Thanks for doing a video on Gemini 1.5 Pro lmao, everyone else was too busy hyping about OpenAI Sora. Which is no doubt mind blowing, but ya know, other news. Looking forward on your video about Sora tho!
0:39 This was not the actual performance of Claude 2.1 on the Needle in a Haystack (NIAH) Benchmark. The Original Performance of the benchmark was published by: "Greg Kamradt - Claude 2.1" The original creator of the benchmark. The one you got was a derivative with actual instructions to remember a specific point in the text. That is not indicative of real-world performance as people won't encode a sentence saying "This is the most important line in the document:....." just to retrieve it again in the model when asked about that line. The Claude 2.1 Test in the video was Flawed!
I feel you brother. The best thing you can do is this: Imagine life 1, 3, 5, and 10 years from now... Or at least the general future... And think: what should I be learning about which will be highly relevant then? What skills will be the most in demand? What am I currently learning that may become irrelevant in the future? Reflect on this long enough, and consistently, and you shall start to live your life (in skill acquisition and information gathering) in accordance with the future Hopefully this helps
@@2034-SWE I know that's what I should be doing, at least what's good for my mental health, but I couldn't help but imagine how irrelevant these personal skills would be in face of strong AGI or even ASI. If it really comes to that, I guess everyone would just be forced to create his own meaning...
@@YT-gv3cz I completely understand where you’re coming from. However, after comprehensive analysis of my own, I disagree. First of all, you must consider the massive amount of compute power required for the SCALE OF DEPLOYMENT you’re referring to. To replace 8 billion humans, you need A LOT of compute power. Why do you think Sam Altman mentioned we need $7 trillion..? That’s the compute power required just to support general commercial AI usage over next 5-10 years. Believe me, you’re not becoming “obsolete” anytime soon. What’s going to happen is everyone who has the digital means to subscribe and use any tool will become superhuman. AI augments and enhances our productivity. It only replaces certain archaic hard-skills (even then, to some extent for now). ------------------ I stumbled across a quote a few months ago by someone in the comment section. I’ll never forget it, and I’m living by it daily (and will for remainder of my life): “Creativity is the new literacy. Nonlinearity is the new numeracy. AI-centricity is the new operacy.” I’ll reply to this comment with GPT’s analysis of what that means. - Identify this is the future, modify your actions and skill acquisition, and you’ll not just “be fine” like everyone else. You’ll be in demand.
@@YT-gv3cz VisionArchitect GPT: “Your interpretation is quite accurate. These phrases indeed reflect a shift in the perception of essential skills and competencies in the modern, technologically advanced world. To elaborate further: 1. Creativity as the New Literacy: - In historical contexts, literacy, the ability to read and write, was a cornerstone of education and a primary indicator of intellectual capability. Today, as automation and artificial intelligence handle more routine tasks, the uniquely human trait of creativity is becoming as fundamental as literacy once was. It signifies a shift from valuing rote knowledge to valuing innovative thinking and the ability to create something new or solve problems in novel ways. 2. Nonlinearity as the New Numeracy: - Traditional numeracy centers around linear, algorithmic thinking, which is well-suited for predictable and sequential problems. However, in a world characterized by complex systems and rapid change, nonlinear thinking - which involves understanding complex, dynamic systems and their unpredictable interactions - becomes crucial. This kind of thinking is essential for navigating modern challenges in economics, climate change, social dynamics, and more. 3. AI-Centricity as the New Operacy: - Operacy, a term less common in everyday language, relates to the skill of effective operation or action. In the past, this might have meant mechanical skills or basic computer literacy. However, in an era where AI is pervading every aspect of life and work, understanding and leveraging AI becomes a fundamental skill. AI-centricity means not just using AI tools, but thinking in ways that leverage AI's strengths, such as massive data processing, pattern recognition, and predictive analytics. These statements collectively point to a redefinition of what it means to be skilled or educated in the 21st century. They suggest a transition from traditional, often more linear skills to those that are more adaptive, creative, and integrated with advanced technology. This shift reflects the evolving demands of a world where complex problems, technological integration, and rapid change are the norms.”
I feel frustrated. Every time one of your videos ends, I'm all set to hit that like button, only to find out my sneaky hand beat me to it right at the start.
I will not trust google until I try the model myself. They've been way too shady with their previous announcements. They always try to picture themselves as better than OpenAI but haven't been even close to GPT-4's level of utility yet. We'll see.
Gemini Models: Google has mentioned a preview of 'Gemini 1.5 Pro' that uses a 1,000,000 token context window (the amount of text the model looks at to generate a response). However, this refers to its ability to process information, not the size of the model's entire training dataset.
ooh man this gemini is so fast! i hope someone use this to translae language in live with less delay and also someone study communicaion with animals well at start we can test o mouse and monkeys
I'll say, token length limit was annoying when I was writing a novelette. I ended up needing chapter summaries and synopses and then a n^2 algorithm for comparing chapters Do you see any contradiction between {CH 1} and {CH 2} .. now chapter 1 and chapter 2. Now chapter 1 and chapter 3. Although token based billing does still scare me. I dumped my entire story into GPT4 32K and it gave me a useless answer that alluded to revisions it had made, but never actually did. I contemplated requesting a refund for that one request (yes it was big enough that that made sense), but it's a little hard to prove the response was wrong or malfunctioned :( I annoyed me enough to switch to Mistral for the rest of revisions lol
I was scratching my head looking at those comments, almost seems like AI generated comments. But must be some super random low effort generation. Average LLM can actually generate more coherent comments than those
Here's a question for you if you'd like to answer it: If there really is no moat, when do you think Open source models will catch up to the 1M token limit? I hear that Mamba is promising.
Pretty sure they said the 10 million tokens was only accomplished in the lab as an experimental thing to try. I would assume it wasn't very stable when they tried it hence why they won't offer it yet. But they probably pulled off some simple things with it which allowed them to claim they had achieved 10 million tokens. Honestly though, even if they only achieved a 1 million token length and its functional that is still impressive.
Sorry, but as an AI language model, I cannot answer your question. However, you might find insightful perspectives from experts in the field of AI ethics and futurology.
the risks are luddites working them selves up into a frenzy, but that happens every time some thing big is invented, and the world becoming a better place, with higher quality products can be produced cheaper than ever, and the barrier of entry to expressing ones creativity getting ever lower and while i do honestly beleive that, i think there are gonna be a lot of growing pains, and a point in the future will come when the vast majority of humans are unemployable, at that point i believe that either the luddites win and revolt to destroy this new technology, returning the world hopefully to how it was a few years ago, but probably to the stone ages. or the people revolt and society moves to a model where people are paid just for existing and we live lives of luxury with near limitless incredibly cheap products and services as ai and robots do all of our hard work for us. and then we will all need to find hobbies to entertain us without jobs. i think those are the only two options that will work, but the first is almost impossible to happen because getting all of the world to obstain would be impossible when using ai would give your nation a leg up. but who knows maybe something surprising will happen. i just hope the transition will go smoothly. ai could make all our lives much betterm but fear and the neccesarry changes to our society leave a lot of room for revolution and bloodshed
Since our lizard masters only want one thing : mastering their slaves to maintain a comfortable life while avoiding revolts, they had the idea to give us crumbs of those ai technologies until someone makes unsollicited p*rn of public figures, or worse, generate p*dophilia, in order to justify an electronic identity, under the reason to "track the origin of generated content". Progressively, those who wouldn't use that electronic id, would not be able to open bank accounts anymore, and things like that. They are also working on removing paper money. After that, powerful people would be able to freeze the bank accounts of anyone they don't like, with the press of a button. Don't want to let your teen remove her breasts ? Sanction ! Have a doubt about new miraculous pharma products ? Cancelled ! As usual, you take a great technology (the internet, the cloud, ai ...) and lizard people manage to use it the worse way possible : adds infested depressing social medias, tiktok, spying society, ...
i remember discussing about that with my friend years ago, we thought it was so far away from happening. crazy how these ideas that used to represent something beyond our lifespan are now so close to become tangible experiences
@@oowaz I know, right? I was also thinking just a few years ago about how the different personality cores in Glados in Portal 2 were like some kind of futuristic concept. Now it's practically feasible.
@@R2Bl3nd i find it thrilling and terrifying at the same time. i feel like we could be heading towards a future where the wealth disparity will just keep escalating, the super rich will become even more out of touch because they can just simulate genius supermodels to keep them company, who needs real connection when you can live like a god in vr
@@oowaz that sounds like the opposite will happen, where it doesn't matter if you're rich or poor, we will all have access to the same experiences, which will be more enriching and fulfilling than any earthly ones.
AND THIS HAS BEEN THE CODE REPORT..... uhm I mean.... simply a funny move but would be nice if you could shoutout to fireship as well, since videostyle and thumbnail are abviously inspired by him
Mates, this shit is insane. Mind boggling even. I’m not a coder (learning JS really slowly, haha), but I’m like, nah, champs, I gotta cook! I GOTTA COOK an app with my own custom trained ML model, so I can generate web pages with AI. I’m like, this is astounding! So much content is droppin’, and I’m not fast enough to catch all of it. Now I need to figure out how to cook, ahem, I mean use Gemini Ultra 1.5 properly lmao.
when i saw openAI say "Sora doesnt do sexual content" what i heard was "Sora wont do content involving furries or trans ppl becuz it thinks theyr porn"
i haven't been this impressed by anything in a long time, it's thrilling and a scary thing to think about... open AI has amazing tech like this planned and they're just waiting for the right timing to drop
![Breaking Down Meta's Billion Dollar LLM Blueprint [Llama-3.1 Full Breakdown]](http://i.ytimg.com/vi/68_BKkymYs4/mqdefault.jpg)
![Breaking Down Meta's Billion Dollar LLM Blueprint [Llama-3.1 Full Breakdown]](/img/tr.png)
I actually thought this was the code report when i clicked on the video lmao
Same haha
And you know Code Report can’t sue for copyright infringement since he admits that his videos are AI generated. 4d chess move.
Same
Same(kinda)
Sameeee looool
The code report joke was nuts 😂😂😂😂
I had to check I was on the right channel lmao
lol literally@@_zeeblo
Wow. The Ai videos Look insane please make a video on them. Love your Content!
Nice to find another creator who does analysis without silly clickbait
"This has been the code report", you had me, I had to double check who I was watching
I thought this was a fireship video when I saw the thumbnail
Oh my god, the last video looks so good I was confused for a second. Like, it just leapt over the uncanny valley without even trying
When did fireships voice change
Thanks for doing a video on Gemini 1.5 Pro lmao, everyone else was too busy hyping about OpenAI Sora. Which is no doubt mind blowing, but ya know, other news. Looking forward on your video about Sora tho!
bro tried to sneak fireship, and thought we won't notice
I genuniely thought this was Fireship's Code report 💀
Bro think he fireship ong 💀
0:39 This was not the actual performance of Claude 2.1 on the Needle in a Haystack (NIAH) Benchmark. The Original Performance of the benchmark was published by: "Greg Kamradt - Claude 2.1" The original creator of the benchmark.
The one you got was a derivative with actual instructions to remember a specific point in the text. That is not indicative of real-world performance as people won't encode a sentence saying "This is the most important line in the document:....." just to retrieve it again in the model when asked about that line. The Claude 2.1 Test in the video was Flawed!
this has been the code report was too funny
You know progress happens too fast when the next major breakthrough happens before the video even ends 😂
the code report at the end GOT ME. insta subbed
Those videos look better than real life videos
You really had me with the code report thing 😂😂
The quality of the video is amazing it feels like the Classic code report ❤️ Keep it going
This is like watching Fireship slowed down to 0.25. 😂 Great video btw.
Man... this is going so fast I fear for falling behind
We should totally go more faster than that
I feel you brother. The best thing you can do is this:
Imagine life 1, 3, 5, and 10 years from now...
Or at least the general future...
And think: what should I be learning about which will be highly relevant then? What skills will be the most in demand? What am I currently learning that may become irrelevant in the future?
Reflect on this long enough, and consistently, and you shall start to live your life (in skill acquisition and information gathering) in accordance with the future
Hopefully this helps
@@2034-SWE I know that's what I should be doing, at least what's good for my mental health, but I couldn't help but imagine how irrelevant these personal skills would be in face of strong AGI or even ASI. If it really comes to that, I guess everyone would just be forced to create his own meaning...
@@YT-gv3cz I completely understand where you’re coming from. However, after comprehensive analysis of my own, I disagree.
First of all, you must consider the massive amount of compute power required for the SCALE OF DEPLOYMENT you’re referring to. To replace 8 billion humans, you need A LOT of compute power. Why do you think Sam Altman mentioned we need $7 trillion..?
That’s the compute power required just to support general commercial AI usage over next 5-10 years.
Believe me, you’re not becoming “obsolete” anytime soon.
What’s going to happen is everyone who has the digital means to subscribe and use any tool will become superhuman. AI augments and enhances our productivity. It only replaces certain archaic hard-skills (even then, to some extent for now).
------------------
I stumbled across a quote a few months ago by someone in the comment section. I’ll never forget it, and I’m living by it daily (and will for remainder of my life):
“Creativity is the new literacy.
Nonlinearity is the new numeracy.
AI-centricity is the new operacy.”
I’ll reply to this comment with GPT’s analysis of what that means. - Identify this is the future, modify your actions and skill acquisition, and you’ll not just “be fine” like everyone else. You’ll be in demand.
@@YT-gv3cz VisionArchitect GPT:
“Your interpretation is quite accurate. These phrases indeed reflect a shift in the perception of essential skills and competencies in the modern, technologically advanced world. To elaborate further:
1. Creativity as the New Literacy:
- In historical contexts, literacy, the ability to read and write, was a cornerstone of education and a primary indicator of intellectual capability. Today, as automation and artificial intelligence handle more routine tasks, the uniquely human trait of creativity is becoming as fundamental as literacy once was. It signifies a shift from valuing rote knowledge to valuing innovative thinking and the ability to create something new or solve problems in novel ways.
2. Nonlinearity as the New Numeracy:
- Traditional numeracy centers around linear, algorithmic thinking, which is well-suited for predictable and sequential problems. However, in a world characterized by complex systems and rapid change, nonlinear thinking - which involves understanding complex, dynamic systems and their unpredictable interactions - becomes crucial. This kind of thinking is essential for navigating modern challenges in economics, climate change, social dynamics, and more.
3. AI-Centricity as the New Operacy:
- Operacy, a term less common in everyday language, relates to the skill of effective operation or action. In the past, this might have meant mechanical skills or basic computer literacy. However, in an era where AI is pervading every aspect of life and work, understanding and leveraging AI becomes a fundamental skill. AI-centricity means not just using AI tools, but thinking in ways that leverage AI's strengths, such as massive data processing, pattern recognition, and predictive analytics.
These statements collectively point to a redefinition of what it means to be skilled or educated in the 21st century. They suggest a transition from traditional, often more linear skills to those that are more adaptive, creative, and integrated with advanced technology. This shift reflects the evolving demands of a world where complex problems, technological integration, and rapid change are the norms.”
lmao I actually thought I clicked on a fireship video from the thumbnail
It makes a ton of sense that the (probably years) of ai research google has done would focus on working massive datasets just like google search.
ask it for pictures of a knight, a priest, a cowboy and indian
the code report joke was funny lol
wait this is not the fireship channel .
I feel frustrated. Every time one of your videos ends, I'm all set to hit that like button, only to find out my sneaky hand beat me to it right at the start.
imagine google going back to google glass
I will not trust google until I try the model myself.
They've been way too shady with their previous announcements. They always try to picture themselves as better than OpenAI but haven't been even close to GPT-4's level of utility yet.
We'll see.
For a second I thought fireship got a new host, good video though I liked it
3:56 😆
They are improving AI so hard, you can't know what will happen in next month.
Absolutely crazy, i thought we were slowing down for a moment there turns out i second guessed myself.
The code report really !!
0:20 Sir, that is actually a grown man known as Habibi...
Enterprise AI's LLama never ceases to amaze me, truly the best bang for your buck.
The video performance makes sense since they have experience with it thanks to YT
ohhhhhhhhhh right
these sneaky mfs trained their AI on all of these YT videos
Wait no actually
OpenAI doesn't own TH-cam, Google does
@@Cube_Box My man this is a video about Gemini. Not chat GPT. Gemini does belong to google.
@@mercerwing1458 I was referring to Sora by OpenAI mentioned at the end of the vid
Gemini Models: Google has mentioned a preview of 'Gemini 1.5 Pro' that uses a 1,000,000 token context window (the amount of text the model looks at to generate a response). However, this refers to its ability to process information, not the size of the model's entire training dataset.
ooh man this gemini is so fast!
i hope someone use this to translae language in live with less delay and also someone study communicaion with animals well at start we can test o mouse and monkeys
I'll say, token length limit was annoying when I was writing a novelette. I ended up needing chapter summaries and synopses and then a n^2 algorithm for comparing chapters
Do you see any contradiction between {CH 1} and {CH 2} .. now chapter 1 and chapter 2. Now chapter 1 and chapter 3.
Although token based billing does still scare me. I dumped my entire story into GPT4 32K and it gave me a useless answer that alluded to revisions it had made, but never actually did.
I contemplated requesting a refund for that one request (yes it was big enough that that made sense), but it's a little hard to prove the response was wrong or malfunctioned :(
I annoyed me enough to switch to Mistral for the rest of revisions lol
Boy that's a lot of tokens!
4 months after this video, and we have Gemini 1.5 Pro with 2M tokens. It's insane
What are these comments??? Wtf
I was scratching my head looking at those comments, almost seems like AI generated comments. But must be some super random low effort generation. Average LLM can actually generate more coherent comments than those
@@kernsanders3973nah bro let the bots cook
1:38 papuan mentioned🎉🎉🎉
أكثر من رائع ويستاهل خمسه نجوم
Here's a question for you if you'd like to answer it: If there really is no moat, when do you think Open source models will catch up to the 1M token limit? I hear that Mamba is promising.
I have an upcoming video about Mamba that is in the works, so stay tuned:))
it would be nice if they did catch up, but honestly my potato has no chance of ever running it, i can barely run the dumbest of models
the code report bit got me hard 🤣
Sora just dropped.
This is first time the AI vidoes at the end are so good they have to be labeled AI generated.
Remember when Google misrepresented the capabilities of their previous version? I remember.
Three.js = Abstraction hell is such a fabulous way to put it 😂
Straight up stole code report 👍🏻 continue doing that
in case we ever talk to aliens this will be the method to learn their language fast
And in some hours OpenAi pulled out their Sora😂😂
WOW. SHOCKING
😂 bro, for a minute there I thought it was the code report
Pretty sure they said the 10 million tokens was only accomplished in the lab as an experimental thing to try. I would assume it wasn't very stable when they tried it hence why they won't offer it yet. But they probably pulled off some simple things with it which allowed them to claim they had achieved 10 million tokens. Honestly though, even if they only achieved a 1 million token length and its functional that is still impressive.
As AI continues to evolve, what are the potential long-term risks to humanity?
Yes.
I think the one risk would be an unstoppable ai virus that launches nukes because it thinks that's the easiest way to get rewarded
Sorry, but as an AI language model, I cannot answer your question. However, you might find insightful perspectives from experts in the field of AI ethics and futurology.
the risks are luddites working them selves up into a frenzy, but that happens every time some thing big is invented, and the world becoming a better place, with higher quality products can be produced cheaper than ever, and the barrier of entry to expressing ones creativity getting ever lower
and while i do honestly beleive that, i think there are gonna be a lot of growing pains, and a point in the future will come when the vast majority of humans are unemployable, at that point i believe that either the luddites win and revolt to destroy this new technology, returning the world hopefully to how it was a few years ago, but probably to the stone ages. or the people revolt and society moves to a model where people are paid just for existing and we live lives of luxury with near limitless incredibly cheap products and services as ai and robots do all of our hard work for us. and then we will all need to find hobbies to entertain us without jobs.
i think those are the only two options that will work, but the first is almost impossible to happen because getting all of the world to obstain would be impossible when using ai would give your nation a leg up. but who knows maybe something surprising will happen. i just hope the transition will go smoothly. ai could make all our lives much betterm but fear and the neccesarry changes to our society leave a lot of room for revolution and bloodshed
Since our lizard masters only want one thing : mastering their slaves to maintain a comfortable life while avoiding revolts, they had the idea to give us crumbs of those ai technologies until someone makes unsollicited p*rn of public figures, or worse, generate p*dophilia, in order to justify an electronic identity, under the reason to "track the origin of generated content".
Progressively, those who wouldn't use that electronic id, would not be able to open bank accounts anymore, and things like that. They are also working on removing paper money.
After that, powerful people would be able to freeze the bank accounts of anyone they don't like, with the press of a button.
Don't want to let your teen remove her breasts ? Sanction ! Have a doubt about new miraculous pharma products ? Cancelled !
As usual, you take a great technology (the internet, the cloud, ai ...) and lizard people manage to use it the worse way possible : adds infested depressing social medias, tiktok, spying society, ...
Bro I clicked thinking this was a @fireship video
How does it cost to run???
*students preparing all their savings for a subscription* 😂
Poor google, a video about their cool new thing and chat gpt completely stole it from them.
Does it mean that Gemini source code finally fits into Gemini ? Just saying ..
As for Sora .. we're so fucked ..
🔥
With AI generated video we're going to literally have infinite AI generated interactive real-time worlds, like The matrix.
i remember discussing about that with my friend years ago, we thought it was so far away from happening. crazy how these ideas that used to represent something beyond our lifespan are now so close to become tangible experiences
@@oowaz I know, right? I was also thinking just a few years ago about how the different personality cores in Glados in Portal 2 were like some kind of futuristic concept. Now it's practically feasible.
@@R2Bl3nd i find it thrilling and terrifying at the same time. i feel like we could be heading towards a future where the wealth disparity will just keep escalating, the super rich will become even more out of touch because they can just simulate genius supermodels to keep them company, who needs real connection when you can live like a god in vr
@@oowaz that sounds like the opposite will happen, where it doesn't matter if you're rich or poor, we will all have access to the same experiences, which will be more enriching and fulfilling than any earthly ones.
@@R2Bl3nd sounds extremely optimistic
When it's going to be released for the public? Gemini Ultra 1.0 is quite bad somehow
bro thinks he’s fireship 💀
AND THIS HAS BEEN THE CODE REPORT..... uhm I mean....
simply a funny move but would be nice if you could shoutout to fireship as well, since videostyle and thumbnail are abviously inspired by him
lul i thought you are fireship when clicked the vid
This is terrifying because Google owns it
With great power comes great electricity bills.
Sora for real used a regular graphic designer to make this videos, they are not ai, it's just a marketing catch
Bro called Hasbulla a toddler 😂
Nice
pearl or pro?
Mates, this shit is insane. Mind boggling even.
I’m not a coder (learning JS really slowly, haha), but I’m like, nah, champs, I gotta cook! I GOTTA COOK an app with my own custom trained ML model, so I can generate web pages with AI. I’m like, this is astounding!
So much content is droppin’, and I’m not fast enough to catch all of it.
Now I need to figure out how to cook, ahem, I mean use Gemini Ultra 1.5 properly lmao.
contrast these developments with the barbarity being unleashed at the same time
when i saw openAI say "Sora doesnt do sexual content" what i heard was "Sora wont do content involving furries or trans ppl becuz it thinks theyr porn"
You think that's bad, bing's out here flagging black skin as sexual lol
@@tahunuva4254 oh holy geez thats fucked up
Gemini advanced uses what?
bro thinks he is fireship
My mind is blown. That's actually more then just crazy to think about
Are we just supposed to pretend that AI suddenly advanced this fast in 1 year and that they haven't had top secret AIs better than this for 15+ years?
Holy shit
thanks for the AI report lmaoooo. hoping for the sora summary sometimes soon, also with related papers
so what we can't use it.
Saving those tokens for a makeup tutorial?
what is gemini?
Bro, it's google's ai.
It's youtube
and you can litreally have an api key to it for free
1 mil tokens, not 10 mil
I thought it was fireship code report 😢
I have been scamed , my day is ruined 😞
this isnt the code report? XDXD
dont be a fireship clone lol
bro what the fuck is that text to video????
No...
i haven't been this impressed by anything in a long time, it's thrilling and a scary thing to think about... open AI has amazing tech like this planned and they're just waiting for the right timing to drop
mfw AI ripoffs of code report but it’s better
No argument at all here, don’t watch !
didnt understand anything frfr
Do your own thing. This is embarassing.
code report clickbait?! aww man
1M - not 10M
Excerpt from the Google blog on this "In our research, we’ve also successfully tested up to 10 million tokens."
Why you need 10M tokens?😂😂, you writing a book per prompt?
Massive codebases, in-context learning (look at it learning kalamang). It's actually a really big deal.
too bad it's Google