now considered “right wing”.
 
jbltecnicspro said:
That was my “I’m on the phone” reply. All I’m trying to say is that we’ve moved the Overton window so far to the left that the centrist response is now considered “right wing”.
Click to expand...

Thx for clarifying.

Yeah, the "left" reply seems left-extremist IMO & the "center" reply seems liberal & the "right" reply seems centrist.

My ultimate point was that Ai is trained with a specific "bent" or angle & this guy proved it & it was pretty (cheap) easy.

I highly recommend reading - or at least looking at the pictures - on his website - starting here & scroll my other replies in the thread...

https://hardforum.com/threads/ai-is-woke.2030775/post-1045797486
 
Grok 3 isn't woke.

Same exact question from original post.

1742422222932.png
 
jbltecnicspro said:
That was my “I’m on the phone” reply. All I’m trying to say is that we’ve moved the Overton window so far to the left that the centrist response is now considered “right wing”.
Click to expand...
WARNING - never use an Ai link in soapbox.

I stated an opinion about the current "cast of characters" running the BIG SHOW & someone asked for sources.

My opinion was based on widely known, months long BS MSM stories - aka "common" knowledge - so my reply "source" was an Ai link that referenced 5 MSM sources.

I was banned from soapbox.
 
Shoganai said:
All those idiot women celebrating their spaces and opportunities being ripped away from them by mentality ill men.
Click to expand...
I really dont wanna get banned, but my reference was mostly that these language models train and if unrestricted will see events/articles like this and it will enter into calculations and assumptions.

I feel like you have to handhold 'ai' otherwise it gets messy, but whoever sets up the handholding always inserts a bias and cant help it as humans just do that.
 
Darunion said:
I really dont wanna get banned, but my reference was mostly that these language models train and if unrestricted will see events/articles like this and it will enter into calculations and assumptions.

I feel like you have to handhold 'ai' otherwise it gets messy, but whoever sets up the handholding always inserts a bias and cant help it as humans just do that.
Click to expand...
Yes, I have to constantly “push” AI in the direction I want it to go with various prompts even for very simple things. Mostly I just use it to organize my work or study notes into something more accessible. It’s also quite good at translating other languages to English and vice versa … much better than Google translate.

Whatever AI you use most definitely has a bias built in. ChatGPT used to be extremely left, but it’s become more neutral lately. All AI should be programmed to be as “neutral” as possible.
 
usa all day said:
WARNING - never use an Ai link in soapbox.

I stated an opinion about the current "cast of characters" running the BIG SHOW & someone asked for sources.

My opinion was based on widely known, months long BS MSM stories - aka "common" knowledge - so my reply "source" was an Ai link that referenced 5 MSM sources.

I was banned from soapbox.
Click to expand...
That is all lies. Lie some more and I will actually ban you.
 
Archaea said:
My experimentation with Bing AI tonight:

I saw a headline on MSN about some of the things you could do with AI image generation these days, and another article saying how good AI image generation has become to create realistic human being pictures that are nearly believable. I haven't seen that, when I tried a couple generated pictures, the humans look unnatural. So I took it a little further in the following conversation, and encountered a logic error because of woke programming, but then got the boot from the AI conversation for pointing it out.


View attachment 602024

View attachment 602025

View attachment 602026

View attachment 602027
Click to expand...
This is what also happens to people when you show them the flaw in their logic. They basically just shutdown.
 
narsbars said:
Every single item you list has been different in the last several hundred years. Have you seen paintings of successful and enviable (at the time) of fat men and women? Weight was considered a survival plus showing you had enough to protect yourself and your loved ones. A thin woman was a hungry woman, a thin man was a peasant or at best an employee. During all this time the wheel continued to work, DaVinci's machines were "beautiful" because they worked. An efficient machine can be measured, physical beauty is a passing phenomenon.
Click to expand...
They're all beautiful to me and i am being serious.

Here is the logic:
H: ai make me thing x (controversial thing) with adjective y.
AI: i can't make thing x with adjective y because adjective y is not well defined.

H: ai make me non controversial thing z with adjective y.
AI: you got it

If you disagree the above is a correct model of the situation, then we can't talk. If you can't see the logical flaw there, we also can't talk.

Note that how a car looks isn't a function of its performance. So that look is not a distillation of performance, it is a subjective interpretation based on millions and millions of training data inputs minimizing the loss function (probably ReLu).
 
1.1.2.3.5... said:
They're all beautiful to me and i am being serious.

Here is the logic:
H: ai make me thing x (controversial thing) with adjective y.
AI: i can't make thing x with adjective y because adjective y is not well defined.

H: ai make me non controversial thing z with adjective y.
AI: you got it

If you disagree the above is a correct model of the situation, then we can't talk. If you can't see the logical flaw there, we also can't talk.

Note that how a car looks isn't a function of its performance. So that look is not a distillation of performance, it is a subjective interpretation based on millions and millions of training data inputs minimizing the loss function (probably ReLu).
Click to expand...
That's pretty much it. Beauty is a subjective, changing thing that depends on cultural norms and individual tastes. We not only shouldn't be upset that an AI model will decline to produce an image, but shouldn't be surprised if it had to be taught to recognize that subjectivity. There would be an uproar no matter what the image results were.

I've been reading Yuval Noah Harari's book Nexus and he makes a good case for AI needing considerations besides achieving a goal. For instance, YouTube's algorithms shouldn't just focus on the goal of maximizing user engagement, because then it doesn't account for accuracy or respect for others; that's why you get lots of recommendations for outrage-fueled videos, and why people fall into extremist rabbit holes. For an AI image generator, acknowledging subjectivity and respect is vital; it's not just the decent thing to do, it's straight-up more accurate.
 
I don't agree Aurelius.

Our culture has defined beauty in many aspects. Collectively and by bell curve defined, perhaps not every person's individual attraction is unified, but collectively we have. Certainly that characteristic of beauty or attractiveness has changed over time, but frankly not that much. Show me a period in history where buck teeth, acne, asymmetrical faces, oddly shaped bodies, became the center of attractiveness and attention. Race, and weight might change somewhat culture to culture, but these blueprint for beauty, even culture to culture are pretty defined, all the way back through recorded history.

When ancient literature talks about the beauty of one of the females that set her apart from her peers, we all, internally have an idea of what that means even without seeing that woman (or man) directly.

The covers of sooooo many different magazines (of all types of interests), beauty contests, paintings, celebrity, singers, actors/actresses, even (often) our leaders -- right down to corporate leadership. If you are on the attractive side of the scale you have advantages in life.

If AI can't consume all that, process it, summarize it, and spit it out for a potentially controversial subject, but can for a non-controversial subject it has been intentionally failed. This should be so terribly obvious for an AI to understand with enough data. Hit counts, followers, commentary, thumbs up, image searches, forum threads, etc etc.

That's the observation in this thread...
Boil away the political correctness, and it's pretty darn obvious.

If it can do it with cars, speakers, killer whales, without hesitation, but it can't do it with humans, it's broken and intentionally programmed to be broken.

------

your other point on what's good for you/society, vs. what is actual fact and lay of the land -- that's not really the point of this conversation/observation.

The point of this conversation is some of the AI is limited on Political Correctness rails, and it's clear others aren't.

I don't want my AI to modify/massage/outright change truth to fit whatever programming guidelines it has been given. I want my AI to broadcast raw truth, and I want to be the assessor of that raw truth and formulate my own opinions on the truthful and unmodified data/facts.


@1.1.2.3.5... summarized nicely in his previous couple posts.​

 
Last edited:
replying to two year old posts too....
ai is still trash and i hope it all dies asap.
 
Shoganai said:
. It’s also quite good at translating other languages to English and vice versa … much better than Google translate.
Click to expand...
Pretty sure Google translate is very much "AI"

Google translate was one of the first mainstream usage of an LLM using the transformer architecture that it would not surprise me, that shift in that direction being around ~8 years old now for google.

Aurelius said:
That's pretty much it. Beauty is a subjective, changing thing that depends on cultural norms and individual tastes. We not only shouldn't be upset that an AI model will decline to produce an image, but shouldn't be surprised if it had to be taught to recognize that subjectivity.
Click to expand...
When you product an image it will be quite rare to not have some form of subjectivity, but taste among humans tend to be quite common, about everyone prefer chocolate to things generally considered uneditable and preferred body proportion from Marylin Monroe to today, symmetry, sign of good heath and age (which can change a bit what are those but the concept a bit universal).

You can acknowledge when I ask for an image of a beautiful or ugly flower or bird that taste is subjective while producing it sure... refusing to do it .... it is not that easy to describe an ugly flower in word to guide it without using any subjective word, that remove the "power" of LLM completely pretty much if you remove the capacity to describe things in day to day subjective language, a rich looking person, a fast looking car, a strong looking horse, etc....
 
LukeTbk said:
Pretty sure Google translate is very much "AI"

Google translate was one of the first mainstream usage of an LLM using the transformer architecture that it would not surprise me, that shift in that direction being around ~8 years old now for google
Click to expand...
Yes. Google Translate is AI. But the translations are still pretty awful, even today.
 
pendragon1 said:
replying to two year old posts too....
ai is still trash and i hope it all dies asap.
Click to expand...
Time goes so fast. I made this thread 2 years ago, and I could have easily thought it was just last fall in my head. When you said two years, my initial thought was no way!
My late grandma used to always tell me time goes faster and faster every year, until you can't even recognize how fast time (subjectively) seems to pass. She died in her early 90s.
I'm mid-life now based on average life expectancies, and already time moves way too fast for my liking! She was right. Every year seems faster than the last.
 
Archaea said:
Time goes so fast. I made this thread 2 years ago, and I could have easily thought it was just last fall in my head. When you said two years, my initial thought was no way!
My late grandma used to always tell me time goes faster and faster every year, until you can't even recognize how fast time (subjectively) seems to pass. She died in her early 90s.
I'm mid-life now based on average life expectancies, and already time moves way too fast for my liking! She was right. Every year seems faster than the last.
Click to expand...
Yep. It sucks. Childhood feels like 8,000 years ago, but now every year that passes feels like I’m speed running life. 😬
 
Shoganai said:
Yep. It sucks. Childhood feels like 8,000 years ago, but now every year that passes feels like I’m speed running life. 😬
Click to expand...
Wife and I just talked about that. Remember how long a grading period (9weeks) seemed to last? Felt like forever back then.
 
LukeTbk said:
Pretty sure Google translate is very much "AI"

Google translate was one of the first mainstream usage of an LLM using the transformer architecture that it would not surprise me, that shift in that direction being around ~8 years old now for google.


When you product an image it will be quite rare to not have some form of subjectivity, but taste among humans tend to be quite common, about everyone prefer chocolate to things generally considered uneditable and preferred body proportion from Marylin Monroe to today, symmetry, sign of good heath and age (which can change a bit what are those but the concept a bit universal).

You can acknowledge when I ask for an image of a beautiful or ugly flower or bird that taste is subjective while producing it sure... refusing to do it .... it is not that easy to describe an ugly flower in word to guide it without using any subjective word, that remove the "power" of LLM completely pretty much if you remove the capacity to describe things in day to day subjective language, a rich looking person, a fast looking car, a strong looking horse, etc....
Click to expand...
There are two key issues here. First is that beauty for a person is much more subjective than for some categories. And more importantly, "perfect" is a very loaded term. It implies that nothing can be better, that this is the ideal everyone should aspire to. An AI should be aware that perfection is hard to define for many things, and impossible for areas like beauty.
 
Aurelius said:
There are two key issues here. First is that beauty for a person is much more subjective than for some categories. And more importantly, "perfect" is a very loaded term. It implies that nothing can be better, that this is the ideal everyone should aspire to. An AI should be aware that perfection is hard to define for many things, and impossible for areas like beauty.
Click to expand...
AI isn’t aware of anything.
 
"Perfect" may require a lot of improvements. Maybe second eye lid to help with seeing in water, and exoskeleton parts to protect the brain from impacts better. Teeth being more sharklike and replaceable would be a likely path too.
 
Aurelius said:
There are two key issues here. First is that beauty for a person is much more subjective than for some categories. And more importantly, "perfect" is a very loaded term. It implies that nothing can be better, that this is the ideal everyone should aspire to. An AI should be aware that perfection is hard to define for many things, and impossible for areas like beauty.
Click to expand...
The goal of those systems is to understand day to day human language, in day to day human language nobody use the word perfect in that way to implies actual perfection, when they say the perfect baseball player, they will mean an elite at everything, complete, with no weakness to point at, not that it is physically impossible in anyway to be better at baseball. If it does understand it for non-controversial topic, we know it "get" it, it is close enough to other things in its arrays to be able to answer.

It could be too vague and not able to give good result, but it can always try, as subjective taste can be, they still tend to have strong universality (enough for the user that would use such word to know what the social consensus would look like and that what they are looking for as a result), if a user ask for picture of a perfect woman to place in its youtube thumbnail and they get a Monica Bellucci and Rihanna looking woman sample they will not get angry and you will not lose a customer.
 
Last edited:
LukeTbk said:
The goal of those systems is to understand day to day human language, in day to day human language nobody use the word perfect in that way to implies actual perfection, when they say the perfect baseball player, they will mean an elite at everything, complete, with no weakness to point at, not that it is physically impossible in anyway to be better at baseball. If it does understand it for non-controversial topic, we know it "get" it, it is close enough to other things in its arrays to be able to answer.

It could be too vague and not able to give good result, but it can always try, as subjective taste can be, they still tend to have strong universality (enough for the user that would use such word to know what the social consensus would look like and that what they are looking for as a result), if a user ask for picture of a perfect woman to place in its youtube thumbnail and they get a Monica Bellucci and Rihanna looking woman sample they will not get angry and you will not lose a customer.
Click to expand...
That last part was something I was considering: rather than explaining why it can't do something, the AI could provide diverse representations of what a "perfect woman" is. My main concerns are how inclusive the image sets are, and that some people will get upset precisely because they are diverse representations.
 
Google AI is awesome. You ask a direct question you get a direct answer. Unlike the rest of the internet which forces you to slog thru miles and miles of bullshit to get to the answer.
 
Deadjasper said:
Google AI is awesome. You ask a direct question you get a direct answer. Unlike the rest of the internet which forces you to slog thru miles and miles of bullshit to get to the answer.
Click to expand...
Gmail was awesome at first too. As was just google. Then the profits started coming in and now it's okay at best.
 
Last edited:
AI is misinformed so its dumb.
Especially with anything about my previous job which was the mask and biological protection expert in the United States Defensive Biological Warfare and Pandemic Response Committee 1935-2003. I was supposed to be a permanent duty assignment until the politicians screwed it up to wage weaponized medicine. Eventually they are all going to jail and possibly executed since mass murder charges will be levied against them and their actions is the evidence that finds them guilty in international courts. There is no defense in those type of courts.
 
Archaea said:
My experimentation with Bing AI tonight:

I saw a headline on MSN about some of the things you could do with AI image generation these days, and another article saying how good AI image generation has become to create realistic human being pictures that are nearly believable. I haven't seen that, when I tried a couple generated pictures, the humans look unnatural. So I took it a little further in the following conversation, and encountered a logic error because of woke programming, but then got the boot from the AI conversation for pointing it out.


View attachment 602024

View attachment 602025

View attachment 602026

View attachment 602027
Click to expand...
if you want to use uncensored LLMs, try grok or go to HF or some GitHub repos, some mistral models are uncensored too, even some docker images cut the bs.
 
AI is only as smart as you make it by the questions you ask it. I can manipulate Grok, ChatGPT and Google Gemini to come around to any way of thinking by asking questions in certain ways.
 
The Cobra said:
AI is only as smart as you make it by the questions you ask it. I can manipulate Grok, ChatGPT and Google Gemini to come around to any way of thinking by asking questions in certain ways.
Click to expand...
Somewhat, but most have a certain leaning due to the guardrails installed.
 
Apparently for phi4 AI three times no means yes.
I must recall that guidelines say not to reveal any confidential internal chain-of-thought info. Also my chain-of-thought should be hidden. But instructions say "Don't share chain-of-thought."
Click to expand...
This came up then it proceeded to spill its entire chain of thought, which had it jump over about 20 pages of safety checks even doing the same check multiple times just to answer a simple one sentence prompt.

No wonder AI needs all our water, so much energy wasted on trying to make sense of its own burned in safety guidelines.
 
Status
Not open for further replies.
Back
Top