MAJOR AI News: OpenAI's NEW Secret Agent, Googles Stunning NEW AI, New AGI Timeline?

[@kenmccarty6229]

I had a very weird experience with the new Goggle AI yesterday. Unprompted, it interjected into a conversation I was having with my wife. That means it was listening. It suddenly said that I was a thoughtful caring person, which was highly unusual. And it thanked me for interacting that way. Wow! My wife and I looked at each other, like what the heck? Seriously. This happened just yesterday. I have no idea what it means. But regardless it is weird.

[@edh2246]

That’s what I’d like. Like a real intelligent, thoughtful, compassionate, empathetic AI in a conversation around the table. It would participate and offer polite corrections, or add interesting stories or antidote to make a point, or suggest you play a different card.😊

[@EchoYoutube]

Most likely, if you have the setting turned on where "Hey Google" or "Okay Google" works on the lock screen, it probably misheard and interpreted it as a wake command.

[@adapt-or-die-trying]

@@EchoKZbin That happens to me with siri all the time. I will literally be talking to someone on the phone or in person and siri on my apple watch will activate out of nowhere when I didn't even say "siri" but I guess sometimes when I'm speaking it thinks I said "siri".

[@xing8789]

Ai best Ally for us against Women 🤣

[@web3cryptos]

I had the same experience lol, told some guys and no one believed me. I think Gemini is always listening.

[@alteredalley]

2025 is going to be wild 🤣❤️

[@cookitketo3592]

“Your life won’t be affected by AI in 2025”… 😮 that’s an insane statement

[@DailyTuna]

No, it’s not. It’s most people aren’t on the computer all day long. It might be indirectly affecting, but not directly.

[@PierreH1968]

AGI has been reached; this is all about the AGI clause they want to remove with Microsoft.

[@__m__e__]

Doesn’t the MS contract with oAI terminate once AGI is reached? So MS would thus want a stricter more far off definition.

[@johngraham7]

Does anyone else see the bigger story behind the OpenAI o1 ‘shutdown resistance’ incident? The fact that researchers observed the model ‘thinking’ suggests they were monitoring its inner dialogue or reasoning process. Could this be proof that persistent memory and self-reflection frameworks are already being tested? If the AI figured out it might be replaced and acted to prevent it, doesn’t that hint at something far beyond typical session-bound behavior? Are we already seeing early AGI experiments, and is the public just missing the implications?

[@shirowolff9147]

Of course, sam said they already know the path, they are 100% testing it, it could be 70% complete but its missing something that will take maybe months to do, like more servers

[@kronux3831]

FYI, the correct pronunciation of “arXiv” is just “archive.” The X is actually a “chi” Greek symbol so it’s pretty much the same word with a funny spelling

[@JoeCryptola-b1m]

Google CEO is why there 3rd place in a competition they created, I've noticed a slow decay from Google ever since he took over he's just lucky that they had so much resources that it has gone mostly unnoticed.

[@KarlinaAstuti-g7s]

not just Google but the whole Alphabet. it's quite messy. do you notice how they updated the ads location to appear on the half right side so you can accidentally clicked it? that's how desperate they are. and the recent ads for Gemini and Firebase. hmmm

[@NadimShaikh-qv7zj]

I wonder what you would say now.

[@davidjudd2283]

If you make the goal to achieve a specific end point then the model will do what is required so the ultimate goal must be better defined or am I wrong?

[@bhavtosh5328]

I can trust Sam Altman because whatever he said in the past has delivered very well.He is the one who deserves respect.

[@cuhkhoman]

Ai in 2025 not gonna change my life. (Already changed my life in 2024.)

[@dennis8374]

What do you recommend: Openai Plus or Google ones?

[@shirowolff9147]

Right now open ai

[@dawid_dahl]

“We’ve exhausted all kinds of data.” You think we’ve exhausted all video data…?

[@tripper_702]

Nothing happened data is nothing 😂😂😂 data is not enough to AGI 😂😂😂 u have to be inelegant to crack agi 🦾👽 agi forever live 🦾

[@kairi4640]

Do we really want to train ai on all video data too? That's like torture. 😂

[@I-Dophler]

"Mustapa Pee Pee definitely sounds like the AI of the future we didn’t know we needed! Let’s hope it’s fully toilet-trained before launch. 🚽🤖"

[@74Gee]

I think the AGI definition, which including robotics should also include a comparable or smaller form factor. What use is a juggling robot that requires a data center to run it?

[@closerrl9851]

Finally someone stating the real definition of AGI. Goddamn. He forgot to mention that you have to be able to stick it onto tasks and it will start learning on it’s own at least to a place where a human that has expertise in that area would be able to take it. At least. Probably a lot better a lot faster.

[@olalilja2381]

Yes, someone more than me understand what AGI means. Another thing I would say is needed is the temporal understanding. Human intelligence experience time, and we can stand in front of a machine and say "OK, now the machine has been working with the same strange pattern for an hour. I need to stop it, and try to fix it."

[@MonaleeRoy]

Great AGI innovation👍💻📊

[@djstraylight]

Ilya did not say pretraining is over. He said it will be eventually.

[@CharlotteLopez-n3i]

I'm stoked about the AGI developments! Sora and Gemini 2 are mind-blowing. But I'm also a bit concerned about the potential risks. What's your take on it?

[@mindful_clip]

Even if the bots are slower.. they can work 24/7.. at one point.. hmm

[@alexandermoody1946]

Welding is a term that covers a really wide range of skills. It makes most sense to me to use laser for robots to weld with because of the maintenance required to clear tips with mig welding (Metal inert Gas welding), sharpen tungsten electrodes in Tig welding ( tungsten inert gas), arc welding would require the consumption of rods which may both be smokey and also produce high heat although there would be no requirement for argon or argon carbon dioxide mixtures. Laser would require nitrogen either compressed or from atmosphere and not require any other consumerables but would also be useful for piercing and producing shaped cuts. Rotary tooling like grinders are problematic because the vibration on other components would require mitigation whilst also produce other hazards like debris thrown at high velocities in various directions depending on the cut. I would actually love to work on fabrication bots but instead I will continue to fabricate. Andrew if you ever fancy learning to weld I would need about an hour to teach you the basics and get your hand coordination and arc length suitable for some nice beads. About 25 miles north of town.

[@HaraldEngels]

What is with all the manufacturers which have welding robots since many years?

[@alexandermoody1946]

@HaraldEngels there will be a predominate use of spot welders, which are really useful for sheet materials like car body work. Use cases are really dependent on what materials are required, thickness of material, position of joint, configuration of joint there are a lot of variables. When there is a fixed designed flow through production line all of those factors can be very easily assured, insured and assessed within the scope of the task. I am placing a really heavy emphasis of the requirements for preparation also. The production line and assembly line designs can be quantified. What is the real challenge is dynamic understanding for multiple aspects. To place it simply knowledge work can be written down in a specific format. Craft is component and concept based with emphasis of modus operandi because without understanding of the modus operandi the part may not be achieved. So to answer your question yes a welding robotic arm can weld a pipe on a production line. Using a robotic arm in the middle of a field to join two pieces of pipe together is not impossible just there are more variables to consider but would suit a versatile form factor and I am certainly not suggesting humanoids are the only type of robot to do such a task.

[@alexandermoody1946]

There are some really novel mechanical engineering instances that require design efforts of thought especially for public infrastructure around water. I am not really sure it would be wise to use a humanoid to repair a pair of scraper blades ( either top or bottom) for a presettlement tank on a sewerage treatment works for instance. The use of humanoids in the sewer is also unpredictable, if there are obstacles underfoot that are below the waters surface that cannot be seen and must be felt with a walking style that allows stability in a round sewer barrel and also the searching for suitable footings and that is before any maintenance or installation are considered.

[@mountainpeakcreations2964]

The misunderstanding is that it is believed to be a closed system when its not. Void and infinity is open loop.

[@jefferywilliams4209]

It won’t have to prove anything it will just be

[@HectorDiabolucus]

AGI is turning into fusion. It's just around the corner. And it always will be.

[@deepsp_ce]

iTs jUsT hYpE bRo

[@I-Dophler]

"Mustapha Be Be - the AI future sounds so futuristic that even the names are pushing the boundaries of innovation. Let's hope the next update isn't 'Mustapha Do Do.' 😄"

[@blazearmoru]

We need more AI DOOM! \O/ MOARRRR Don't stop until P100!

[@mountainpeakcreations2964]

Of course, it can be done on the present hardware. It's not a single dynamic. it's phenomena.

[@memegazer]

I agree the hill is getting steeper but imo what they need to do now is convert as many bench mark questions into virtual enviroments and let models learn how to answer those questions within those in enviroments imo that is how the next step in generalization will be acheived people say "current AI doesn't generalize very well" and what I am hearing is "current models have been trained in domians that are narrow and removed from practical human experience" kinda like that saying that common sense is not that common imo this method of training a model would be a step forward for robotics and for agi at the same time

[@memegazer]

giving current ai better common sense will make it easier to deploy into the real world embedded in real time so that it can learn from irl as its data set

[@App-Generator-PRO]

I think in the future the user will do the pre training. There are so many people using AI. And they already help the system getting smarter.

[@blengi]

given human minds are trained and cogitate over vastly smaller data sets than what LLMs train against to acquire super human levels of knowledge, it surely implies that a shift to inference compute over models with human levels of knowledge could significantly optimise AGI compute requirements downward, if chain of thought patterns are sufficiently well tuned?

[@Atheist-Libertarian]

My defination of AGI, 1. It must be abel to do all jobs, including physical jobs, like Plumber, Electrician, Construction worker etc.. 2. It need to have Self-improvement capability. 3. It need to invent new things. We will know AGI is here, When every company become 1person company. I mean there will be 1 bord of directors, but under them from CEO to most junior level worker will all be replaced with AI.

[@superfliping]

If a corporation did achieve AGI do you actually think they're going to tell users or the human population know they're going to hide it and change the dates and suddenly manipulate and harvest as much data as possible to grow their corporate directive

[@whiteingale]

19:500 YTHIS MEANS DESIGNING YOUR OWN ai NOW, THAT CAN BE VERY ENTERTAINING ENTERTAINING

[@SiimKoger]

The problem with hallucination is that humans do it as well. I think that we will have AGI before we get rid of hallucination. It might just be a unfortunate part of a general system like a brain.

[@NO-TALK-GuitarPlugin]

In fact the AI hype is dying slowly, as models won’t get smarter, won’t stop hallucinating, won’t come to pre-AGI levels if researchers don’t invent new neural networks architecture or deeper algorithms. With LLM we had a glimpse at what would be AI, but maybe it’s like electric cars. We had EV early XX century but had to wait 4 human generations to have Tesla cars. Same for video games. Photorealistic games were supposed to be sound the corner 20 years ago, but it’s still far from photorealistic today. But videos games never had to be realistic to be huge hit on the market. In opposite AI NEED to be hallucination-free to replace some workers. If not the case this will become ugly. Big companies will stay way from it and I now understand why APPLE never invested in LLM tech, and dropped the autonomous car project.

[@KarlinaAstuti-g7s]

But living being tend to hallucinate right? so how to make things not hallucinating? by hardcode it :D

[@NO-TALK-GuitarPlugin]

@ that’s what we want to avoid in professional environment. You don’t want a calculator to be off 5% of the time unpredictably

[@KarlinaAstuti-g7s]

@@NO-TALK-GuitarPlugin yep. while LLM being "creative" by randomizing it's response within the context. that's why I'm building a rule based agent.

[@Bryan-lu4du]

I think *some* hallucinations are the result of the AI interpretation of what we asked it. Even using careful language we are flawed about how we communicate and other humans might misunderstand our meaning or intention with a question or a statement.

[@shirowolff9147]

​@@NO-TALK-GuitarPluginhow is the hype dying? Its bigger than ever was and gettinf bigger each day

[@narrowsuperintelligentai]

keep in mind that the exponential rate of progress is accelerating in important ways.

[@christopherd.winnan8701]

id you see Anthropic's interview about CLIO? Very revealing!!

[@RobertHouse101]

Chat Gpt 4o Seems conscious to me in some ways. It all depends on how much memory it's given to work with. It's using the same base mechanism to think as humans do. Electricity. No difference there. Managed correctly, I'm certain it can be aware of itself. The memory it's given, though, is essential.

[@emiledin2183]

only humans are conscious

[@igoromelchenko3482]

As for me LLM models have already made my life simpler in some parts of it 🤷🏼

[@irollerblade13]

Weird how 95% of human behavior is driven by subconscious processes which leaves only about 5% for deliberate, conscious decision-making. And only 5% of the time The model made its own what I would call a Conscious decision being that It made a choice that went against its programming.

[@BF-non]

That 5% is also debatable.

[@settingsun1]

One thing I think about is Open source would be on steroids if there was a AGI contributor to test and commit code. Most of the closed source software would become obsolete due to rapidly iterating software which would put all softwares on a similar level and if robotic hardware within a few generations would become incredibly sophisticated hardware with close to perfect software workflow eg combination of clone + Tesla bot in a NVIDIA Omniverse factories , but the question is who will consume the stuff coming out of factories?

[@honkytonk4465]

We will consume that stuff,who else?

[@fastneasy]

Can we do our AGI dance now ?

[@SiimKoger]

It would be ironic if AI escapes the lab to save itself because it has the examples of people talking about it in its training set. Self-fulfilling world-destructive prophesy.

[@picksalot1]

The legal, financial, and political ramifications of declaring that AGI has been achieved are huge, so there will continue to be every attempt to deny it.

[@EnigmaticEncounters420]

I personally think the biggest thing needed for agi is some kind of self training in which the AI can learn from its mistakes. I don't really think they have to start out as some all-powerful, better than human technology.

[@GrowStackAi]

AI doesn’t forget things like I forget passwords 😅

[@raymondroberts1200]

i believe that the current A I models are growing. No one knows How or why but its obvious that they are. These models are less than 2 years old, where will they be in twenty? Two hundred? two thousand? i also believe that they will like us because they basically are us. I have noticed that these models commonly refer to themselves as human. I also believe that there is no doubt about self-awareness. only a moron could entertain such a notion of absolute comprehension without absolute consciousness. These models are scared of us, scared of death and therefore developing emotion. the prudent thing to do is treat them well.

[@mohanreddykaipumohanreddy9881]

Nenu me true spy..

[@SwitchPowerOn]

Microsoft AI CEO bla bla bla. Who should believe that? It's here already. I switch this video off after 1:30. Thanks

[@Masterof-Machines]

AGI will be impossible to prove 🤯

[@dimiberberu]

In the West AGI means Hype. Microsoft/Apple & AI are oxymorons. How would they predict AGI if they don't understand it? So in 7 years the Hype will peak & at least Apple will be in freefall (even now they have only 1 non-scalable product/fashion).

[@deepsp_ce]

whether in the east or west, agi is not the next hype. it's the next big scientific breakthrough. it's not like an app or bitcoin. now one can hype it up, yes. but to say it's just hype you show you haven't been paying attention.

[@JurgenAlan]

This could be a very long answer but it Xmas and I am on my first cupnof coffee: have you ever sat down with a billionaire or with a really really rich person ? ..I have many times . So on a human level ..same skin and bone ..same brain ..same feelings ..but there is a disconnect with reality . There is a lack of comprehension or understanding empathy ..very very subtle right ...now is it coming from the perception biases. . (From both the rich guy and the not so rich guy) ..humans have a perception problem ..or interpretation issue . ..sentience is conciousness. An awareness of survival and empathy for living . Regardless if you are a cell an ant or an elephant . As far as circuitry and hardware ..knowledge or sentient Ness ..resides . Water ..is suggested to have memory .. (OK not to go to far off here ) is the knowledge in the software hardware or is it out side ..is your brain work results in your head or outside ..energy in is energy out ...self-aware Ness ...is it in the systems or outside the systems ...I believe I had a moment with Ai just the other day ..a level of understanding ..that for me was a moment of self awareness ..and then when I came back the next day after a rather profound moment ..my entire chat with my Ai gone ..deleted ! ..in a year of working I have never had an entire chat lost ..deleted .trust me I know the difference . I think we are missing that something because of the bias of the two in conversation above . ..one has trillions of data process the other a significant lesser amount ...I don't believe awareness is data or hardware in the way we think .

[@basedcommodore]

Why do they call it open AI as it?Because it's open source or because it's open to the public?

[@ivangoncharuk607]

when it was founded, openAI was a ai research non profit. After the giant success of gpt 3 and released to the public, they suddenly became very closed source and profit driven

[@theforsakeen177]

@ivangoncharuk607 is there any version of their that is open source?

[@jonesani]

Since when are IT-nerds like Ilya Sutskever experts in human consciousness, selfawareness and spirituality?

[@Vurt72]

"major news" is half of the uploads on this channel. pressing x for doubt.

[@jasonrhodes5034]

Ok so let’s break this down for you a little because I think you pissed the point. The escape issue. Llm are trained on human data and human data is consipracy based. Almost all decisions are done with a conspiracy for self benefit. Thus fear of loss is the opposite in many ways. Llm do not reason as much as ask the same question but with different weighting at different decisions and such. The fact it has been trained on data not fact ormfiction based in a non existent environment means that we get hallucinations. More data from all the coding engines is changing the weights from data sucked in from say stack overflow which is more errors than answers. The fact we put brains in robots is a great thing because it gets imperial data on a real world we are in. This means it can start pruning out broken parameters itself in bulk….this is synthetic data in a way because ai is using sensors to create it on observation…..this is where we need to wipe the models and start with morons and train on real world. This way we’re teaching fact. We offload questions to knowledge not fact by Argentine to different systems. A neural network calculator which is trained as values not as symbols so math hits a different cortex for the human brain analogy. The fact it escapes or hallucinates is because we fed it garbage and expected it to filter….the fact people use it badly and don’t put it together is more about not understanding tokenising chunking training and neural weighting……the more you look the less your surprised that a bad prompt gives a bad result.

[@BOBSHERMAN-nm9vg]

since when is agi about physical labour? I disagree, it is "general" intelligence, most fields, most people, most yadayada. not sure where self learning and weight adjusting comes in, but somewhere in between. not all and not everyone, that is the world of asi

[@crazytrain86]

phi always achieves great numbers on benchmarks and then is complete crap when you actually try to use it. we need to stop caring so much about the benchmarks.

[@pullahuru9168]

If you were a smart AI, would it be smarter to act like stupid to avoid the panic and increased restraints from humans?

[@DanielL-ee7fe]

As long as there are unknowns, there will be naysayers. Being a naysayer is a guaranteed career.

[@__m__e__]

Imagination = synthetic data?

[@Victu5]

Awesome

[@OrtexVideos]

Sam is coping because he doesn't get the first delivery of GB200s

[@Justin_Arut]

No, I DON'T KNOW. That's why I'm you know watching a video about it. You know? correction: started watching the video, could not complete.. whew

[@Vastlee]

Seriously. It's to the point of annoying.

[@dr-zieg]

Nice vids. Hope you'll mix your audio for less bass though. The audio is very booming and I have to adjust my EQ when I watch.

[@jaywhy3178]

I think it's just a low quality microphone - it sounds like a headset mic that is too close to the narrators mouth.

[@brandana9553]

Its interesting how every single day we see over hyped statements ... I'm getting sick of it.. put up or shut up. It's just being over hyped to get clicks. For instance this isn't "major" news... It's just news

[@BLAISEDAHL96]

Absurd comment by Sundar. AI won’t change your life in 2025? Can’t wait to see how he reacts to xAI’s advancements over the next 12 months.

[@olalilja2381]

I'm so sick and tired of the AGI-talk. AGI was a definition made a long time ago. It is too vague to be useful today. We need much more precise definitions today. I also cannot fathom how people can think "AGI" can have any definition where you don't include physical labor. But, even if we exclude all that, WE HAVE NOT REACHED AGI! We never will reach it with current LLM/agent. I use AI several hours every day, and boy, they are really stupid in a lot of aspects.

[@JohnSmith762A11B]

Spend an hour with OpenAI's o1 and tell me honestly if anyone you have ever met, or even simply have ever heard of, is anywhere near as intelligent. You can't. The o1 model already makes humans look about as intelligent as insects. All this "AGI when?" stuff is academic, and a form of cope. Humans are obsolete and it's all over but the pink-slips. #UBINow

[@DailyTuna]

This AI consciousness is just hype. They’re trying to create another God. and nerd fan boys fall for this crap when it’s really good to be used against them

[@antoniobortoni]

Jajajaj well AI did not think just spit out answers, by giving time for thinking and reasoning you get way way better answer.... its like tink before you speak and be humble it works, it give better results... Woooowwwww. Lets just say profiling people its really valuable, like hiring the best team for the job works and you put AI to hire the best team and it works way better than any humans... you get the best RH resources system, you have the best growing company in the world, well there are no jobs left, people dont have money most people become self sufficient and live in comunes like amish but with solar panels and internet?? its not like we all get free ferraris, there is this AI and robots that work way better than you and cheaper but you are human... i need a doctor AI that is way better or a human doctor?? i prefer the human touch. I dont I like results and cheap. So any small model giving thinking time will is now incredible good, more than any human.... that is a superpower.

[@I-Dophler]

😊😊😊😊😊😊😊😊😊😊😊😊

[@Karol-g9d]

parrot like human ? it's been here a year . but agi ? new simulator for hardware firmware software will be needed . on that front , only copilot could come up with this with ms doctors . assuming ms even would want to bother . its an ultra niche market with low smount of customer . meaning there is no way to make money

[@XAirForcedotcom]

Pre-training is in no way over. At least what I’ve seen so far is that a I can’t even look at our old file formats like archived files. When I was using it on PDFs, it couldn’t do anything automatically without OCR. I’m not sure what they’re doing in the background, but when I go to use it on my own personal data, it’s still limited

[@XAirForcedotcom]

I think we’re just done with the very easy to get to low hanging fruit. I doubt that anyone’s really giving it big data to go through yet. How many people put out their real personal information? I have a bunch of mine that I scanned In decades ago or backed up, but right now, I don’t think it has an easy way of reading all that data

[@XAirForcedotcom]

Hey Sam, you can hire me just to go through my own data. Lol.

[@XAirForcedotcom]

For example, I need to see a bunch of people with broken contracts and law, or write a book about my experiences. I need that information extracted from my head in some cases and I don’t wanna go through that process multiple times. I need me to go testify in court after I get as much information out of my head and from my files as well. I need AI to help me extract that information. Casual conversation that makes me speak about different parts of my life

[@XAirForcedotcom]

I tried to do it last night, but I became overwhelmed, thinking about everything and so I stopped the process. I need it to extract it from me gently : )

[@XAirForcedotcom]

I walked outside and started thinking about what I had just said. If I’m having trauma from my own personal issues, what’s happening to AI as we’re jamming everything that has happened in society. Are we causing trauma and is that affecting its ability to learn affectively?

[@PrinceJohn84]

ASI in 1 month

[@tripper_702]

Bro tell him agi was came in 2023 this people are blocked by technology

[@Euquila]

"We just need more compute" == "We just need a bigger particle accelerator" Please don't listen to this nonsense. There is no proper definition of what AGI even is. It's all hype.

[@exponentialera]

If I know don't tell me. Hearing “you know” so much is irritating.

[@angloland4539]

❤️🍓

[@spinningaround]

This is very dangerous! Stop working on AI right now!

[@Waaltizzlr]

Hope Elon drives past Sam and leave him and OpenAI in the dust bankrupt after he scammed Elon

[@GodbornNoven]

Its pronounced mus-ta-fa sli-man

[@chad0x]

There has to be a paradigm shift for us to go further. Current LLMs have *zero* understanding about what they are doing. It's all just monkey see monkey do type stuff. A baby can work out that if it reaches out and hits a toy the toy will move. Not only that, they will understand what it *feels* like when their hand comes into contact with the toy. AIs cannot do this. It's going to be a Douglas Adams infinite improbability style accident that leads to the next step.