Google DeepMind's JEST AI Learns 13x Faster 🚀 & SenseTime's AI Beats GPT-4.0! 😱 | AI Revolution - video Dailymotion

Ai Revolution

🚨 AI evolution is happening at lightning speed! Introducing Google DeepMind's JEST AI — a revolutionary model that learns 13x faster than previous systems! 🤖 SenseTime’s new AI is out and beats GPT-4.0 in performance, setting a new standard for what AI can achieve in 2025. From advanced learning capabilities to incredible real-world applications, these AI breakthroughs are changing the game.  🔍 Stay tuned for the AI Revolution that’s just beginning!  ✅ Like, Share & Subscribe to get more updates on the future of artificial intelligence! #JESTAI #DeepMind #SenseTimeAI #GPT4Beaten #AIRevolution #ArtificialIntelligence #MachineLearning #AI2025 #AIProgress #AIAdvancements #DeepLearning #TechInnovation #AIInTech #NextGenAI #AIandTech #FasterAI #AIModels #AIforEveryone #TechBreakthrough #FutureOfAI

Transcript

00:00All right super exciting stuff coming out of Google's DeepMind. So they've come up

00:06with a new way to train AI models that's not only faster but also way more

00:10efficient. This method is called Jest which stands for joint example selection

00:15and trust me it's pretty revolutionary so let's talk about it. All right so if

00:18you know anything about AI you know that training these models usually takes a

00:23ton of time and a lot of computing power. It's like trying to teach a child every

00:27subject in school one by one and it's super energy intensive. For example

00:32systems like chat GPT guzzle a lot of power and water to keep those massive

00:36data centers cool. There are even comparisons to the energy demands of

00:40Bitcoin mining. Yeah it's that intense. But here's where Jest comes into play. The

00:45researchers at DeepMind have found a way to make this whole process way more

00:48efficient. How efficient? Like up to 13 times fewer iterations and 10 times less

00:53computation. That's a game changer for both speed and energy consumption. So how

00:58does Jest work? Instead of picking individual pieces of data to train on

01:02which is what's usually done, Jest looks at groups or batches of data and picks the

01:06best ones that work well together. Think of trying to learn multiple languages.

01:10Instead of learning English then German then Norwegian separately you might find it

01:14more effective to study them in a way where what you learn in one language helps

01:19you understand the others. That's essentially what Jest does but with data.

01:22Here's a bit more on the technical side but I'll keep it simple. The method

01:26Jest uses is called multimodal contrastive learning. This means it looks at

01:30different types of data like images and text together and it identifies

01:35dependencies between them. This approach significantly speeds up the learning

01:39process because the AI can learn from these well-matched data batches much more

01:43efficiently. What makes Jest even cooler is that it starts with something

01:47called a pre-trained reference model. This model helps steer the AI towards the

01:51high-quality well-curated data making the training even more efficient. Now the

01:56study showed that using Jest they could achieve the same performance as

01:59traditional methods with up to 13 times fewer iterations. Also it requires 10

02:04times less computational power. So in terms of energy consumption that's a

02:09massive reduction which is great news for the environment. Now the Jest

02:12research comes not a moment too soon as the tech industry and world

02:16governments are beginning discussions on artificial intelligence's extreme power

02:20demands. To give you an idea AI workloads consumed about 4.3 gigawatts in 2023 which

02:27is almost equivalent to the annual power consumption of the entire nation of

02:30Cyprus and this trend is only going up. A single chat GPT request currently costs 10

02:36times more in power than a Google search. It's estimated that AI might take up a

02:40quarter of the United States power grid by 2030. Therefore any method like Jest that

02:45can drastically cut down on the power required for training AI models is incredibly

02:49significant potentially easing the environmental impact and the financial

02:53costs associated with AI advancements. Furthermore the success of the Jest

02:57method hinges on the quality of the training data. The bootstrapping technique

03:01utilized by Jest requires a highly curated initial data set to function

03:05effectively. This means that while Jest shows great promise for large-scale

03:09well-funded AI projects it might pose challenges for hobbyists or smaller

03:14developers who lack the resources to compile high-grade training data.

03:18Essentially this method emphasizes the old adage of garbage in garbage out

03:22underlining the necessity for top-tier data to achieve optimal results. If

03:27implemented widely however Jest could revolutionize how major players in the AI

03:32industry approach model training possibly reducing costs and power

03:35consumption while accelerating development timelines. Another neat thing

03:39they found is something they call data quality bootstrapping. It sounds fancy but

03:43it's pretty straightforward. It means using a small highly curated set of data to

03:47guide the training on a much larger messier set of data. So even if you start

03:51with a tiny really good data set Jest can help you scale up to train on a vast

03:56amount of data efficiently without losing quality. This is a big deal because

04:01high quality data sets are usually hard and expensive to create. Now this method

04:06isn't just limited to one type of data or task. It works across various

04:10benchmarks which means it's versatile. For example they tested it on a data set

04:14called WebLI which is a large-scale collection of images and text from the

04:18web and the results showed remarkable improvements in both learning speed and

04:22resource efficiency. So if Jest proves effective on a larger scale it could

04:26drastically reduce the power needed to train AI models. This means we could

04:30develop more powerful AI tools with the same resources we use now or use fewer

04:34resources to create new models. Essentially we're looking at faster

04:38advancements in technology, less strain on our energy resources, and a smaller

04:42environmental footprint. It's a win-win-win. Alright now you know the big AI

04:46companies in China have been making some major claims about their large

04:49language models at this huge AI conference happened recently. It's like a

04:53competition to see who can boast the most impressive advances. You know what I mean?

04:57So there's this company called SenseTime. They're kind of big shots in the AI

05:01game over there. They just unveiled the latest versions of their SenseNova

05:05LLMs and get this they're claiming the new 5.5 model is like 30% better than the

05:10previous one from just a few months ago. Wild right? The CEO, Xu Li, he was on stage

05:15hyping it up big time saying the key to success in Chinese AI is basically

05:19building these high-level logic systems using synthetic data for different

05:23industries. Xu claimed that this 5.5 model supposedly outperformed GPT-40, the

05:30latest and greatest from OpenAI, in five out of eight key metrics according to

05:35some benchmarking data from a platform called OpenCompass. I mean take that

05:39with a grain of salt for sure but still that's a pretty big flex. Flexing on OpenAI

05:43like that is a gutsy move. Now despite making these huge claims about their

05:47models performance, SenseTime's stock price actually tanked like 16% on the same day

05:53in Hong Kong. Crazy reversal, right? You'd think investors would be stoked about

05:57supposedly having a model that can take down GPT-4 but I guess the market had

06:02other ideas. And it's not just SenseTime making noise at this conference. Oh no,

06:06Alibaba's cloud division, the folks behind the Tongyi-Chanwen LLMs, they were

06:10bragging about getting like millions of new downloads and a surge in customers for

06:15their AI platform in just the last couple of months. The CTO, Joe Jingren, he was

06:20really pushing this whole open source angle hard saying their models are now

06:24fully open source and closing the gap with closed source ones from the big

06:28Western players. Smart move if you ask me, makes it more accessible and gets more

06:32eyes on their tech. But here's the wildest part of it all. There's this AI

06:36startup CEO, Yan Junji from a company called Minimax and he straight up

06:41predicted that in the future there will only be like five companies making these

06:45huge LLMs globally. Now that's a hot take if I've ever heard one, right? I mean,

06:49could you imagine that kind of consolidation happening? Just five

06:52players controlling the entire large language model space worldwide? Insane to

06:56think about. Of course, Yan didn't actually say how many of those five would

06:59be Chinese companies, so we'll have to wait and see on that one. But the

07:03implication is clear. He thinks there's a real chance for Chinese firms to be among

07:07that tiny elite group someday. Whether that's realistic or not, I'll leave that up to

07:11you to decide. Oh, and back to SenseTime for a sec. They also showed off a couple

07:15of other new models at this big Shanghai conference. The first is called 5.0 and

07:20it's this multimodal model that can combine different data types like text,

07:24images, video, all that jazz into one seamless experience. And they did a live

07:29demo with the 5.0 that was low-key mind-blowing, not gonna lie. So there's

07:33this SenseTime employee on stage, right? And he takes a video of himself just

07:37standing there, seemingly nothing special. But then the 5.0 model instantly

07:41identifies that he's attending this World AI Conference thing just from

07:45analyzing his staff badge and t-shirt in the video. It's wild, but it gets even

07:50crazier. The model could then answer follow-up questions about what the

07:53conference is for and all the deets, like it achieved true understanding from

07:57just that little video clip. I'm talking next-level multimodal AI capabilities on

08:02display here, folks. So yeah, between the bold performance claims, the open-source

08:07pushes, the predictions of an LLM oligopoly, and the awesome tech demos,

08:11it's been an absolute whirlwind of big moves and even bigger ambitions from the

08:16Chinese AI giants at this conference. And that's not even getting into what

08:20companies like Baidu and Tencent were unveiling. It's definitely an exciting time

08:23in the AI space, that's for sure. And it'll be fascinating to see how all these

08:28moonshot plays actually pan out in the real world. Can the Chinese models really take

08:33down the Western juggernauts? Will we see a small group of ultra-dominant LLM

08:38providers emerge globally? So many intriguing possibilities. One thing's

08:42for certain though, these Chinese tech companies are clearly not messing around.

08:46They've got their sights set on AI supremacy and they're going all-in to

08:50make it happen, whether that's through cutting-edge R&D, savvy business

08:53strategies, or sheer bravado and ambition. All right, if you found this

08:57interesting and want to stay updated on more AI insights like this, don't forget to

09:01subscribe to the channel. Thanks for watching, and I'll see you in the next one.

Google DeepMind's JEST AI Learns 13x Faster 🚀 & SenseTime's AI Beats GPT-4.0! 😱 | AI Revolution

Category

Transcript

Recommended