Practically human, then.
A new evaluation of artificial intelligence systems suggests that while modern language models are becoming more capable at logical reasoning, they struggle significantly to distinguish between objective facts and subjective beliefs. The research indicates that even advanced models often fail to acknowledge that a person can hold a belief that is factually incorrect, which poses risks for their use in fields like healthcare and law.
A couple of days ago, Grok confidently told me the most cost efficient route from Saskatoon to Indio, CA went through Oregon via I-5. So there’s that.
I find you get narrative based answers for certain questions. Try ‘is the COVID vaccine safe and effective’
AI is definitely not the be all and end all!
As I continue to say, AI IS BS.
Arrogant Incompetence.
Did anyone really expect a computer program to not reflect those who wrote it?
Garbage In Garbage Out.
G.I.G.O.
And the symbol minded will continue to chant”It says right here”,or the modern version; “Computer says”.
Well?It sure beats thinking.
Confidentally wrong. What I can’t believe is how the arrogant pricks at the AI companies and Nvidia are burning capital like nothing ever seen before and how they all think that when it runs out, the US goverment will bail them out anyway so they can keep doing it. Problem is they might not be wrong.
AI pioneer, Yann Lecun, says Large Language models are limited. He thinks a World Model – looking at the world – is how you create true artificial intelligence.
I’m not a computer geek, but maybe some poindexter could post their opinion on his view.
I am a computer geek and although I don’t work in either generalized AI or LLMs specifically, I have friends who do and I have a general understanding of the current state of the art.
The quote “This indicates the models may be relying on superficial pattern matching rather than a deep understanding of the concepts” summarizes the current problem with “AI”.
First of all, when people say “AI” right now they mean LLMs – Large Language Models. And there’s no “may” about it: relying on superficial pattern matching is all LLMs do, because it’s all LLMs can do. LLMs are nothing more, conceptually, than random sentence generators. They’re very, very good at producing random sentences that are grammatically correct and superficially sensible to a real person but they have no idea what they’re talking about. They have no “world model” that connects words to real things that have specific traits. That’s what Lecun means.
For example, if an LLM is constructing a sentence that starts with “I ate an apple…” it’s going to decide the next word is “pie” and not “iphone” not because it knows that an apple pie is something you can eat and a smartphone isn’t, but because in all the text that was fed into it initially the sentence “I ate an apple pie” appeared a lot more often than the sentence “I ate an Apple iPhone”. It’s playing probabilities, and sometimes it will guess wrong. This is why I don’t like the term “hallucinations” to describe LLM output errors. The LLM isn’t conceiving a different reality; it’s just guessing wrong about what the next word(s) should be.
The second part of the current problem is that everyone knows – or should know – that this is how LLMs work but everyone still somehow thinks being able to speak in grammatically correct sentences equates to generalized intelligence that has a world model that understands concepts in a platonic way. So you get drivel like “Why, it’s almost as if the AI models are behaving in exactly the way the people who designed them told us they did! How shocking!”
Grok is becoming increasingly laughable…going the way of Wikipedia it seems. Garbage in…garbage out.
Looked up the notorious Biden administration,DHS director Alejandro Mayorkas the other day. If he was Surgeon General I’m sure the words “great strides” and “cures for cancer” would be mentioned.
I didn’t know admitting god knows how many millions of the worst that humanity has to offer was really a focus on humanity.
Okey dokey.
“A couple of days ago, Grok confidently told me the most cost efficient route from Saskatoon to Indio, CA went through Oregon via I-5. So there’s that.”
The higher humidity along the west coast causes the engine to intake more water vapour, which increases the efficiency of a gasoline engine. For peak efficiency take Highway 101 (aka Oregon Coast Highway No. 9) during a “pineapple express”.
… “they struggle significantly to distinguish between objective facts and subjective beliefs”
A.I. = Liberals
As a native Californian … I GROKed that i-5 is the absolute WORST highway to travel N-S in CA (and WA and OR) … I learned this long ago. In fact when I travel from SF to LA … I drive further east to use hwy 99 which has 3 or more lanes most of the way, hence the illegal truck drivers cannot clog the two lanes of i-5 while leapfrog passing each other at 55 mph.
Dear GROK … i-5 is the WORST freeway in all of America.