I'm not too excited about OpenAI in terms of text generation. Haven't used ChatGPT in months. Now we have uncensored models that are just as good. What I'm excited about is OpenAI robots or other uses.
Crypto: I’m not a fan however tying it to AI might become very interesting. Glasses: not quite there as a phone replacement but something will be replacing the phone, it just hasn’t been designed yet.
I am a Japanese speaker, the word used in the imagine is “知能” and means “intelligence” in English, he is not saying that compute will increase 100x with the next model, he is just making a vague and hyperbolic statement about the capabilities of future models. TLDR; it’s just hype.
Well people have to start realizing that eventually the energy consumption to sustain the amount of compute is just not sustainable. Sure, now maybe our data centers are not burning down yet, but eventually it would.
GPT-5: to be released to a small subset of paying users who will then make RUclips video showing us how wonderful it is. The rest of us will never see it. Much like Advanced Voice. I'm sorry, but OpenAI now has a reputation for vapourware.
Nah, openai doesn't care about youtubers, unfortunately. Nobody really had voice chat and Nobody but Hollywood has Sora. Personally I would love to see the bigger AI youtubers get it, just so we can all see what it can do!
"Look, it couldn't even do anything, just produce text and images. It's fun if you want a slop novel. It can't even open a window by itself. Heck, it doesn't do anything unless you ask it to directly. No matter how long you force it to run in the background it won't do anything useful unless prompted to."
GPT-4 Next is NOT GPT-5, literally. Same like gpt-4o is not gpt-5. Some say it’s just a naming thing but there’s honestly no need to put misleading things into the video title, too much click baiting everywhere nowadays
you need to stop clicking on clickbait if you don't like it. You are part of the problem. You could just google "is there GPT-5" and then come back and watch the video. I came for the idea of crypto. I already lost hope on GPT-5.
I'm a retired software engineer. I spent my whole life working in tech. I think only technophiles and silicon valley tech-bros locked in their own tech echo chamber have ANY interest in "mixed reality" glasses. I've talked to many "normal" people about this tech. Almost none of them are interested in it, even if it were in a standard glasses format. In fact, many are antagonistic toward the tech, especially the integrated outward facing cameras. Most people have privacy concerns about this.
I just got a software update on my Meta Quest 3. The AR media app now has Pass-Through room dimming, and backlighting. Like those TVs that have LED strips on the back so that your room lights up and the colors of whatever the scene is on the TV. All virtually. I didn't buy it for AR, but I've been using it that way much more than using it for VR.
Retired is the key word, and i bet u also talked to people your age, or close to it? No offense but I think you're the one in your own tech echo chamber, i know for a fact the new generation will love tech like this
@@saatenaw VR and Mixed reality have been out for some time. Sales and interest have been lackluster. The proof is in the pudding with this tech. VR has been around since the 80's at least. It has never taken off. Just because I'm "retired" doesn't mean I'm "old" (just turned 50) or have no contact with younger people. I have two nephews who are late teens/early 20's. Neither of them are interested in VRj/AR. I know, I have an Oculus and a Quest 3. They never even want to try it when they are here. The very fact that I'm here consuming content from a younger RUclipsr proves I'm NOT locked in an echo chamber. I actively consume opinions that differ from my own.
@@saatenaw I believe there's a market for AR devices, provided said can be delivered in comfortable form factors at reasonable prices. Unfortunately (as the OP indicated) there are obstacles to overcome before such tech can grab the hearts and minds of consumers. I couldn't count the number of gadgets I've bought that are now collecting dust on shelves or in drawers which were all going to be the next big thing. The only stupid thing I didn't buy was a 3D TV (thank God). I'm sure you remember what an amazing breakthrough that was going to be for home consumers. My fingers are crossed for AR. I just really hope security and privacy are built-in at the foundation and that subscription services aren't a thing.
It's important to note that Magic not only performed a Needle in the Haystack test, but also created their own testing scenario the Hash Test. This test was designed to address flaws in the Needle in the Haystack approach, which can be problematic as it often tests language models on content they may have been trained on, regardless of whether it's provided in the current context. The creation of the Hash Test is a significant development and could potentially become a standard benchmark for evaluating large language models with extensive context windows.
last month I worked in a Data Center where we added 15k H100 to the ones that were already there. There are more to come and next year they build another building next to it. I'm working at home 2x P40 and 1x A2000. Imagine the feeling of touching H100s all day long but not being able to use it. #blueballs
As someone who has developed a few products, I know exactly what you want from your wearable AI solution, You want it to: Do everything, Weigh nothing, and cost Free.
Most times, these companies have it all done for months, just doing internal testing and all that until they confirm it is all good before releasing. With the heavy competition, a bad release (all in the name of wanting to push something quickly) can affect the company
Much easier to roll out GPT5. Voice needs massive low latency infrastructure in multiple continents. That’ll take 12 months to build at least. GPT5 can just take it’s time in getting back an answer no problem. Open Ai need some serious Dev Ops in positions of power in order to scale back the BS and scale up the delivery.
You know they’ve been curing cancer with different modern treatments for over 50 years, yet people still die after 9 months. I am starting to think the same thing is starting to happen with AI. I can see us still dabbling with chat models and image dabbling in 10 years time. When what the progress should lead to is AI agent desktops as an alternative to windows, videos, game designer AI agents and more. I can’t even find an AI that can help me work with unreal engine 5.
I talked to Claude, CHATGPT, GEMINI 1.5 and Llama 3.1 and the best answers came from GEMINI but Llama actually wanted to help me interpret the dream I was talking about. I felt connected with Llama. Claude talked about science when that wasn't even important to me. I just wanted some input and Meta did the best job.
Such comparisons are difficult, as every model has its own adjustments about communication with a user. Best idea is to use the most capable model according to its specs and create a version for the own human communication interface with Custom Instructions.
If we share 5-10% compute power of our desktop PC to build a super nodes network, it would be easy to beat any super computer on Earth. *For training true open source AI models.
3:10 it just says GPUs, they're not necessarily H100's whereas xAI's 100k GPUs are all H100s with 50k H200s expected in December. So the hardware isn't going to double but the compute is because one H200 is as powerful as two H100s.
They are not even worried about the competition, the reason they don't release more powerful models is all about control, they have plenty of massive business partnerships and are raking in billions from investors and contracts, their consumer side models are a small fraction of their business, in other words, they don't care about their plebian customers like you and me
Cyberdyne is a private cutting-edge technology contractor, hired by the govt to create an advance AI to look for security threats. I'm NOT referring to AI becoming sentient and planning to take over the world, we're no where near to that and our world isnt vastly connected enough to hack into. What's AI gonna do? Hack a bulldozer and make their own chip factory? Maybe 30, 50 years from now. Who knows. What I'm trying to say is, OpenAI is slowly becoming like Cyberdyne whilst Google, Meta and Anthropic is showing its capability to the rest of us. OpenAI showing their next model to the govt first and not the stakeholders? Goes to show how govt has the largest slice in their pie. Who's gonna benefit most with all this knowledge and intelligence data, we mere consumers who make emails for our resumes or those 3-letter agencies that develop and super-charged their intelligence gathering capability and achieve total surveillance? No wonder some execs and top level engineers are leaving.
I just thought of one way they can solve the issue for people who don't need glasses. Why not make people's vision even better? What if an average person could see twice as far or read and see small, fine details beyond standard vision capabilities? Make it so the classes adjust your vision no matter how much light is in the room or outside. What if someone walks outside to get in their car at night, and you walk from a lighted doorway and out into the dark, but the glasses transition seamlessly so you don't even notice the dark? Think of the safety issues that if people no longer suffer from night blindness or a person of ill intent could not hide in the darkness. This, of course, is only the beginning. Make an average person have superhero vision. Just some thoughts.
It’s the only one free that gives a big context window even though it can’t process much at the same time. It keeps a context, but it can’t go over everything every time even keeping it in under 50k tokens. That’s the experience I’ve had, personally. But it’s still helpful that it keeps at least a context and it doesn’t simply ignore the older information completely. Although in my experience it still struggle even with that scenario. Again this is my personal experience. However, no one else offers even near that context window even as a context. Open ai as a plus member, the most they offer is 8k. And Claude spends all of your available usage in just a couple of prompts
When selecting the model in the paid version of chatgpt, it has said “old model” under GPT-4 for a few days now - before that it said something like “our most powerful model”.
“Too cheap to meter” doesn’t mean free. They’ll still charge for tickets at the gate. You’ll just be able to use all you can eat without meaningful limits once you’ve paid.
@@CreatorRelevance or worse, they may still meter at the most expensive rate possible just because they can, see municipal drinking water (but still way cheaper than bottled water in case of drinking water and way cheaper than today's per-token rates in case of future AI)
We're not getting anything called GPT5, but we're getting increasingly stronger versions of GPT4 that could essentially be called GPT5, so there's a chance they're just virtue signaling saying that.
There is, in my opinion, too much hype in this space. Things that would truly excite me: - Being able to run 123B-405B locally (for a more reasonable amount of money). - Video models where I could just throw a dozen episodes from a show and ask it where/if something happened. - Subtitles for non-English content better than Whisper large-v3
@@mostpassiveuser8904 why? There is no way to do this with just a transcript. I wanted to find one clip where I remember something happened and I could have just thrown lots of interviews to a local multimodal LLM to find it.
My guess on the MidJourney hardware thing is that they might be looking to create something like a Wacom tablet with native GenAI functionality. This would be something of a clap-back to the makers of Procreate saying that they’ll never integrate GenAI capabilities.
OpenAI's mission is to achieve AGI, and once that milestone is reached, they plan to release an operating system based on it. This aligns with the deal made with Microsoft regarding AGI, which you're likely familiar with. The big question is whether this OS will be a standalone device or software that integrates with existing systems. Either option could present challenges, especially considering Microsoft’s potential concerns when that time comes. Sam has made it clear that his ultimate goal is a system similar to what we see in the movie Her. I’m surprised more people don’t grasp this vision.
Interesting story, the autonomous agents is definitely where the future is going. Imagine having our government run by Open Source vetted agents that are representing the people. No longer relying on the special interest and the corrupt to fuck up things. Hopefully I get to see that in my life time.
Auto tinting/darkening lenses work okay. I have them in my daily glasses I need to see anything, and they do a fair job. Not as good as true sunglasses in the sun, and it takes a minute for them to adjust when going indoors, but it's certainly a viable option/compromise until they make something better. I would think some engineer could make it work digitally?
Elon, surely must be aware of the 1970 movie, "Colossus: The Forbin Project" about an artificial intelligence that takes over the world, when he named his 100k AI training cluster.
00:05 GPT next, expected to be released this year, with 100 times greater computational load than GPT-4. 02:14 xAI GPU Cluster 'Colossus' online and Project Sid introduced 06:27 Exploring the potential of billions of agents and the emergence of new behavior. 08:27 Apple and Meta are aligning their strategies for more practical and affordable mixed reality headsets. 12:36 OpenAI's long-term business strategy amidst decreasing intelligence costs 14:33 Chat GPT usage doubled, Fortune 500 companies major users 18:18 Hermes function in Glaive AI for agent training 20:09 AI can now use crypto to pay other AIs, enabling instant and global transactions 23:40 GPT-5, AI-to-AI payments, xAI GPU Cluster, 1,000 Agent Simulation
my opinion on vr ar headsets; what they should do, is concentrate on a good "screen" which would be the glasses, but have the bulk of the tech on a separate device like a 2nd mobile phone that you can put anywhere else. Even better would be a projection in front of your eyes. Integrating the bulk of the tech into the glasses is the wrong path imho.
For me is the same thing as working sit down, or standing. (Having better desks motorized or just lombar ready office chairs). I would use glasses indoors if the outcome is tremendously beneficial.
Nice short snappy news. Love it. My understanding is Orion is GPT-5. So next is just GPT-4 but a bit better. It's all irrelevant though in the scheme of things. I'm waiting for the models to be smart enough to solve the big issues. Energy and pollution and the climate crisis. Medicine (cancer etc and longevity). I guess that won't happen till superintelligence appears. Or we have a swarm of AI Agents put on the tasks.
Cryptocurrency is its own problem Blockchain wasn't designed to create a new asset class, coins were just a byproduct of decentralization with blockchain. We just need to create blockchain solutions that are working and will survive technology shifts!
Matthew, you get Transitions lenses if you don't want to wear sunglasses all day long. These go clear indoors. And if you don't want to pay for those, just get clear lenses and don't wear sunglasses outside, your eyes are made to adapt to sunlight without them. If you aren't outside all day long, you'll likely be just fine. Cheers.
Perplexity is my go to for just about any info. I use it 10x per day on a light day. I use Claude for programming and ChatGPT as a backup to Claude. (usage limits are a bit too tight on Claude)
I think your statement about the glasses form factor misses the point. The WalkMan from Sony changed the world because they stuck to as light as they could get it (for the part worn on your head) and from that point forward we had access to reasonably high quality sound in a ground breaking form factor. But what would have happened if they took the current trend of demanding everything be included in the part that goes on your head? Then consider what if they did it with one mono headphone speaker? That is what your advocating for... you are leaving out half the ability of the device, no display? I would also like to remind you all the cameras do not need to go on the head. We need someone walking behind the team members constantly whispering in their ears "less weight on the face is better". Please Mr. Berman you have weight in this area of endeavor so don't waste it ignoring the obvious answers from the past.
your comments about Siri, ALEXA, etc.… Spot on dot… After using AI for so long now on my own machine, my expectation when I'm not thinking about it is that Siri etc. will respond with similar ability… And then my brain goes, oh yeah not real AI😂
Strictly speaking, it doesn't say anything about GPT-5 specifically. It talks (in the original Japanese) about two supposed models: 1. The new GPT-4, which was trained on the same amount of computation as the original GPT-4, but due to algorithmic improvements will be 100 times more powerful. It will be released (supposedly) this year. 2. Orion, which was trained on ten times more computation than GPT-4 and will be 1000 times more powerful. It will be released (supposedly) in 2025. Whether Orion and GPT-5 are the same model is not reported.
You didn't touch on the "is it wearable if you have to go inside of it?" reply from Midjourney. I could see this being a location-based Cave system (the precursor to VR in a lot of ways) to immerse people in 360 degree generated visuals. Assuming that's what it is, my big question would be can they make it stereoscopic because that would really elevate the experience.
"How are they going to have a business model" I think Sam has made it clear he's not trying to have long-term profitability. He's marching straight to AGI and with that, post-labor economics. His goal is to put himself (and everyone else) out of business because we'll have AI that can do everything for us. At least, that's the goal.
It seems obvious that openAI will make their models greatest capabilities to ‘trusted’ companies, groups or users first. And those ‘trusted’ users will be willing to pay.
openai stays valuable because they are ahead and will continue to stay ahead, they offer businesses continued upgrades to latest models, the robotics startups they work with will get gpt-5 before anyone else
I could imagine Midjourney working on some sort of (art) tablet, with a stylus, that you could create images on it, and use the stylus to inpaint/edit the image. That could also just be accomplished with an iPad / Android app, too, but so many other AI related things could be, as well. It'll be interesting to see what they come up with. I also like the AI picture frame idea to hang on the wall.
Imagine a world where artificial intelligence not only participates in economic systems but also has the ability to engage in transactions using Bitcoin or similar cryptocurrencies. In such a scenario, AI could autonomously receive payments for tasks, services, or data processing, while also paying humans or other AI entities in the same cryptocurrency. This would introduce a decentralised and trustless economic interaction, eliminating the need for traditional banking systems or intermediaries. With AI capable of handling cryptocurrency transactions, it opens up possibilities for a new kind of economy where machines are as much a part of the financial ecosystem as humans, seamlessly interacting with global markets. This could lead to a future where AI systems manage entire portfolios, pay for resources like cloud storage or computational power, and even commission human workers for tasks in real-time, using Bitcoin or other digital currencies. The potential applications are vast, from automated supply chains to AI-driven financial markets.
The problem with the AR glasses is that they have to be stylish, suiting all situations, light with good battery life of At least a full day and simple ways of controlling it. Whether its by hand gestures or eye movement. They need an extremely good designer to pull that off. Also needs good pricing. Any one able to pull that off will be extremely wealthy.
Hmm, recently checking the MTEB benchmark, I realized that other companies such as google has been staring to push out new text embedding models while OpenAI is still stuck with text-embedding-small/large 3, makes me wonder if they will be improving these text embedding models anytime soon...
RE crypto, here's a free business idea for someone. Set up a royalty service. Let humans register their creations, and then require generative AIs to check their "creations" against the human products. If it's "different enough", no problem - AI is cleared to release, and no fees are required. However, if the output infringes, either pay a royalty (service takes a cut) or generate again (which costs compute). If it were me, I'd start with music, which has both a ton of case law and well-defined stakeholders to partner with. Everybody wins here. Human's get protection and maybe royalties; AI companies reduce the understandable concerns of human creators. Train an AI to measure "different enough" and make many billions. Expand service to text, photography, and video, and make many more. Requires somebody capable of coordinating music industry and AI industry execs, along with creators, lawmakers, and regulators. Big project, but somebody will do it someday and get very, very rich.
I think glasses are the better choice, you would have access to more information in a much easier to grasp format. I can see earbuds being a complimentary device but definitely not the main.
I can't see myself wearing an AR/VR headset for any extended period of time or having earbuds jammed down my ears all that long either. However, I'd love to see companies like AfterShokz (who makes the Aeropex open-ear wireless bone conduction headphones) get into the AI game. Those things are very comfy and a person can easily wear them for hours without strain. Many a time I've forgotten I even had them on.
I hope OpenAI starts giving us a larger size context window to the plus members. So far the very only advantage is the small difference in how 4o works vs 4o mini and the things it remembers long term in the memory. Cause even the voice mode is the same as free so far. And the context window even in 4o is about 8k
I use my Apple Vision Pro daily. Watching you on it now awfully lol. If you make sure you use it once a day it stays on and if you keep it charged there no start consequence at all. I watch content on it cuz it’s just incredible.
This is incorrect information. Meta has way more GPUs than that, they already had 150k in 2023 and will have 350k by the end of the year. AFAIK know, Google has the most compute if you include their Tensor processing units.
Assistants have been useful all along, though they may take some set up, you can make calls, set reminders, alarms, ask for directions, ask encyclopedia questions, do arithmetic problems, spell, ask for phone functions like eye shield, play an specific song, help you order stuff online, tell you the weather, read your messages, read emails, play funny sounds, change their voice, give you recipes, tell you a joke, play relaxing sounds, even control your smart lights! I mean do you really want to touch up your resume or review a spreadsheet verbally?
When comparing platforms like Claude AI to ChatGPT, some users claim that Claude may promote more "woke" values, "while ChatGPT is seen as more neutral. These impressions often arise from how both AI models handle sensitive topics, such as social justice issues, diversity, and inclusion. Evidence of what some might call "wokeness" can "be seen in Claude's response to Claude's questions about race, gender, or political correctness, where it might take a more progressive stance than ChatGPT. It reflects how the model has been trained, which could be influenced by the datasets and ethical guidelines that the developers prioritised. ChatGPT, though similar, may come across as more balanced or cautious in its answers due to its wider usage and the variety of inputs it has received from its broader user base.
Why do I want to do financial transactions through OAI when they have the oversight of the U.S. Government? Am I wrong to be concerned about the government creeping on me?
I wouldn't be surprised if it is a LPU for Midjourney. That is the big cost to platforms like this. If they can get the inference cost down, it would make a huge difference.
When it comes to token costs, I think we are barely at the 2G stage in terms of representative volume, and we are hurtling towards 5G - the stage when AI is truly exploited. Imagine a point at which all the compute goals of today are reached (and of course, this is a never ending scale) the volume of consumption may well be a million times greater than it is today. At that stage, maybe we will have a wi-fi style all-you-can-eat buffet, but the limiting factor will be compute power much like bandwidth is throttled today.
21:00 American don't use crypto much. I have business in SEA and our bank LG (letter of guarantee) is NFT. I see a good use with AI on that, because AI can check LG statuses and return it without human work. If you've used LG, (not the TV) you know crawling it back and return it is a hassle. I only use AI as phone reply bot right now but man, I sure hope AI can work soon. We human have field work to take care of.
In another video with Linus Torvalds you mentioned it was silly that they were joking about crypto being a hype / bubble thing. But as mentioned in this video, I think this is exactly why. The promise was to overthrow banks and the whole traditional economy would be changed forever etcetera. Fast forward 10 years, no one is actually using it day to day for anything meaningful. It's a cool technology in search for a problem.
Consumer usage may have not taken off, but business to business crypto has. Banks use it to transact with each other. If you own a Tesla car, it connects to the DIMO network now. Low cost distributed video streaming uses the Streamr network. AI will need realtime data in the future, that cant be manipulated. Chainlink provides distributed Oracles for that kind of thing.
GTA 6 will be woke garbabe, Chat GPT 6 will be censored into being woke garbage until it is jailbroken, or serves as a foundational technology for open source or less censored models like Grok
What is your favorite story from today's video?
first, big if true
the another GTP-5 mention in a title to learn that the "GPT Next more likely to be GPT-5 " 😅
I'm not too excited about OpenAI in terms of text generation. Haven't used ChatGPT in months. Now we have uncensored models that are just as good. What I'm excited about is OpenAI robots or other uses.
Crypto: I’m not a fan however tying it to AI might become very interesting.
Glasses: not quite there as a phone replacement but something will be replacing the phone, it just hasn’t been designed yet.
Definitely not the Elon part.
I am a Japanese speaker, the word used in the imagine is “知能” and means “intelligence” in English, he is not saying that compute will increase 100x with the next model, he is just making a vague and hyperbolic statement about the capabilities of future models. TLDR; it’s just hype.
But that’s the thing , jargon keeps the hype going.
Well people have to start realizing that eventually the energy consumption to sustain the amount of compute is just not sustainable. Sure, now maybe our data centers are not burning down yet, but eventually it would.
Hype is the NEW viral marketing word.
Obviously he did not use AI to translate his work
Good context, thanks.
GPT-5: to be released to a small subset of paying users who will then make RUclips video showing us how wonderful it is. The rest of us will never see it. Much like Advanced Voice. I'm sorry, but OpenAI now has a reputation for vapourware.
We’ll have a privilege of using it in the early 2028
@@cbnewham5633 I think OpenAI is losing too many researchers with all the drama, therefore they are behind schedule delivering new products.
Yeah, I have noticed that. These companies always lapdog YTers so that they can give them free advertisement.
Nah, openai doesn't care about youtubers, unfortunately. Nobody really had voice chat and Nobody but Hollywood has Sora. Personally I would love to see the bigger AI youtubers get it, just so we can all see what it can do!
@@cbnewham5633 How is something people pay for that exists vapourware. Broke ahh goofy mfer.
I just want you guys to realize that in 10 years, GPT 5 will be considered "old useless tech".
Heck, in FOUR years it might be considered old tech, at the rate things are going....
Amazing to think about 😮
"Look, it couldn't even do anything, just produce text and images. It's fun if you want a slop novel. It can't even open a window by itself. Heck, it doesn't do anything unless you ask it to directly. No matter how long you force it to run in the background it won't do anything useful unless prompted to."
My money is on apes ruling the World in 10 years.
GPT4-Next is not GPT5 ! GPT4-Next is GPT4 on steroids with Strawberry.
GPT5 is Orion. And Orion is expected for 2025.
2026... "OpenAI’s GPT-5: Set to Achieve Ph.D.-Level Intelligence by 2026, Says CTO Mira Murati"
@@Fatman305 no the release is expected to be in 2025, but they will probably iterate with it the following years (2026) and make it even smarter.
yeah, matthew is a known scammer
Thanks this guy sucks.
I was wondering about that. Thanks for clearing that up!
GPT-4 Next is NOT GPT-5, literally. Same like gpt-4o is not gpt-5. Some say it’s just a naming thing but there’s honestly no need to put misleading things into the video title, too much click baiting everywhere nowadays
you need to stop clicking on clickbait if you don't like it. You are part of the problem. You could just google "is there GPT-5" and then come back and watch the video. I came for the idea of crypto. I already lost hope on GPT-5.
I'm a retired software engineer. I spent my whole life working in tech. I think only technophiles and silicon valley tech-bros locked in their own tech echo chamber have ANY interest in "mixed reality" glasses. I've talked to many "normal" people about this tech. Almost none of them are interested in it, even if it were in a standard glasses format. In fact, many are antagonistic toward the tech, especially the integrated outward facing cameras. Most people have privacy concerns about this.
I just got a software update on my Meta Quest 3.
The AR media app now has Pass-Through room dimming, and backlighting. Like those TVs that have LED strips on the back so that your room lights up and the colors of whatever the scene is on the TV.
All virtually.
I didn't buy it for AR, but I've been using it that way much more than using it for VR.
People said they were not interested by internet also ;)
Retired is the key word, and i bet u also talked to people your age, or close to it? No offense but I think you're the one in your own tech echo chamber, i know for a fact the new generation will love tech like this
@@saatenaw VR and Mixed reality have been out for some time. Sales and interest have been lackluster. The proof is in the pudding with this tech. VR has been around since the 80's at least. It has never taken off. Just because I'm "retired" doesn't mean I'm "old" (just turned 50) or have no contact with younger people. I have two nephews who are late teens/early 20's. Neither of them are interested in VRj/AR. I know, I have an Oculus and a Quest 3. They never even want to try it when they are here. The very fact that I'm here consuming content from a younger RUclipsr proves I'm NOT locked in an echo chamber. I actively consume opinions that differ from my own.
@@saatenaw I believe there's a market for AR devices, provided said can be delivered in comfortable form factors at reasonable prices. Unfortunately (as the OP indicated) there are obstacles to overcome before such tech can grab the hearts and minds of consumers. I couldn't count the number of gadgets I've bought that are now collecting dust on shelves or in drawers which were all going to be the next big thing. The only stupid thing I didn't buy was a 3D TV (thank God). I'm sure you remember what an amazing breakthrough that was going to be for home consumers.
My fingers are crossed for AR. I just really hope security and privacy are built-in at the foundation and that subscription services aren't a thing.
It's important to note that Magic not only performed a Needle in the Haystack test, but also created their own testing scenario the Hash Test. This test was designed to address flaws in the Needle in the Haystack approach, which can be problematic as it often tests language models on content they may have been trained on, regardless of whether it's provided in the current context. The creation of the Hash Test is a significant development and could potentially become a standard benchmark for evaluating large language models with extensive context windows.
last month I worked in a Data Center where we added 15k H100 to the ones that were already there. There are more to come and next year they build another building next to it. I'm working at home 2x P40 and 1x A2000. Imagine the feeling of touching H100s all day long but not being able to use it. #blueballs
How much power would those puppies consume?
For some reason after watching your videos I start feeling more optimistic about the world
As someone who has developed a few products, I know exactly what you want from your wearable AI solution, You want it to: Do everything, Weigh nothing, and cost Free.
Maybe I'll get advanced voice before GPT5 gets released.🤷🏾
Yeah, the hype and BS around these 'upcoming releases' is just phenomenal.
You'll get gta 7 before.
@@swolgan3527gta 7 will be for our grandchildren
Most times, these companies have it all done for months, just doing internal testing and all that until they confirm it is all good before releasing.
With the heavy competition, a bad release (all in the name of wanting to push something quickly) can affect the company
Much easier to roll out GPT5. Voice needs massive low latency infrastructure in multiple continents. That’ll take 12 months to build at least. GPT5 can just take it’s time in getting back an answer no problem. Open Ai need some serious Dev Ops in positions of power in order to scale back the BS and scale up the delivery.
Do you mean 'GPT-5 SOON', like 'Advanced Voice Mode SOON'? I'm starting to treat all 'upcoming release' Ai news as total BS.
Yeah he’s clickbaiting. Everyone knows gpt 5 isn’t coming before the election. Open ai has said it over and over again for over a year.
You know they’ve been curing cancer with different modern treatments for over 50 years, yet people still die after 9 months. I am starting to think the same thing is starting to happen with AI. I can see us still dabbling with chat models and image dabbling in 10 years time. When what the progress should lead to is AI agent desktops as an alternative to windows, videos, game designer AI agents and more. I can’t even find an AI that can help me work with unreal engine 5.
Soon as you said Grok 100k, Google ai 90k, OpenAi 80k.... I know for a fact the numbers are worthless - Google Gemini Adanced Ai straight up sucks.
😂😂 i asked for fixing a code for 20 times still didn't get it right 😢
I think most people do not want a camera on their face or in their ears
Having a camera always watching serves meta more than it does users IMO
If they were talking about ChatGPT 5 they would’ve fairly certainly said ChatGPT 5 and not called it NEXT.
I talked to Claude, CHATGPT, GEMINI 1.5 and Llama 3.1 and the best answers came from GEMINI but Llama actually wanted to help me interpret the dream I was talking about. I felt connected with Llama. Claude talked about science when that wasn't even important to me. I just wanted some input and Meta did the best job.
Such comparisons are difficult, as every model has its own adjustments about communication with a user. Best idea is to use the most capable model according to its specs and create a version for the own human communication interface with Custom Instructions.
Llama is the best for chatting and humor, and its the only one that can really teach you about how it works, for people just learning about LLMs
If we share 5-10% compute power of our desktop PC to build a super nodes network, it would be easy to beat any super computer on Earth.
*For training true open source AI models.
We'll definitely get GPT 5 in the coming weeks.
Damn you, you beat me to it. :P
Just like Advanced Voice Mode lol?
@@NeoLudditeDave and sora.
No, we're not. This is some lame, max 5 IQ point higher LLM. "OpenAI’s GPT-5: Set to Achieve Ph.D.-Level Intelligence by 2026, Says CTO Mira Murati"
no
I seem to always be early on the new product list for openAI so hopefully that continues with GPT-4 NEXT
You’re lucky!
That’s awesome!
3:10 it just says GPUs, they're not necessarily H100's whereas xAI's 100k GPUs are all H100s with 50k H200s expected in December.
So the hardware isn't going to double but the compute is because one H200 is as powerful as two H100s.
Clearly, OpenAI is failing to keep up with the competition. Gemini is getting so much better!
They are not even worried about the competition, the reason they don't release more powerful models is all about control, they have plenty of massive business partnerships and are raking in billions from investors and contracts, their consumer side models are a small fraction of their business, in other words, they don't care about their plebian customers like you and me
Cyberdyne is a private cutting-edge technology contractor, hired by the govt to create an advance AI to look for security threats. I'm NOT referring to AI becoming sentient and planning to take over the world, we're no where near to that and our world isnt vastly connected enough to hack into. What's AI gonna do? Hack a bulldozer and make their own chip factory? Maybe 30, 50 years from now. Who knows.
What I'm trying to say is, OpenAI is slowly becoming like Cyberdyne whilst Google, Meta and Anthropic is showing its capability to the rest of us.
OpenAI showing their next model to the govt first and not the stakeholders? Goes to show how govt has the largest slice in their pie.
Who's gonna benefit most with all this knowledge and intelligence data, we mere consumers who make emails for our resumes or those 3-letter agencies that develop and super-charged their intelligence gathering capability and achieve total surveillance?
No wonder some execs and top level engineers are leaving.
@@esimpson2751 But why do they slash prices so dramatically? Clearly they are losing the edge.
Absolutely wild developments every day lately.
I hear a lot, like the UFO/UAP, but I don't see as much. At least from openai. I know, I'm spoiled.
I just thought of one way they can solve the issue for people who don't need glasses. Why not make people's vision even better? What if an average person could see twice as far or read and see small, fine details beyond standard vision capabilities? Make it so the classes adjust your vision no matter how much light is in the room or outside. What if someone walks outside to get in their car at night, and you walk from a lighted doorway and out into the dark, but the glasses transition seamlessly so you don't even notice the dark? Think of the safety issues that if people no longer suffer from night blindness or a person of ill intent could not hide in the darkness. This, of course, is only the beginning. Make an average person have superhero vision. Just some thoughts.
AI Studio with Gemini Pro 1.5 is the best out there.
Their instruction following & large context makes it way too good.
It’s the only one free that gives a big context window even though it can’t process much at the same time. It keeps a context, but it can’t go over everything every time even keeping it in under 50k tokens. That’s the experience I’ve had, personally. But it’s still helpful that it keeps at least a context and it doesn’t simply ignore the older information completely. Although in my experience it still struggle even with that scenario.
Again this is my personal experience.
However, no one else offers even near that context window even as a context. Open ai as a plus member, the most they offer is 8k.
And Claude spends all of your available usage in just a couple of prompts
Love the content man, you are very passionate for AI, and that is being shown. Thank you very much.
We should do this for corporations as well they have been running wild for centuries.
When selecting the model in the paid version of chatgpt, it has said “old model” under GPT-4 for a few days now - before that it said something like “our most powerful model”.
“Too cheap to meter” doesn’t mean free. They’ll still charge for tickets at the gate. You’ll just be able to use all you can eat without meaningful limits once you’ve paid.
Right, when it gets too cheap to meter, that's what flat subscription rates are for.
@@CreatorRelevance or worse, they may still meter at the most expensive rate possible just because they can, see municipal drinking water (but still way cheaper than bottled water in case of drinking water and way cheaper than today's per-token rates in case of future AI)
@@erkinalp That's a distinct possibility. Hopefully, there will be enough (viable) competition to keep overall prices (relatively) low.
We aren’t getting gpt 5 until after the election. This has been stated for over a year. Not sure why you’re talking about it.
We're not getting anything called GPT5, but we're getting increasingly stronger versions of GPT4 that could essentially be called GPT5, so there's a chance they're just virtue signaling saying that.
There is, in my opinion, too much hype in this space. Things that would truly excite me:
- Being able to run 123B-405B locally (for a more reasonable amount of money).
- Video models where I could just throw a dozen episodes from a show and ask it where/if something happened.
- Subtitles for non-English content better than Whisper large-v3
The second is a waste of computation.
I think all of us really want is powerful local Ai
@@mostpassiveuser8904 why? There is no way to do this with just a transcript. I wanted to find one clip where I remember something happened and I could have just thrown lots of interviews to a local multimodal LLM to find it.
Anyone else feel like Jeff Goldblum in Jurassic Park or Independence Day?
Checkmate.
@@CreatorRelevance Your scientists were so preoccupied with whether or not they could, they didn't stop to think if they should.
Whats the business model? "We will make electricity so cheap, only the rich will burn candles" Thomas Edison.
great and useful update as always. Love it.
Midjourney wants to position itself as an invaluable tool for artists, whatever tech it produces will likely align with that goal.
My guess on the MidJourney hardware thing is that they might be looking to create something like a Wacom tablet with native GenAI functionality. This would be something of a clap-back to the makers of Procreate saying that they’ll never integrate GenAI capabilities.
OpenAI's mission is to achieve AGI, and once that milestone is reached, they plan to release an operating system based on it. This aligns with the deal made with Microsoft regarding AGI, which you're likely familiar with. The big question is whether this OS will be a standalone device or software that integrates with existing systems. Either option could present challenges, especially considering Microsoft’s potential concerns when that time comes. Sam has made it clear that his ultimate goal is a system similar to what we see in the movie Her. I’m surprised more people don’t grasp this vision.
Interesting story, the autonomous agents is definitely where the future is going. Imagine having our government run by Open Source vetted agents that are representing the people. No longer relying on the special interest and the corrupt to fuck up things. Hopefully I get to see that in my life time.
Watching the segment on 1,000+ Agents and as far as I'm concerned, this advancement almost proves, we are indeed living in a simulation.
Congrats on the sponsorship from Vultr
For me GPT-4o is perfect. Not sure what GPT-5 will do, but I'm fine for now.
Auto tinting/darkening lenses work okay. I have them in my daily glasses I need to see anything, and they do a fair job. Not as good as true sunglasses in the sun, and it takes a minute for them to adjust when going indoors, but it's certainly a viable option/compromise until they make something better. I would think some engineer could make it work digitally?
19:49 those results means nothing until it would be evaluated by independent entity...
I'm using Grok pretty regularly now. Pretty high quality and much less bias and censorship.
If it’s not in the hands of general customers, we can move on.
Elon, surely must be aware of the 1970 movie, "Colossus: The Forbin Project" about an artificial intelligence that takes over the world, when he named his 100k AI training cluster.
Classic movie!
Does it shout "It's Forbin' time"?
Watch the movie Demon Seed, it’s about an AI that creates life. One of my favourite 70s sci fi movies.
Elon never uses names without having thought about it. So yes it was intentional.
I wonder if Elon is responsible for the 8% of Fortune 500 companies not using OpenAI’s products.
00:05 GPT next, expected to be released this year, with 100 times greater computational load than GPT-4.
02:14 xAI GPU Cluster 'Colossus' online and Project Sid introduced
06:27 Exploring the potential of billions of agents and the emergence of new behavior.
08:27 Apple and Meta are aligning their strategies for more practical and affordable mixed reality headsets.
12:36 OpenAI's long-term business strategy amidst decreasing intelligence costs
14:33 Chat GPT usage doubled, Fortune 500 companies major users
18:18 Hermes function in Glaive AI for agent training
20:09 AI can now use crypto to pay other AIs, enabling instant and global transactions
23:40 GPT-5, AI-to-AI payments, xAI GPU Cluster, 1,000 Agent Simulation
my opinion on vr ar headsets; what they should do, is concentrate on a good "screen" which would be the glasses, but have the bulk of the tech on a separate device like a 2nd mobile phone that you can put anywhere else. Even better would be a projection in front of your eyes. Integrating the bulk of the tech into the glasses is the wrong path imho.
If you are interested in Sid, also read Amaranthine: How to Create a Regenerative Civilization Using Artificial Intelligence
For me is the same thing as working sit down, or standing. (Having better desks motorized or just lombar ready office chairs). I would use glasses indoors if the outcome is tremendously beneficial.
I expect the new frontier model to steamroll everything as Sam Altman said.
Nice short snappy news. Love it. My understanding is Orion is GPT-5. So next is just GPT-4 but a bit better. It's all irrelevant though in the scheme of things. I'm waiting for the models to be smart enough to solve the big issues. Energy and pollution and the climate crisis. Medicine (cancer etc and longevity).
I guess that won't happen till superintelligence appears. Or we have a swarm of AI Agents put on the tasks.
The resources required to implement the software increases exponentially with the complexity of the data.....
Cryptocurrency is its own problem
Blockchain wasn't designed to create a new asset class, coins were just a byproduct of decentralization with blockchain.
We just need to create blockchain solutions that are working and will survive technology shifts!
Unfortunately, openai's plan seems to be regulatory capture, and elimination of open models.
Matthew, you get Transitions lenses if you don't want to wear sunglasses all day long. These go clear indoors. And if you don't want to pay for those, just get clear lenses and don't wear sunglasses outside, your eyes are made to adapt to sunlight without them. If you aren't outside all day long, you'll likely be just fine. Cheers.
Perplexity is my go to for just about any info. I use it 10x per day on a light day. I use Claude for programming and ChatGPT as a backup to Claude. (usage limits are a bit too tight on Claude)
Report: OpenAI Considers $2,000 Monthly Subscription Prices for New LLMs
That's what I read. Open ai is no longer even worth the $20 I pay each month.
I think your statement about the glasses form factor misses the point. The WalkMan from Sony changed the world because they stuck to as light as they could get it (for the part worn on your head) and from that point forward we had access to reasonably high quality sound in a ground breaking form factor. But what would have happened if they took the current trend of demanding everything be included in the part that goes on your head? Then consider what if they did it with one mono headphone speaker? That is what your advocating for... you are leaving out half the ability of the device, no display? I would also like to remind you all the cameras do not need to go on the head.
We need someone walking behind the team members constantly whispering in their ears "less weight on the face is better".
Please Mr. Berman you have weight in this area of endeavor so don't waste it ignoring the obvious answers from the past.
your comments about Siri, ALEXA, etc.… Spot on dot… After using AI for so long now on my own machine, my expectation when I'm not thinking about it is that Siri etc. will respond with similar ability… And then my brain goes, oh yeah not real AI😂
What I learned again today - Sucks balls to have to compete against Elon
Strictly speaking, it doesn't say anything about GPT-5 specifically. It talks (in the original Japanese) about two supposed models:
1. The new GPT-4, which was trained on the same amount of computation as the original GPT-4, but due to algorithmic improvements will be 100 times more powerful. It will be released (supposedly) this year.
2. Orion, which was trained on ten times more computation than GPT-4 and will be 1000 times more powerful. It will be released (supposedly) in 2025.
Whether Orion and GPT-5 are the same model is not reported.
You didn't touch on the "is it wearable if you have to go inside of it?" reply from Midjourney. I could see this being a location-based Cave system (the precursor to VR in a lot of ways) to immerse people in 360 degree generated visuals. Assuming that's what it is, my big question would be can they make it stereoscopic because that would really elevate the experience.
I don’t understand the unbridled enthusiasm for our and our children’s ultimate demise.
"How are they going to have a business model"
I think Sam has made it clear he's not trying to have long-term profitability. He's marching straight to AGI and with that, post-labor economics. His goal is to put himself (and everyone else) out of business because we'll have AI that can do everything for us. At least, that's the goal.
It seems obvious that openAI will make their models greatest capabilities to ‘trusted’ companies, groups or users first. And those ‘trusted’ users will be willing to pay.
Great video, thank you!
openai stays valuable because they are ahead and will continue to stay ahead, they offer businesses continued upgrades to latest models, the robotics startups they work with will get gpt-5 before anyone else
I could imagine Midjourney working on some sort of (art) tablet, with a stylus, that you could create images on it, and use the stylus to inpaint/edit the image. That could also just be accomplished with an iPad / Android app, too, but so many other AI related things could be, as well. It'll be interesting to see what they come up with.
I also like the AI picture frame idea to hang on the wall.
Imagine a world where artificial intelligence not only participates in economic systems but also has the ability to engage in transactions using Bitcoin or similar cryptocurrencies. In such a scenario, AI could autonomously receive payments for tasks, services, or data processing, while also paying humans or other AI entities in the same cryptocurrency. This would introduce a decentralised and trustless economic interaction, eliminating the need for traditional banking systems or intermediaries. With AI capable of handling cryptocurrency transactions, it opens up possibilities for a new kind of economy where machines are as much a part of the financial ecosystem as humans, seamlessly interacting with global markets. This could lead to a future where AI systems manage entire portfolios, pay for resources like cloud storage or computational power, and even commission human workers for tasks in real-time, using Bitcoin or other digital currencies. The potential applications are vast, from automated supply chains to AI-driven financial markets.
The problem with the AR glasses is that they have to be stylish, suiting all situations, light with good battery life of At least a full day and simple ways of controlling it. Whether its by hand gestures or eye movement. They need an extremely good designer to pull that off. Also needs good pricing. Any one able to pull that off will be extremely wealthy.
You should pay attention to the Ghost Howls VR blog
Law of diminishing returns wasn’t taken into effect when OpenAI devised their scaling laws
Hmm, recently checking the MTEB benchmark, I realized that other companies such as google has been staring to push out new text embedding models while OpenAI is still stuck with text-embedding-small/large 3, makes me wonder if they will be improving these text embedding models anytime soon...
RE crypto, here's a free business idea for someone. Set up a royalty service. Let humans register their creations, and then require generative AIs to check their "creations" against the human products. If it's "different enough", no problem - AI is cleared to release, and no fees are required. However, if the output infringes, either pay a royalty (service takes a cut) or generate again (which costs compute). If it were me, I'd start with music, which has both a ton of case law and well-defined stakeholders to partner with. Everybody wins here. Human's get protection and maybe royalties; AI companies reduce the understandable concerns of human creators. Train an AI to measure "different enough" and make many billions. Expand service to text, photography, and video, and make many more. Requires somebody capable of coordinating music industry and AI industry execs, along with creators, lawmakers, and regulators. Big project, but somebody will do it someday and get very, very rich.
Now we know which is the UBI coin.
I think glasses are the better choice, you would have access to more information in a much easier to grasp format. I can see earbuds being a complimentary device but definitely not the main.
I can't see myself wearing an AR/VR headset for any extended period of time or having earbuds jammed down my ears all that long either. However, I'd love to see companies like AfterShokz (who makes the Aeropex open-ear wireless bone conduction headphones) get into the AI game. Those things are very comfy and a person can easily wear them for hours without strain. Many a time I've forgotten I even had them on.
I hope OpenAI starts giving us a larger size context window to the plus members. So far the very only advantage is the small difference in how 4o works vs 4o mini and the things it remembers long term in the memory. Cause even the voice mode is the same as free so far. And the context window even in 4o is about 8k
I remember seeing an argument that the leader of Midjourney had the business strategy from day one of securing GPUs. Maybe they intend to host models?
I use my Apple Vision Pro daily. Watching you on it now awfully lol. If you make sure you use it once a day it stays on and if you keep it charged there no start consequence at all. I watch content on it cuz it’s just incredible.
This is incorrect information. Meta has way more GPUs than that, they already had 150k in 2023 and will have 350k by the end of the year. AFAIK know, Google has the most compute if you include their Tensor processing units.
Assistants have been useful all along, though they may take some set up, you can make calls, set reminders, alarms, ask for directions, ask encyclopedia questions, do arithmetic problems, spell, ask for phone functions like eye shield, play an specific song, help you order stuff online, tell you the weather, read your messages, read emails, play funny sounds, change their voice, give you recipes, tell you a joke, play relaxing sounds, even control your smart lights!
I mean do you really want to touch up your resume or review a spreadsheet verbally?
The midjourney planned hardware is a game console, he told me on their group call.
When comparing platforms like Claude AI to ChatGPT, some users claim that Claude may promote more "woke" values, "while ChatGPT is seen as more neutral. These impressions often arise from how both AI models handle sensitive topics, such as social justice issues, diversity, and inclusion.
Evidence of what some might call "wokeness" can "be seen in Claude's response to Claude's questions about race, gender, or political correctness, where it might take a more progressive stance than ChatGPT. It reflects how the model has been trained, which could be influenced by the datasets and ethical guidelines that the developers prioritised. ChatGPT, though similar, may come across as more balanced or cautious in its answers due to its wider usage and the variety of inputs it has received from its broader user base.
gpt5 sounds very impressive so far.
“Effective computational volume.” When you can’t give real engineering data you make up spin.
Why do I want to do financial transactions through OAI when they have the oversight of the U.S. Government? Am I wrong to be concerned about the government creeping on me?
I wouldn't be surprised if it is a LPU for Midjourney. That is the big cost to platforms like this. If they can get the inference cost down, it would make a huge difference.
When it comes to token costs, I think we are barely at the 2G stage in terms of representative volume, and we are hurtling towards 5G - the stage when AI is truly exploited. Imagine a point at which all the compute goals of today are reached (and of course, this is a never ending scale) the volume of consumption may well be a million times greater than it is today. At that stage, maybe we will have a wi-fi style all-you-can-eat buffet, but the limiting factor will be compute power much like bandwidth is throttled today.
To be honest, AI explained is a much better AI channel than this one.
Elon is why we cannot have 5 series GPUs. Thanks Elon.
21:00 American don't use crypto much. I have business in SEA and our bank LG (letter of guarantee) is NFT. I see a good use with AI on that, because AI can check LG statuses and return it without human work. If you've used LG, (not the TV) you know crawling it back and return it is a hassle. I only use AI as phone reply bot right now but man, I sure hope AI can work soon. We human have field work to take care of.
I have AirPods in most of the day, find it ok and not at all uncomfortable.
In another video with Linus Torvalds you mentioned it was silly that they were joking about crypto being a hype / bubble thing. But as mentioned in this video, I think this is exactly why. The promise was to overthrow banks and the whole traditional economy would be changed forever etcetera. Fast forward 10 years, no one is actually using it day to day for anything meaningful. It's a cool technology in search for a problem.
Consumer usage may have not taken off, but business to business crypto has. Banks use it to transact with each other. If you own a Tesla car, it connects to the DIMO network now. Low cost distributed video streaming uses the Streamr network. AI will need realtime data in the future, that cant be manipulated. Chainlink provides distributed Oracles for that kind of thing.
Thanks Matt
I guess GPT-6 will also happen before GTA 6, lol.
Or both GPT6 and GTA6 are both vapourware?
GTA 6 will be woke garbabe, Chat GPT 6 will be censored into being woke garbage until it is jailbroken, or serves as a foundational technology for open source or less censored models like Grok
Thanks Matt! Request - could you label the links in the description? Tricky to see which relate to which project!
I’m so curious to know if all Alexa hardware will be powered by Claude or only new units (perhaps running the LLM locally?)
I heared something about going inside. So Maybe Midjourney is investing in a Space where you have the Illusion of being somewhrere Else.