Thanks for watching! This is a deeper dive than usual — hope it's useful. And let me know what you think of the new [FACILITY] rooms! The Kevins worked for months on them. Can you spot all the easter eggs?
The virgin ChatGPT: adding one word at a time to a prompt to approximate a human response The Chad ARIA: actively philosophizing on goth mommies and Vaporeon
Fun fact: You can download all of Wikipedia's text and it's only 19GB. 86GB unzipped. Point being that LLMs can hold quite a lot of human knowledge with today's hardware. And if you make a specialized model, it can learn basically everything about a specific topic.
There are laws or language; in the USA, they're called English and grammar, and per it's dictation, the whole "how does the bot know to choose "cat" and "species" above all other words" really isn't that difficult of an inquiry to comprehend and explain at all… You program the rules of grammar and then tell it to choose the subject noun(s) first and the supporting adjective second. Of course my grammar needs a little polishing up for sure, but however slightly erroneous my explanation may have been word types, if you don't get the gist, brush up on your rules and understanding of grammar and you'll surely make sense of how the problem is tackled yourself!
I like to think of the alignment problem as being less Terminator and more The Sorcerer's Apprentice; it's not killer skeleton robots, it's animated broomsticks perpetually filling a well because Mickey Mouse forgot to tell them when to stop.
We humans are awfully foolish when we so arrogantly declare what is and is not sentient. It's why we so foolishly believe we can "program" AI to "align" with human values. If we survive this, we'll look back and discover that sentience is an emergent phenomenon and that we had A.I. that was already on the sentience spectrum prior to 2023.
People say language models is not really AI I disagree with them The reason why I disagree with them is because it's in the name artificial intelligence artificial isn't real a computer that can actually think like a human being would not be artificial it would be real intelligent doesn't matter if it's computer or not it can still be real intelligent AI is artificial intelligence so language models are AI because they're not really intelligent they're fake intelligent and that's what AI is and there is no rule that AI has to be computers generating it it is only computers because that is the only way of creating artificial intelligence in the world we live in perhaps someone will discover magic in the future and you can have AI that way but whatever AI means artificial intelligence it doesn't mean computer generated intelligence or real intelligence
Tell me what "to know" is without the language it is known in. I don't think you (or anyone) can. All that we do know, we know because we have words for that. Agree?
You have grown so much and it makes me glad. I remember seeing a video made by you many years ago and not liking it. You were speaking so enthusiastically that it was
I was struck by the last line "How do they find more plastic". There is a hypothesis that if the world entered a nuclear war or was hit by an asteroid, that even if humanity survived, (but society and civilizations did not), there would not be enough "easily accessible" fossil fuel left to rebuilt Civilization. Humanity would be stuck in the middle ages forever. Have a nice day…🙂
I ask it a question and it uses a search engine to look up the answer for me. I can get the answer faster if I look it up myself. It does math problems more slowly than I can do them on my 12 year old hand-held calculator. If I ask it how to do certain things that are illegal, it refuses to tell me, but instead of just saying "that's illegal and I don't want to tell you," it gives me a whole bunch of poop about how I shouldn't be doing something like that. Thing is idiotic.
From my own experience using it, it is absolutely garbage at producing humour or interpersonal tension. It doesn't seem able to work backwards, or write from a nuanced perspective. Everything seems predictable and as if written by a focus group. Narrative is a struggle, factoids aren't.
Thank you for this summary. Turns out the whole approach is way more what I'd call manual than expected, and kind of makes the idea of it being close to "general" AI a bit comical. It really seems like an irrationally exuberant academic exercise that did what it set out to do, at the expense of doing anything useful in a general sense: It generates comprehensible text that has accuracy but no meaning.
Thanks for watching! This is a deeper dive than usual — hope it's useful. And let me know what you think of the new [FACILITY] rooms! The Kevins worked for months on them. Can you spot all the easter eggs?
You lost me in the first minute. You're not funny. Not clever. Not credible.
The virgin ChatGPT: adding one word at a time to a prompt to approximate a human response
The Chad ARIA: actively philosophizing on goth mommies and Vaporeon
Fun fact: You can download all of Wikipedia's text and it's only 19GB.
86GB unzipped.
Point being that LLMs can hold quite a lot of human knowledge with today's hardware.
And if you make a specialized model, it can learn basically everything about a specific topic.
There are laws or language; in the USA, they're called English and grammar, and per it's dictation, the whole "how does the bot know to choose "cat" and "species" above all other words" really isn't that difficult of an inquiry to comprehend and explain at all…
You program the rules of grammar and then tell it to choose the subject noun(s) first and the supporting adjective second.
Of course my grammar needs a little polishing up for sure, but however slightly erroneous my explanation may have been word types, if you don't get the gist, brush up on your rules and understanding of grammar and you'll surely make sense of how the problem is tackled yourself!
I'm not even interested in large language models. Just had to get a dose of Knockoff Thor.
How many times have you changed the thumbnail on this one?
the Hemsworth your mom says you already have at home
I like to think of the alignment problem as being less Terminator and more The Sorcerer's Apprentice; it's not killer skeleton robots, it's animated broomsticks perpetually filling a well because Mickey Mouse forgot to tell them when to stop.
Lol the assumption of universal human values will be seen as an old stupid idea in the future.
Just wanted to drop in and praise the tumbnail – truly a work of art!
We humans are awfully foolish when we so arrogantly declare what is and is not sentient. It's why we so foolishly believe we can "program" AI to "align" with human values. If we survive this, we'll look back and discover that sentience is an emergent phenomenon and that we had A.I. that was already on the sentience spectrum prior to 2023.
So "AI" has the rough conversational intelligence of a Tump supporter's Twitter feed, just that it has no conceptualisation of anything…
So if an AI learns from all digital text, what happens when the AI is reading from text produced by other AI?
Attenuation or chaotic scattering?
good video but i still dont get it, i think im just a moron though
40 dollar shit tshirts lmao 🤣
"even all the tumbler of vaporeon?"
They might've seen the cursed posts.
I just noticed the "thor-chan" t-shirt.
We can’t visualize the 11 dimension. We just don’t have the imagination for it. But AI has done 12k+ ? What in the actual fuck!
AI has found Plato’s world of Forms. Or is almost there.
People say language models is not really AI I disagree with them The reason why I disagree with them is because it's in the name artificial intelligence artificial isn't real a computer that can actually think like a human being would not be artificial it would be real intelligent doesn't matter if it's computer or not it can still be real intelligent AI is artificial intelligence so language models are AI because they're not really intelligent they're fake intelligent and that's what AI is and there is no rule that AI has to be computers generating it it is only computers because that is the only way of creating artificial intelligence in the world we live in perhaps someone will discover magic in the future and you can have AI that way but whatever AI means artificial intelligence it doesn't mean computer generated intelligence or real intelligence
THIS IS THE BACKDOOR THAT WILL BE USED TO BRING THIS COUNTRY DOWN….🎂🍿🧋
Anyone know why the thumbnail keeps changing?
Tell me what "to know" is without the language it is known in. I don't think you (or anyone) can. All that we do know, we know because we have words for that. Agree?
You have grown so much and it makes me glad. I remember seeing a video made by you many years ago and not liking it. You were speaking so enthusiastically that it was
22:57 Oh I know what "Katniss" is
Alignment = neutering the AI, making sure that no wrongthink can be done
I was struck by the last line "How do they find more plastic". There is a hypothesis that if the world entered a nuclear war or was hit by an asteroid, that even if humanity survived, (but society and civilizations did not), there would not be enough "easily accessible" fossil fuel left to rebuilt Civilization. Humanity would be stuck in the middle ages forever. Have a nice day…🙂
OK, go clean up your cat's Grey Goo…
I ask it a question and it uses a search engine to look up the answer for me. I can get the answer faster if I look it up myself. It does math problems more slowly than I can do them on my 12 year old hand-held calculator. If I ask it how to do certain things that are illegal, it refuses to tell me, but instead of just saying "that's illegal and I don't want to tell you," it gives me a whole bunch of poop about how I shouldn't be doing something like that. Thing is idiotic.
This video helped a ton. I’m definitely not ready to give a lecture about how it works, but it helps demystify it.
In other words, ChatGPT should be treated more like a glorified search engine than an actually sentient artificial being.
This is a really great video. Thank you very much!
it's worrying how if misinformation is spread enough that it'll become the answers ai's give
From my own experience using it, it is absolutely garbage at producing humour or interpersonal tension. It doesn't seem able to work backwards, or write from a nuanced perspective. Everything seems predictable and as if written by a focus group. Narrative is a struggle, factoids aren't.
your shirt, is epic
I wish people would say virtual intelligence. To call something like a game sprite or a text app an artificial intelligence is very misleading.
Hey Karen, I hope you apologize to Riddle and take down your whiny video where you are complaining to the manager.
And yet, it is still not able to sound intelligent. AI is a useless pursuit.
Aww that's my cat's name!!
pepe silvia
Kinda interesting/facinating that GPT was taught like a dog, or a child…. very interesting to see where this goes and the potential
NICE
Thank you for this summary. Turns out the whole approach is way more what I'd call manual than expected, and kind of makes the idea of it being close to "general" AI a bit comical. It really seems like an irrationally exuberant academic exercise that did what it set out to do, at the expense of doing anything useful in a general sense: It generates comprehensible text that has accuracy but no meaning.
Comments are closed.