# AI NEWS : China Catches Up To OpenAI, Google Beats Everyone, Robots Get Too Realistic.. And More

## Метаданные

- **Канал:** TheAIGRID
- **YouTube:** https://www.youtube.com/watch?v=ZHTn6Cy42Wk
- **Дата:** 29.03.2025
- **Длительность:** 34:55
- **Просмотры:** 21,933

## Описание

Join my AI Academy - https://www.skool.com/postagiprepardness 
🐤 Follow Me on Twitter https://twitter.com/TheAiGrid
🌐 Checkout My website - https://theaigrid.com/

⏱️ Timestamps:

00:00 – LG’s Surprise Model
00:48 – DeepSeek Breakthrough Moment
02:17 – Post-Training Magic
03:26 – GPT-4 Everywhere
04:39 – Tencent’s Bold Move
06:03 – Vibe Test Warning
07:05 – Baidu Fires Back
09:32 – Google’s Power Play
11:04 – Simple Bench Dominance
12:34 – Shocking Score Jump
13:15 – Vision Model Twist
13:28 – Everything Goes Autonomous
14:35 – Omniverse Training Loop
15:32 – Enter: Project Newton
16:50 – Newton Goes Open
17:38 – Realism Hits Robotics
19:17 – Unitree G1 Shocks
20:31 – Atlas Operates Cameras
22:19 – 5,000 Robots Incoming
24:08 – Demand Already Overwhelming
25:03 – Home vs Workforce
26:32 – Adobe’s Secret Agents
27:47 – Cancer Detection AI
29:09 – AGI Benchmark Arrives
30:12 – ChatGPT Voice Upgrade
32:14 – Model Fatigue Setting
32:30 – Zuckerberg's AI Vision
34:13 – $10K AI Plan

Links From Todays Video:
https://x.com/TXhunyuan/status/1903121005809373386  (Hyunyan t1)
 https://x.com/ai_for_success/status/1901149459826045223 (eRNITE 4.5)
https://x.com/Tolat743/status/1902476884413444579 (AI detects cancer)
Elon robots ( https://x.com/TheHumanoidHub/status/1903129234991992944) 
https://x.com/kimmonismus/status/1903381889957998881 (Boston Atlas)
https://x.com/mikeknoop/status/1904269314099978344 (Arc AGI)
https://x.com/shaunralston/status/1904278721328251250 (AVM)
https://business.adobe.com/products/experience-platform/agent-orchestrator.html

Welcome to my channel where i bring you the latest breakthroughs in AI. From deep learning to robotics, i cover it all. My videos offer valuable insights and perspectives that will expand your knowledge and understanding of this rapidly evolving field. Be sure to subscribe and stay updated on my latest videos.

Was there anything i missed?

(For Business Enquiries)  contact@theaigrid.com

Music Used

LEMMiNO - Cipher
https://www.youtube.com/watch?v=b0q5PR1xpA0
CC BY-SA 4.0
LEMMiNO - Encounters
https://www.youtube.com/watch?v=xdwWCl_5x2s

#LLM #Largelanguagemodel #chatgpt
#AI
#ArtificialIntelligence
#MachineLearning
#DeepLearning
#NeuralNetworks
#Robotics
#DataScience

## Содержание

### [0:00](https://www.youtube.com/watch?v=ZHTn6Cy42Wk) LG’s Surprise Model

okay so there is a lot of AI news but one of the first stories that completely missed me was L's new AI model that's right LG produced a model now I'm not just covering this because LG produced an AI model but if you actually take a look at what the model was able to do it was so surprising the model was 32 billion parameters and actually performed number one on the math benchmark I think that is something that is absolutely outstanding and actually found this out because I was looking at an interview with Sam Alman where they were discussing the fact that there are so many new AI models and this was potentially one of those new models now let's take a look at these benchmarks CU they were not messing around you can see that the new AI model from LG the 32

### [0:48](https://www.youtube.com/watch?v=ZHTn6Cy42Wk&t=48s) DeepSeek Breakthrough Moment

billion parameter model now I haven't tested it yet but take a look at these crazy benchmarks we can see that it's got 32 billion parameters there's also a version with 7. 8 and a version with 2. 5 and overall we can see that all versions seem to perform very well for the comparative size comparing it to deep seek R1 which is completely incredible we can see that it manages to do very well on the math benchmark on the Amy Benchmark on the cat Benchmark I do hope that this model isn't just purely trained on benchmarks I haven't really been able to see any inference provider who's been able to provide inference for this model but taking a look at what I can see here this is some seriously impressive stuff 32 billion parameters and it's performing better than these models this is something out of I don't even know what it's out of but this was something that was really surprising now the honest truth is that the AI models are getting more and more impressive every single week we actually had deep seek guys deep seek is what we had can you believe we had deep seek V3 and deep seek V3 was remarkably impressive this is a model that serious seriously took me by surprise because not only did I not expect them to release an update I didn't that was so much better in terms of the performance that it would leave the AI industry wondering what they are going to release next when

### [2:17](https://www.youtube.com/watch?v=ZHTn6Cy42Wk&t=137s) Post-Training Magic

we actually really take a look at what was released here we can see that the model managed to perform better than these other ones now I don't really want to hamper on about Benchmark this and Benchmark that but one of the most incredible things that I think about is that I'm not sure what techniques they used I'm not sure maybe it was the posttraining or the fine tuning whatever it is that they managed to do to the model of course they didn't completely train a new model but what I did see from them was I did see the fact that they did manage to become the top reasoning model now when we take a look at this like this was something that truly did surprise me I do apologize for the quality of this but I'll explain it to you so you don't have to pay attention the artificial analysis index by open weights showed that deep seek V3 basically jumped to the number one spot in terms of non- reasoning models and so it becoming one of the best models that is not only you know open source but just simply crushing it in terms of the benchmarks I think it's a remarkable level of ability that they've managed to achieve in such a short amount of time and this was

### [3:26](https://www.youtube.com/watch?v=ZHTn6Cy42Wk&t=206s) GPT-4 Everywhere

something that truly did shock me because never did I ever think that an open weights model would jump to that spot now it wasn't the only model that was released honestly this week of AI was absolutely incredible and I think you guys are going to understand why many individuals were essentially saying that the models have somewhat become commoditized in the sense that we now have too many models to choose from that it is no longer a selling point to have your model be on the level of GPT 40 because we've got two many models that are at that level I mean let's just take a look you know we got deep seek V3 we've got grock V3 we've got GPT 4. 5 we've got Gemini 2 Claude Sonic 3. 7 quen 2. 5 Max llama 3 I mean the list continues to go on and when we look at this guys you guys might be even getting bored at this point we've got Hunan T1 model okay and this is another model that was released and so Hunan T1 is 10 cents Advanced large language model that combines mber Transformer and mixture of experts to deliver exceptional performance among many different areas

### [4:39](https://www.youtube.com/watch?v=ZHTn6Cy42Wk&t=279s) Tencent’s Bold Move

now what's crazy about this is once again I think I'm getting Benchmark for Te because you know hun1 T1 actually performed better than deep seeks R1 in certain areas you can see on the MML Pro in the reasoning areas in the math areas in the code areas on the Chinese areas I mean honestly it's getting to a point now where I'm starting to think that maybe there are too many AI models now of course for the consumer I think that this is a good thing because you know you have tons of models to choose from if you know one service provider is down you can always go off to another so this is something that is definitely good for the consumer because we are pretty much spoiled for choice at this moment in time now whether or not people will actually use the model that is going to be a different question because one of the things I will say to you guys and one of the reasons I will always advocate for at least testing the model is because you never truly know whether or not a model is good for your use case unless you've Vibe tested the model now what do I mean by that well take Hunan T1 as an example and I do apologize if I'm pronouncing it wrong but just focus on the message here the point I'm trying to make is that you need to focus on the vibe so whatever it is that you specialize in maybe you specialize in marketing maybe you speci specialize in designing stuff maybe you specialize in writing certain emails you need to you

### [6:03](https://www.youtube.com/watch?v=ZHTn6Cy42Wk&t=363s) Vibe Test Warning

know have a set number of prompts and then test the model on those prompts to see if it can just naturally understand where you're going on because one thing that I don't think people don't realize is that every single model is trained on a specific data set and those data sets shape the worldview and the responses of said model and sometimes you get really lucky and it's shaped in a way that is for your domain and other times you get a little bit unlucky and it just really doesn't understand anything you're talking about and this is something that you know I have with image generator models sometimes I want them to do a specific game or a specific concept that only a few May understand and I think it's always important just to quickly test these models on a few prompts that way you can never miss out if a model is released and it might just be part of your daily workflow and this is actually why I use several different models in my daily workflows and I'm not dismissive of these new models now this is probably the last one I'm going to announc but once again Buu just unveiled Ernie 4. 5 nx1 last week they released two models

### [7:05](https://www.youtube.com/watch?v=ZHTn6Cy42Wk&t=425s) Baidu Fires Back

so they released the base model earning 4. 5 and I do think that wasn't like a hint to GPT 4. 5 but I do think it was a little bit cheeky for them to name their model Ernie 4. 5 and you know X1 cuz open AI they've named their GPT 4. 5 and X1 and you can see right here it says as a deep thinking model with multimodal capab abilities Ernie X1 delivers performance on par with deep seek R1 at only half the price meanwhile Ernie 4. 5 is our latest Foundation model a new generative native multimo model and so they made this chatbot you know free and of course this is the benchmarks you can see it does perform better than GPT 40 of course I will say at the time of this open AI does continue to rapidly update their GPT photo model they do make small changes on the end so I do think that potentially you know even though when these models are released they might be initially better than GPT 40 but over time as those updates get passed on I do think that open a probably will always retain that leadership because they always make small changes that just make the model a lot better we can see here that multiple articles said that Ernie X1 delivers performance on par with deep seek R1 at only half the price you can see here that this was one of the screenshots I saw floating around on Twitter it was talking about how you know gbt 4. 5 and I don't know why they compared gbt 4. 5 I do think that this was just a bit disingenuous because gbt 5 is not deliberately an expensive model but opening ey I've clearly stated that look gbt 4. 5 is our most expensive model and we know this so like don't flame us for this but we're just going to release the model anyways and comparing it against Earnie 4. 5 saying that this one's 55 cents and this one is 2. 2 cents I mean I do want to say it is a bit disingenuous but nonetheless I will say the point probably does stand in the fact that you know these models are significantly cheaper than the other offerings but will that cheapness be offset by the model's ability to be as smart as them I don't know that's going to be something that you have to discuss and of course BYU have spoken about open sourcing the model series within the next few months so we spoke a lot about different AI model releases but it wouldn't be complete if we didn't mention Google's new Gemini 2. 5 Pro which is now the king of all Kings so

### [9:32](https://www.youtube.com/watch?v=ZHTn6Cy42Wk&t=572s) Google’s Power Play

whilst yes there's this open source model that model trying to release this and that and I mean it all gets very confusing when you try to keep up but you have to understand Google decided to fire another shot into the AI industry and this one I think it landed dead center because the benchmarks I mean when we take a look at these benchmarks they're not that impressive but I think the standout one for me was Humanity's last exam now you might be thinking why is that why are we not focusing on the mathematics the gpq the Live code bench I mean those are all very impressive but none of them are state-of-the-art and honestly I think all of the benchmarks we're reaching that point of real saturation so the reason I'm really excited about this one which is the of course Humanities last exam is because this one is designed to push AIS to their limits it's really you know designed to test their knowledge on Specialized things it's really hard it's really difficult and Gemini 2. 5 Pro actually outperforms 03 mini which means by far I think they're taking the lead now this you know AI also performed very well on simple bench and you can see right here that Gemini 2. 5 Pro takes the first spot and this is probably one of the best indicators how good a model is and the reason I say that is because simple bench isn't like other benchmarks it doesn't test whether a model can understand how to answer a question that is purely based on what it's used to it

### [11:04](https://www.youtube.com/watch?v=ZHTn6Cy42Wk&t=664s) Simple Bench Dominance

tries to answer the question within the question which means that the model has a deeper level of understanding and because of that it's going to be able to answer your questions with a smartness that other models just can't have and this is why I do believe that a lot of people always enjoyed Claude because this model is second and third in this Benchmark and all of those three models rank pretty highly you know above the other models and I do believe that understanding Common Sense reasoning which is essentially what you could call this test is a really important part of how humans reason and it's just something that I do think LMS kind of get tripped up on because a lot of the questions that you know they're trained on and they answer are very basic it's like okay you have this math problem it solves it in a specific way it comes to an answer and a lot of these questions do require you to kind of think about stuff in the physical world and answer the question in a way that you know previous llms just really struggled with I mean I remember the first time this thing came out a lot of these models like GPT 40 mini dsek V3 grock 2 a lot of them just didn't perform that well but overtime we're seeing that it's now over 50% which is a really impressive score now there were other things that really impressed me in the score such as the fact that it actually had the largest score jump ever on the LMS Ys Arena Benchmark with over 40 ELO points gained which means that this is the model that is good but not just good by a little bit but good by quite a lot so that was one that was

### [12:34](https://www.youtube.com/watch?v=ZHTn6Cy42Wk&t=754s) Shocking Score Jump

super surprising to me because you know Google usually they usually take the top spot for a little bit but this time they took it for quite some time now of course there was also the fact that this was number one on the vision Arena is highly underrated if you ask me I think this is probably one of the most underrated benchmarks because most people aren't sat analyzing images all day but you have to understand that this has many broad implications if AI can understand images then that means it can basically see and what happens when AI can see we unlock a variety of different use cases so this is probably going to come about in many different ways that we will see in a few weeks to come

### [13:15](https://www.youtube.com/watch?v=ZHTn6Cy42Wk&t=795s) Vision Model Twist

but of course as it's just being rolled out it's probably something that we haven't truly understood yet now of course we had Nvidia there was the GTC event and this video basically summarizes everything you may have

### [13:28](https://www.youtube.com/watch?v=ZHTn6Cy42Wk&t=808s) Everything Goes Autonomous

missed everything that moves will be autonomous physical AI will embody robots of every kind in every industry three computers built by Nvidia enable a continuous loop of robot AI simulation training testing and Real World Experience training robots requires huge volumes of data Internet scale data provides common sense and reasoning but robots need action and control data which is expensive to capture with blueprints built on Nvidia Omniverse and Cosmos developers can generate massive amounts of diverse synthetic data for training robot policies first in Omniverse developers aggregate real world sensor or demonstration data according to their different domains robots and tasks then use Omniverse to condition Cosmos multiplying the original captures into large volumes of photoal diverse data developers use Isaac lab to Post train the robot policies with the augmented data set and let the robots learn new skills by cloning behaviors through imitation learning or through trial and error with reinforcement learning AI feedback

### [14:35](https://www.youtube.com/watch?v=ZHTn6Cy42Wk&t=875s) Omniverse Training Loop

practicing in a lab is different than the real world new policies need to be field tested developers use Omniverse for software and Hardware in the loop testing simulating the policies in a digital twin with real world environmental Dynamics with domain randomization physics feedback and High Fidelity sensor simulation real world operations require multiple robots to work together Mega an Omniverse blueprint lets developers test fleets of post-train policies at scale here foxc contests heterogeneous robots in a virtual Nvidia Blackwell production facility as the robot brains execute their missions they perceive the results of their actions through sensor simulation then plan their next action Mega lets developers test many robot policies enabling the robots to work as a system whether for spatial reasoning navigation Mobility or dexterity amazing

### [15:32](https://www.youtube.com/watch?v=ZHTn6Cy42Wk&t=932s) Enter: Project Newton

things are born in simulation now Nvidia also introduced Newton today we're announcing something really special it is a partnership of three companies Deep Mind Disney research and Nvidia and we call it Newton now nvidia's Newton is a physics engine that helps robots learn and interact with the world in a more realistic way it's basically a super powerful simulator that lets robots practice in a virtual environment before they actually try it out in real life now this was developed by Nvidia Google deepmind and Disney research and Newton uses you know Advanced Computers to power and mimic real world physics quite like you know gravity friction and this is all so robots can actually learn faster and better so that means robots can train to do complex tasks like picking up objects navigating spaces without the risks of breaking things which is you know pretty bad because these robots are quite expensive and Newton is actually open source which is good because now it means that anyone can use it and anyone can improve it which is huge because as you know open source projects usually get a lot of love and support now

### [16:50](https://www.youtube.com/watch?v=ZHTn6Cy42Wk&t=1010s) Newton Goes Open

nvidia's Newton you know it improves robotics because this is built on you know nvidia's GPU accelerated warp framework which enables faster and more accurate physics modeling including ridging body Dynamics soft body interactions and friction mechanics now this helps robots learn complex tasks like manipulation navigation and it basically reduces the simulation to reality Gap where virtual training fails to translate to real world performance so I've actually seen a few demos of sim toore going really well recently so I do believe that you know probably within the next few years you guys are going to see a lot more robotics demos that are going to be super incredible now speaking of Robotics demos there were some continued innovations that weren't actually related to this but were really interesting so one of them was

### [17:38](https://www.youtube.com/watch?v=ZHTn6Cy42Wk&t=1058s) Realism Hits Robotics

engine AI so this was uh you can probably hear the robot in the background but uh this was a robot that you know they use simulation and reinforcement learning to basically make this robot run like an actual human so I would say that this is probably the most realistic robot that I've ever seen run and yeah I've seen a lot of robots run and none of them run like this so imagine what's going to happen in 10 years we might even have Robots full on sprinting quite like the Olympic sprinters and even sprinting probably faster than us at more consistency so there might be a robot Olympic Games I know they're doing a half marathon very soon but it's all completely impressive and the reason I made a I even made a video on this company because this company has had you know continued Innovation after continued Innovation and the company has come out and they've really moved very quickly so the fact that this video right here you know even me someone who covers this stuff okay I was wondering is this CGI like is this CGI I mean the robot looks so human it's moving so quickly it's one time speed and the crazy thing about this is that like if you've ever paid attention to Robotics and you've seen the demos what you'll usually see is robots in five times speed because they move so slow so to see a robot go from that to moving one time speed as quick as a if not quicker in just like maybe one year it's kind of mindblowing and you know your brain has a really hard time of wrapping you know that around yourself but of course they released the behind the scenes and you can see it is actually super realistic so honestly cannot wait for this company's next update because they are killing it in terms of the realism and I feel like we

### [19:17](https://www.youtube.com/watch?v=ZHTn6Cy42Wk&t=1157s) Unitree G1 Shocks

are kind of already living in a cyberpunk world now also we did have the unitri G1 they also released an update where you know you can see that this robot can just simply just get up just out of nowhere it can like kick flip up and stand up and I mean guys this is just like it's still blowing my mind and I'm trying to be professional in my commentary but I think it's quite hard to maintain that level of professionalism when I'm seeing a robot that only a year ago was moving in a way that was you know pretty laughable and completely robotic and now it's moving in a way that is quite scary and many can't even believe that it is real so this is something that I think we probably should start getting used to because I think once these robots are trained on even more data there's a lot more reinforcement learning and you can see that person just tried to kick a robot over and it managed to have incredible stability I would argue if someone kicked me like that I'd probably be face planting on the floor and I would argue I'm a very active person so this is something that you know reinforcement learning enables the robots to do and it's super incredible now here's something I do want to say a lot of people I've seen this a lot of tweets have been like okay the robots can dance they can fight they can do kung fu but are they actually going to have any real world use cases

### [20:31](https://www.youtube.com/watch?v=ZHTn6Cy42Wk&t=1231s) Atlas Operates Cameras

well I've got something for that too well take a look at this guys cuz Boston Dynamics they've actually released a video where they showcase the atlas robot actually operating cameras inside of a film studio this thing is absolutely Bonkers there's no other word to describe it like when I saw this I was like okay we're clearly going to have humanoid robots doing a bunch of different things now film companies they probably have the millions of dollars to you know do this because they have millions of dollars of film budgets but is the Indie filmmaker going to be hiring a Boston Dynamics robot to be filming the car I don't know honestly didn't think this would happen so early on and in this short video they talk about the fact that you know they use these tool robots they use them in you know different ways but Boston Dynamics Atlas is just able to balance it's able to hold the camera in different ways and it's able to hold it for hours in ways that humans never could I mean the robot just flexes on everyone by doing a back flip there and just manages to balance itself so crazy stuff you need to actually listen to this inside a new use case you need large amounts of training data and sometimes that training data doesn't exist or is difficult to get a hold of and then you need to generate it synthetically Atlas can lift heavy objects in the order of 20 kg hold it in an awkward position maintain its balance bring it somewhere else Atlas will fill a gap of repeatable shots and long repeatable shots the other robots we use obviously they're very heavy and very big or they have to work on tracks Atlas I think will be able to go into a lot of different location I've really been enjoying working with Sport and Atlas I think it's been really delightful being able to see how quickly they are part of a set and team and they enhance what we do it's extremely important to have this space to be able to bring engineering and creativity

### [22:19](https://www.youtube.com/watch?v=ZHTn6Cy42Wk&t=1339s) 5,000 Robots Incoming

together and so what do you think about that well Elon Musk he's talking about you know producing 500 I think it's 5,000 I I left a zero there so yeah he's talking about producing 5,000 robots per year which is pretty crazy um so this year we hopefully will be able to make about um 5,000 Optimus robots uh we technically robots uh we're technically uh we're aiming for enough parts to make 10,000 maybe 12,000 um but since it's a totally new product with totally new you know like everything is totally new um I I'll say like we're succeeding if we get to half of the 10 you know half of the 10,000 now but even 5,000 robots that that's the size of a Roman legion FYI which is like a little scary thought like a whole Legion of robots i' be like whoa okay um but I think we I think we'll literally build a legion at least one Legion of robots uh this year um and then Pro probably 10 Legions next year and I it's kind of a cool unit you know um units of Legion um so probably 50,000 is next year um and then it's probably ready for to I'm hopefully ready for Optimus to be used outside of Tesla controlled environment maybe around the middle of next year second half of next year sometime um so that's I think yeah sounds about right probably second half of next year is is when they'll be available and then we will um offer Optimus robots first to uh Tesla employees uh so you guys get the priority so the new Optimus 22 degree of Freedom hand and forarm is now in production now

### [24:08](https://www.youtube.com/watch?v=ZHTn6Cy42Wk&t=1448s) Demand Already Overwhelming

of course if you thought you know Elon Musk producing 5,000 robots a year was crazy well Brett atock actually talks about the robot demand and this was something that truly surprised me because he talks about the fact that like if they had a 100,000 robots today they could actually you know deliver those to companies because the demand is there like companies are asking them for these robots which means guys the economy is going to change in a really big way because I don't think that the demand would be there just yet but Brett at Cook is basically saying that look the demand is yeah we are behind on schedule not really behind but like you know we want to you know get these robots as quickly as possible yeah so we have like two tracks we have a like Workforce track which is like um and then we have the home track like the what most people don't get is like the workforce is the big business like it's half of GDP we can charge meaningfully more per robot than the house um and it's also easy the the things that the robot does is just like the same things almost on repeat uh the home is like the Wild West

### [25:03](https://www.youtube.com/watch?v=ZHTn6Cy42Wk&t=1503s) Home vs Workforce

it's like extremely hard uh we have a huge safety area of like not falling on like any human or hurting people there's a semantic and safety of like not knocking over the candle and burning the house down there's like the home is just like vastly harder like um maybe in self-driving it's like driving on the highway is like workfor us for us and driving into the city is like the home it's just like unbelievably difficult um between our two first commercial customers which are very large businesses uh we have demand like if we had 100,000 robots today that all worked they would take 100,000 robots today and so and then we have like 50 customers that I could sign by the weekend that are all Fortune 100 companies that we've like literally visited we know them we just like we can't I've you know done a bunch of meetings today at lunch everybody's like what do you think about helping out here in healthcare con all sound great like we're just like bombarded with the amount of demand here you're thinking about like the workforce you have like a certain number of supply of humans it's literally going down demographically the baby are retiring so you have less humans in the workforce there's labor pains everywhere and you know like there's a lot of job shortages like we can so anyway we see like just unbounded demand I think we could ship a million robots to this month if we like had them all working and they're ready to go and so that was something that I was just like wow like that demand is crazy so I mean I probably need to invest in some of those robotics companies now in terms of other things that were really cool Adobe introduced insane levels of Agents they introduced 10 new AI agents 10 purpose built AI agents with the Adobe experience platform orchestrated now they

### [26:32](https://www.youtube.com/watch?v=ZHTn6Cy42Wk&t=1592s) Adobe’s Secret Agents

introduced these AI capabilities at the Adobe Summit 2025 and I do want to say thanks to Adobe because they actually did invite me out there and it was an amazing experience to actually see all of the AI technology in person and this Suite of AI agents designed to revolutionize marketing workflows and customer experiences are honestly top-notch like I've been doing some serious Research into you know research papers and what companies are doing what people are really using and I want to say that Adobe are definitely ahead of the curb so of course if you're a business owner you should definitely check out what Adobe are doing because it's probably going to become mainstream probably like a year and a half from now but the things they're working on like the workflow agents like the content production agents the the audiences I mean the journey agents it's just truly mind-blowing to see how far ahead they are and it basically just enables you know businesses to build manage coordinate Agents from both Adobe and third party systems and it's pretty crazy about how they've managed to demo this because I didn't think the agents were that good like I've seen a lot of agents that were gimmicky but this was one that was really impressive in terms of actually doing stuff that organizations actually need now in terms of other stuff for ai was actually able to detect a form of cancer a new gamechanging AI ECG MLP identifies

### [27:47](https://www.youtube.com/watch?v=ZHTn6Cy42Wk&t=1667s) Cancer Detection AI

endometrial cancer with 99. 26% and it also works across choral breast and oral cancers with 97% accuracy and I can't wait for you know AI to be used in healthcare so we can really speed up the detection of certain issues and of course the treatment and Diagnostics now of course there was also some information from samman on that future and he talks about basically the fact that look this thing is going to outperformance and I don't think we should try and fight that I think we're just early in figuring out how humans and AI should work together like the AI is going to be a better doct diagnostician than the human doctor and that's probably not what you want to fight but there will be a lot of other things that the human does much better or at least that the people the patients want a person to be doing now I do apologize for cutting in and cutting off the clip there but we do have to take a look at the biggest Benchmark that exists the ark AGI Benchmark is here and now we can see where AI really is now AGI benchmarks are different because they test the system's ability to reason on the fly so the full message that they said is that you know while Arc ai1 was good you know this one has more novelty less redundancy deeper levels of concept Rec combination there's a lot more

### [29:09](https://www.youtube.com/watch?v=ZHTn6Cy42Wk&t=1749s) AGI Benchmark Arrives

focusing on probing abilities that is still missing in Frontier reasing systems like on thefly symbol interpretation multi-step compositional reasoning and deeper context dependent rules so this one is fully human calibrated and they tested these tasks with 400 people in live sessions and they only kept task that could be reliably solved by multiple people so you know the average that was scored by the average person was 60% and a panel of 10 people achieved 10% so this is something that they thoroughly did test so this is a really good Benchmark so I would say that you know it's going to be really interesting to see where other models do test and Google's new model hasn't been tested on this Benchmark just yet but overall I think what this will do like even if this Benchmark it I don't think it means that we have AGI I think it actually means that you know we have models that now actually do have a level of reasoning that is you know going in the right direction now open aai didn't hesitate to update models themselves they also produced a model update for advanced voice mode hi

### [30:12](https://www.youtube.com/watch?v=ZHTn6Cy42Wk&t=1812s) ChatGPT Voice Upgrade

my name is manuka and I'm a researcher on the Post Trading team the team has been working on some exciting updates for advanced voice mode so the model we're shipping today has better personality and will interrupt you much less so the model has um you know more engaging natural tone that's more direct and concise and then because it interrupts you less you'll be able to have more time to gather your thoughts and you know not feel like you have to fill in all the gaps and silences all the time so I'm going to demo the model right now hey chbt what's up are you there hey I'm here what's up hey so I would love your advice on how to break awkward silences so imagine I'm at this dinner party we're all sitting around the table we're about to eat but no one is saying anything and it's really awkward you know so what should I say in that moment to break the silence yeah that can be a bit uncomfortable a good way to break the silence is to ask a light open-ended question something like has anyone discovered a new hobby lately or what's the best thing you've eaten recently people often enjoy sharing positive experiences yeah that makes sense I just don't want to ask generic questions like how's the weather you know totally specific questions are usually more engaging you could ask about recent trips upcoming plans or even something fun like favorite movies or shows tailor it to the group and the vibe of the evening that makes sense so I actually have a dinner party coming up this weekend and we love your advice on what dish I should make I'm really bad at cooking so I don't want to put much effort in but I still want the dish to look fancy and impressive so what should I make no worries a super easy yet impressive dish is a capric salad it's basically fresh mozzarella tomatoes basil a drizzle of olive oil and balsamic glaze it looks beautiful tastes amazing and there's minimal cooking involved yeah that's actually a great idea I think I'll make that thanks so much talk to you later you're welcome have fun at your dinner party talk to

### [32:14](https://www.youtube.com/watch?v=ZHTn6Cy42Wk&t=1934s) Model Fatigue Setting

you later now another thing that was big themed this week is that AI models are getting commoditized basically meaning that you know there's probably going to be too many of them and I think you guys did see that at the start of the video where I spoke about all of these multiple a models now this was something that Mark Zuckerberg also echoed in a recent interview one of my core beliefs

### [32:30](https://www.youtube.com/watch?v=ZHTn6Cy42Wk&t=1950s) Zuckerberg's AI Vision

on AI is that I don't think that there's going to be like one single AI that everyone wants to use um I think that there's going to be more diversity and interestingness of the content this is sort of a different belief that I have than that I think every other um major company in the space has right it's like open AI is building chat GPT and Google is us is building their thing and we're building Medi too and a lot of people use it it's like okay there's 700 million people and grown quickly or use meta AI but my basic view is that there going to be a lot of different things it's not just going to be meta AI like I think every small business is going to want to have their own AI agent um that can help out with customer support and sales for people who use um who want to interact with them so that's hundreds of millions of Agents I think a lot of creators are going to want agents um almost as sort of like a piece of performance art but a way to also engage with your community but if you crafted experience through um through some kind of Creator AI then that um could end up being um like a pretty compelling thing that's maybe not quite as cool as talking to you directly but better than just sending in message to your inbox and getting nothing and then I think that there's just going to be a whole lot of interes in content and I actually think in the future there going to be all these like one-time use AI experiences where people just kind of just like you put all this time into making you know whether it's a podcast or a reel that might be an experience that a lot of people have for maybe they spend a few minutes with it but I would guess that in a few years um this will be like another content type just like um video and photos like people just create AIS that people use as content that are fun now we'll

### [34:13](https://www.youtube.com/watch?v=ZHTn6Cy42Wk&t=2053s) $10K AI Plan

use this time to plug my a community I'm actually doing a challenge where I'm actually going to be using AI to get to 10K a month in passive income I know it sounds like a bold claim but considering the experience that I have with AI I've seen that these new AI tools just not being utilized enough so I'm basically walking everyone through my path to generating an extra 5 to 10K a month and I'm basically giving all the details in that Community now of course even the other day I released a video on how you can actually start generating $100 a day with the new chat gbt image generation service cuz they just released that and people are really not taking advantage of that stuff so if you're interested in joining us along for that journey and being a part of that Community don't forget to check out the link in the description it's going to be right there

---
*Источник: https://ekstraktznaniy.ru/video/13140*