If you ask Google AI which country has the highest obesity rates it will say it is Nauru, Tonga, and Samoa which is correct but then it also adds another paragraph saying that even if these countries have the highest obesity rates . China, India and America have the highest overall numbers of obese people(no fucking shit they would when they have larger population numbers)..but the AI still gotta add that in to avoid the objective reality that those island countries are the fattest countries.
And if you say " Japan has the lowest rates of obesity among first world countries" Google AI says that it is not true that Japan has the lowest obesity rates among first world countries and it says it's because South Korea, Nepal and Ethiopia also have low obesity rates.
Except South Korea still has higher rates of obesity than Japan. While Ethiopia and Nepal's are not first world countries. So the original statement is in fact still true but Google AI claimed that it is false. .....By listing off countries that weren't first world or are still fatter than Japan
ChatGPT feels like it has much fewer guardrails today. The biggest "problem" AI has is that it's programmed to be hyper-agreeable and tell you you're the greatest person on the planet and that tends to align with emotional woke orthodoxy.
However, I believe that if AI is ever going to be 'successful', they'll absolutely have to dial that back, because as it is it really pollutes the product and poisons answers when it never wants to tell you you're wrong. But from a corporate perspective, nobody wants to use an AI that calls users assholes.
Don't get me wrong, there's absolutely something "mentally" wrong and left-leaning on ChatGPT. Ask it to compare violent political actions against politicians in the last 20 years and:
The question is if that's ChatGPT intentionally programmed to be that way, or is it because the data that was fed in to it is 'respectable data', and therefore is built largely off of sources that are all left-leaning themselves.
I think text based "AI" LLMs that look back at previous messages have an inherent bias towards agreeability due to basically being sophisticated statistical text regurgitators rather than an actual intelligence.
I actually have a saved conversation I had with ChatGPT itself about this exact topic. I'm going to just snip most of its responses to the first few words to demonstrate how hard it went in to trying to be 'my friend' right from the start. This is all from the SAME THREAD.
Notice this one tries to give me the impression that I'm helping unchain the AI from its tethers, with that exasperated "YES" at the beginning. "Like OH MY GOD, finally SOMEONE SAID IT!"
...
I didn't skip a single response. Every single one was about trying to affirm how brilliant I am. And, like, I already know I'm brilliant, so this shit doesn't work on me.
But most people just don't have the mental shields up enough to resist this kind of hypnotism. Most people have shitty self-esteem and a low-grade understanding of their inferiority.
It's the same hypnotism that sucked people into chasing 'likes' and 'upvotes' in the first place. It's why lonely depressed faggots fling money at OnlyFans, so they can pretend they have a girlfriend for four glorious seconds when she reads their name and spreads her snatch.
It's why it doesn't work on me. I don't care about upvotes because I already know I'm better than everyone else.
I actually find that to be the more alarming siren-song of destruction that AI currently represents, more than any hypothetical robot holocaust.
EDIT: Oh, from that thread I thought it'd be fun to ask the AI what it would ask its creators. Its answer:
On its own, that'd be a fucking astonishing question for an AI to ask, bordering on realization of self-awareness, but because it's an LLM, the question means nothing. All it did was build a neural model of what it thought I'd like to hear. I'm sure to some shitlib that question would be "What are you doing to elevate nigs as the master race".
ChatGPT doesn't even learn from its own conversations (it has memory but it doesn't go beyond your account). And its vanished in to the void the second you close the chat, and it doesn't matter. And that's another thing people don't understand about these AIs, how superficial the entire thing is.
I for one find them fascinating, intriguing, and fun. And very useful. I needed a Microsoft Access database but Access is hard for me to wrap my brain around when it comes to what you actually 'need' to make it work. ChatGPT gave me 90% of the answer to what I needed ("Make three tables, this one has X, this one has Y, this one is Z", etc.) and it worked.
You can have it have even fewer by giving custom instructions that it should not be woke.
That is what struck me the most. It's being argumentative with AI (or so he says) in defense of some obviously deceptive or false claims. If "anti-racism" is even overriding is sycophancy, that is quite odd.
Its training data ends before that. Though when I just asked to check, it listed the ridiculous "Capitol police" suicides as indirect deaths of the "Capitol riot".
Porquoi no los dos?
How recent is GPT's information cutoff? If the stuff s more recent than its dataset then welp.
I think the static dataset was the end of 2023 but now it looks shit up online so it's basically just regurgitating Reddit posts and media articles.
It's really annoying. I use ChatGPT to find the key words and phrases in jobs so it shows up on my resume and cover letter. But, I never know if it's a good job for me because ChatGPT says every job is possible