@@clearsight655 seriously lol hahaha it’s always the thing said and I’m like really? Is there nothing else? Plus yeah these days most people watching these vids aren’t able to afford so many flights they need to deploy agents imo
@@brianWreaves something tells me it won't do the hard part of finding the best route/price, and then filling out the forms etc. It will just do what Kayak does and find some flights for you.
But this was all huge news! I too am tired of those intros from here and Matt Wolfe, but then I listen to it all, and I’m like… ok yeah, that was huge news. No? It’s crazy how week after week I keep getting shocked and I study this shit near full time every day
@@entropystar80231 Almost all YT thumbnails / titles are clickbait nowadays. It probably helps with views / money, but doesn't necessarily mean the content is crap.
Agree, i'm just surprised that Matt didn't understood yet what "apple intelligence" really is. The ability to sell an inferior product at a higher price just by mastering the propaganda, and they do it again and again, 16 count so far.
If you put your content into the public space it becomes public knowledge. There is no difference between a person summarising public information and an AI doing it. The argument is a non sequitur.
The issue is the level of access. If you meet Joe Blow at an event, he currently can't just snap a picture of you and get all of the negative details about your life. But that is coming soon.
We also have a brain with a cerebral cortex that has an estimated 60-240 trillion synapses. The human brain is organized into neural networks of 100 neurons (probably around 700,000 synapses). It has 300 million of these networks, organized hierarchically. The current largest LLM we have has an estimated 1.8 trillion parameters (function as synapses). It could probably not drive a car safely on all roads using only cameras for input. If you wanted to try it anyway, to run inference, for one vehicle, you would need in the region of 13-25 server racks, pulling 70-140 kw. Do you think Tesla could fit that into a car? Tesla FSD is nothing but hype, designed to make one man (narcissistic autist) as rich and powerful as possible. Waymo is the real deal.
@@antonystringfellow5152 Wait, you think with cameras you would need 25 server racks to control a vehicle, but simply by adding Lidar it's suddenly easily doable for an onboard computer? You paid by Google or something?
I think that we will have more sensors than cameras when Tesla FSDs have accidents another sensor would have prevented. The latest version of FSD is very capable even with just cameras, which is very impressive!
I was doing some research of course with Chat GPT, about Tesla's vehicles and I found out that at one point in time they were using directional sonar in order to scan the road but it was too expensive, I told them instead to just put regular directional microphones on the undercarriage of the car and just use an AI model to determine a larger map the cameras can't even see. It would be cheaper, would probably break less, and it would add a layer of security. Human beings do not just use their vision when they drive. Your hearing is actually really important when you're driving.
I tweeted elon, or x'd....and told him to just doing flying cars that use google maps, so that they are flying the exact same pathways and you just lock them in. done. im going on break
Great content but want to recommend revisiting your self lighting set up. Consider a ring light camera config-ideally you want self lighting nearly straight on (not three point lighting, which you want for your set/stage) and angled slightly above eye sight line on your face so the light casts slightly downwards on you. This is what we do for models on set, removes shadows on the face.
Long before Anthropic and Google, a couple years ago I built a JARVIS-like bot to take screenshots of my display and use various algorithms with YOLO to fully control my pc. Not difficult but extremely time-consuming for me at that time. It's actually easy nowadays with the new hardware processing power hat yolo can run on.
Yeah and probably way better. Perplexity is just a gpt reskin. 50% of the time with more in depth research questions it just gives up and says ‘listen dude I’m just a openai api call, cant do much for ya”
Matt, do you ignore the announcements? Image generation should come with 18.2, other features with 18.3 and 18.4…. And - EU will get support with 18.3 or 18.4 and it was never the AI Act or DMA, although Apple claimed this publicly. To port this to several languages is the real problem. Honestly: even 70b models do have serious troubles to “understand” German and other - more complex - languages… The promise was a public beta with 18.1. And this was done.
@ what has the taxes thing to do with AI act, MDA and other data protection laws? No company wants to pay taxes anywhere, it would be a miracle if Apple was different. 😁 But as I saw on my invoice, VAT was calculated and paid.
💥THE BEST AI NEWS CHANNEL - Congratulations, Matt, you got a news format that has tons of the most important news, concise script, no lose of time like Mattvidpro, which talks so much nonsense. Your format is for us, people which don't have time to lose, and need the latest news with short examples, and comments beforehand, your experience in using them, so that we don't lose our time on bad apps. Continue this format and the pace is excellent. Tons of useful information. A real masterpiece on how to make good RU-vid videos. 🎉❤❤❤
Apple leasing an AI model and rebranding it as "Apple Intelligence" is the most Apple thing of all time. This company truly stopped innovating when Steve passed.
re: COPYRIGHT @ Matthew: 1. If a student were to go to their local library and watch your videos/read your articles as part of their PhD research, how would you feel about that? 2. If you were one of hundreds, thousands or even millions of authors and creators the person learned from, how would you feel about that? 3. The student earned their PhD, and creates some new technology for the world, eventually writing a paper of their own. You MAY (or may not) be cited depending on how much content came from you, uniquely. How would you feel about that? 4. Assume that the student has eidetic memory....does that change your response? Personally, if a model derives a significant portion of its response from a single source, then it should provide references to the sources. Incidentally, I would like to see references to give these models credibility...especially where research, laws, facts, etc. are used to give a response weight. Otherwise, the idea that ONLY humans may derive new works from the existing works of other humans will never endure. At what point do humans have to start providing references to genAI created content? Haha.
@@lordnikon6809thx. i stay for a few reasons for the iphone including mainly the ecosystem. if you know you know. i am in punta cana my favorite place other than my home in new jersey 🇺🇸
The camera only argument makes no sense. What is wrong with cameras PLUS other sensors if the other sensors help with limitations of cameras (like the example you pointed out), ignoring costs (which will probably come down)
Vision only systems have to deal with sensor issues. Like from the sun. Self driving systems should at a minimum include front and rear radar modules. Probably ultrasonic sensors as well (they are cheap). LiDAR seems like a nice to have, not a must have.
@matthew_berman when it comes to AI and your own content, I mostly agree with your statement about "I want the option to decide if AI ingests my content or not", but I would add one extra step to that. I think the rule should be, you should be allowed to choose if AI ingests your content or not, but by default the option is for AI to NOT ingest your content", so that way your content is protected by default and you can choose to allow it if you wanted to. At the moment the default is that the AI can just take your content, which means it might already have ingested your content before you even set the option so that it isn't allowed to. The same sort of thing used to happen with Google when you published a page, but you hadn't setup the robots.txt or meta tags yet. Google could add your page before you told it not to.
We dont only have eyes. We use our ears and can feel the road beneath us and we have social understanding that allows us to interact with human drivers. But even ignoring that, why limit yourself to just our senses when we often start off acknowledging that human drivers arent great and we'd prefer better? Why not use new senses that accommodate for where we fall short? If we want to improve as much as possible, why allow our humanity to still be the asymptote? Idk thats where im at on that
It's funny because I just saw a video on Reddit where a Tesla on fsd just plowed through a deer without stopping because it's cameras couldn't see far enough in the dark to figure that out. Yeah, we need some other sensors too to augment safety! 😅
I think glasses are the best form factor already? They are super easy to wear, and already exist in society. The best way to get a constant display in your vision is to wear glasses? Everyone acting like glasses are a bad form factor like there are millions of people who already have to wear glasses. How do you think they feel? 🤣 It's already a tried and true form.
I think the claud update for controlling your computer is a massive deal due to it could solve the context window problem for large programs. Progressed in the rite direction this could be the Deven tool we wanted.
Installed it yesterday hoping for decent Open Ai like voice chat with Siri and it's actually worse than the old Siri for me. it now hangs on 95% of questions I ask it, Siri just sits there with the icon showing it's lisening but nothing happens. Underwhelmed is accurate.
What a dilemma for Google. "Google" the search engine product - and the brand - would have to attack itself with an Agentic search tool, because when they do that, they will cannibalize their own legacy search engine. And its revenue model. The reaction to agentic search will be very dismal if the agent compiles great answers for you and liberally intermixes them with Ad services. How will they resolve that dilemma? (Answer - they might not and the future of google might depend on something other than search...)
I’m pretty sure they have a solution ready to go that they are just waiting to release, but are doing a cost benefit analysis given the 9x higher cost to deliver results.
Leading scientists in the field of autonomous driving don't see the approach of Tesla with vision only as too promising. And to be honest most of us could probably agree that our vision is quite inssufficient in challenging weather and light conditions...
wait, the TSMC "production record", was that after a lot of employees quit and they were (are) having a major hr shortage? when was that comparison made? 🤔
Don't understand why browser control is considered difficult, not why screenshots (and mouse "emulation") are used. (chrome/firefox/edge)-devtools-(frontend) Eg (ie) chrome-devtools-frontend Panel extension, websockets, code injection (inject a repl), all the access to browser, the dom, events, the console, local overrides and the PC it is running on, through (all of) the browser, browser extension and devtools apis.
I completely get your point about not wanting AI to copy your content.... But the tricky challenge with AI is that it is not a scraping bot, but more like a person (brain) watching your video, it will not copy paste, but rather assimilate and remember key points just like humans do. Regardless of how powerful and talented that brain is, is it fair to block it from watching what you have made public to other humans?
Apple has historically committed to only releasing things when they are really solid. Remember how long it took for cut and paste? I don’t think they will do the nice fast and break shot approach that we saw did not work well for Google…
re: copyright... My feeling is that if it's legal for a human to do it it should not be illegal for an AI to do it. Just because the LLM is faster and more comprehensive shouldn't make something legal, illegal. So if it's legal for me to go to your YT channel, consume your video, and then write my own take on it in a blog, I don't see why that same activity would be illegal just because an AI does it.
The thing is fixed cameras may not provide enough info for autonomous driving. Digital sensors have always been worse than the human eye. Driving requires focal points, dynamic exposure through adjustable aperture. If these camera were DSLR's with pivoting motors it could make sense. The most important thing in driving is looking were you are going. Its vary important in motorcycle riding because momentum keeps you up so anticipating and looking into a corner is needed to not looking everywhere and driving forward.
That being said the vehicle and computer can project physical properties like speed, steering position and anticipate position but without strong depth perception it would be stuffed knowing were the most important place to look is.
I have no idea how you all can stand how bad perplexity hallucinate. Every time i have tried to use it. It made up so much s it was infuriating to work with. Don't you not even notice that it hallucinates? Or how can you stand it?
I'm not using google much at all now. It's chat GPT 4.o. It can be a bit sassy and keeps ending my questions with a question of it's own. I told it that asking questions all the time at the end is not normal communication. It didn't give a damn.
Notification rollups are definitely a standout feature, but Apple's AI feels like a letdown compared to the hype. The false advertising really made it seem more revolutionary than it actually is. Hopefully, future updates deliver something more substantial. Anyone else disappointed by this?
Spoken like someone who doesn't know what to do with AI right now. I have the opposite problem, theres too much useful AI out there, and still much room for improvement. We are only at the start of the bubble, and while many projects will fail, many new fortunes are yet to be made. Mark my words.
AI starts clunky then beats you up if you don't use it after a few months. I really appreciate the snapshot of where it is but I hesitate to discount it yet
We don't yet have close to the intelligence in AI required to drive a car safely on all roads. We can do it but our cerebral cortex has about 100X the scale of the largest LLM and is several orders of magnitude more complex in its structure. It's made up of neural networks containing around 100 neurons each (up to 700,000 synapses in each network) and it has about 300 million of these networks. If you think the smartest LLM we currently have could achieve FSD (I don't) then all you need to do is figure out how to fit in the region of 13-25 Nvidia server racks, along with a 70-140 kw power supply into a car and you're ready to go! Good luck with that!
Yes, let’s race toward every conceivable AI tool and product, turning our entire technological landscape into a Nexus system, accelerating headlong into AGI. I may not have a PhD, but I can confidently say that this path risks the very end of our species. Apple prioritizes safety over profit when it comes to AI, and there’s solid reasoning behind certain features not being integrated yet - primarily for security and safety.
15:23 We only have eyes for sensory input?? When was the last time you looked in the mirror? Humans have something like 14 senses. You know the big 5. But for human navigation, we also have equilibrioception (balance, spatial orientation), proprioception (body position and movement), nociception (pain), thermoception (heat), mechanoreception (vibration), interoception (internal bodily states), and echolocation (sound reflections, rare).
18.2 is supposed to bring more "Apple Intelligence". However, this was never going to be the seller that Apple hoped it would. I'd much rather keep my AI out of my OS.
While I am watching this, I get a notice that Claude can now run code? "We are introducing the analysis tool, a new built-in feature for Claude ai that enables Claude to write and run JavaScript code. Claude can now process data, conduct analysis, and produce real-time insights. Think of the analysis tool as a built-in code sandbox, where Claude can do complex math, analyze data, and iterate on different ideas before sharing an answer. The ability to process information and run code means you get more accurate answers-building on Claude 3.5 Sonnet’s state-of-the-art coding and data skills." You are not done working for the day, lol. Ok, lol. I see further on, you talk about that a bit.
I’m kinda selfishly sad that MS is going the platform agnostic route, as that was one of the main selling points of the AI tools we have been building. Now I have to stop telling people to avoid MS for this reason. Hmmm
16:59 Has Google not learned anything from Microsofts Recall's sh*itstorm? Who in their right mind would allow Google to continuously take screenshots of their desktop, upload them into the cloud and give an AI permission to make unsupervised purchases with their credit card?
Sorry, but putting OpenAI's Advanced Voice Mode instead of Siri would mean giving OpenAI access to *all* of your private data and that is an absolute no-go.
Yes, not impressive for those of us accustomed to a dedicated ai. But i think apple intelligence will gradually find its place … snd, for instance, there will b convenience from not having to go off to a different app for basic ai functionality
10:45 If the information is out there, "FAIR USE" starts to take hold. If someone is simply reporting that you made a video with certain information, it is no longer copied, it is a derivative. It doesn't matter if it's used in AI or a News blogger. Same with art. If it learns to "paint" from looking at a Bob Ross painting, it's no different than my 5-year-old doing the same thing. NOW..... If it simply regenerates exact replica's of art or content, there is an issue. But this isn't happening.
what's so special about the Claude update? it's not running the code server side, it's just another client side "artifact", right? It generates javascript but running it in the client machine (browser).. Nothing really new here, it's still limited to ReactJS or plain javascript/html, no other frameworks or libraries etc., still limited in context size (the csv that you "upload" just to be pushed back in your browser for the javascript it generates).
I paid good money for an iPhone 15 just for Apple Intel So far Extremely unimpressed All I want is Advanced Voice as Siri All I want is Advanced Voice as Siri All I want is Advanced Voice as Siri
the meme about companies is fun but misses another reality. it's no use to burn cash with 10 people searching everywhere in the wrong place, it's more efficient to have 1 person digging at the right spot so that the 10 others will all focus on their specialized task when they are required :)
hm.. on the one hand you say cars should just drive by vision like us, but then you say Chromes AI shouldn't act just by vision like us but read the code ... Your car argument is perfectly applyable here too ;P Reading the webside's code is good for now, but outdated for tomorrow
"we only have eyes.....why can't we get networks to do the same?" well, human vision works around 28 fps, should we then limit cameras on cars to that frame rate just because it is what we have? Wasn't the idea with autonomous driving to make driving better and safer? It is a bit silly to think you can get better by trying to imitate the way we humans have been driving up until now.
