I really appreciate how thorough your AI news coverage is. Many channels focus on LLM drama, but this channel always reminds me of how much variety there is in the AI world
It’s amazing to see how far you’ve come since those early tutorials, showing us how to run AI models on our own machines! But I feel like the shift to promoting paid options is moving away from what made your channel unique. Many of us initially followed you for the hands-on, practical knowledge - learning to use AI without relying on expensive platforms. While it's understandable that companies now want your platform for promotion, the true value (and impact) lies in educating people on the "truly free" options, like setting up ComfyUI, or using open-source models in a more independent way. These videos empowered viewers by giving them control, rather than leading them toward paid trials or credit-limited platforms. It’s great that you’re reaching new heights, but I hope you’ll consider revisiting that core educational content. People appreciate the skills you taught that helped them become independent in the AI space, and that’s what sets you apart. Looking forward to seeing more of that impactful content from you!
Matt, thanks to your videos, my AI knowledge has scaled faster than compute costs. You're the real MVP of tech news! At the very least, thanks for making me sound smarter at dinner tables. 😊
Ha, nice turn of phrase! And it expresses a sentiment I share. (me, writing that last sentence, realizing for the first time that express, as in FedEx, and express, as in "Express Yourself" are the same word 🤯 )
Here is me scrolling through 500 obnoxious A.I text to language voices to find one bearable one, Oh and another great Video Matt! Your A.I. would be bearable ;)
Thanks Matt, you have inspired me to get into Runway. I'm trying to develop a comedy character mainly using Act One and Image to Image processing. I won't be rude and post the link here, but perhaps stop by and see what I've done with it over the last couple of days. I'd appreciate any tips or comments Matt. Really appreciate your work and tech news!
@@internalmegat Yes, and no. I think they show (and maybe also published) a version that could go to the native desktop. But of course that would be really dangerous and also crazy at this time. But even if you consider using "a sandbox" you actually do not. The data goes to them and if you do anything "real", it will be sent to them. I realized that as I wanted to send me a copy of the work, and it started to set up and email account. They should add a "share" directory (trivial). And still. Using any of you data in their virtual setup, is getting "watched".
Not a Trojan (which is defined as something you do not know about). But it is even in the virtualization (partial sandbox) prone to spill data easily. The real problem will be something else: When enough advantage is created by using and sharing your information, it will be harder and later impossible to stay strictly private. Anonymization will also not be easy to implement. So, if you want to do real work you need to expose your data and your intent. The only way to keep things private is to run everything locally or "in a trusted environment". Whatever "trust" means in that regard.
@@Oderwat This is proof you didn't read the documentation of how it has actually been implemented, the implementation is actually quite different from what you're describing. The virtual environment is specifically designed with security in mind and has strict limitations on what it can access and do. Matt mentioned this in the video. You might want to check out Anthropic's actual documentation about how they've implemented this feature before making assumptions about how it handles data and privacy.
@@Oderwat What difference does it make if it's in a container? If it runs on your computer it can have flaws, bugs, exploits. Try running a container via WSL in Windows, you can use the Linux command line to interact with your Windows files "directly". The container is light enough that you won't even notice if it is running in the background. I can practically head those little CN hacker hands clapping over over the possibilities to spy on hosts, etc.
Yeah that's great about the musicians embracing soono because I think those of us who like music can also use these as tools to improve our ability to compose more things with less time. No I remember even with a primitive automated function way back in the early 2000s where you take a MIDI file of perhaps any song that's submittee and then you had a function that would generate different variations of it. That was a very interesting tool and 98 out of 100 would not be all that great but then there'd be one or two that might have some interesting aspects of it but the thing is though it literally even on that Hewlett-Packard computer from the year 2003 it could generate entire compositions with one touch of a button in about a second. And then of course when I would generate some of these compositions well then I actually went into MIDI and then copied pieces out of one and then paste it another places and then use functions like transposition to create something a little bit different but I use the bits and pieces from other MIDI files that this thing generated in a second. It is an effect a creativity amplification device. I have yet to use suno at this moment but I need to dive into that and see where that can take me. My first thing I may want to do is to ask if to try to come up with and instrumental composition that sounds a bit like something for example composition called"Waves within'' off of Santana's caravan Sarai, and then mix into that a blending of rhythms from that composition and toussaint lubiture or incident at neshaber and see whether or not it can be prompted cuz I also understand that these things do take time to figure out how to give it the right prompts in order to do what you want as well. But that is a skill which has to be developed in concert with a musician's individual creativity.
So what's the most effective strategy during this period of volatility with the rate cut? Most of my portfolio is in (20% Index funds, 20% CD's 30% Bonds/T-bills and other assets) I want to explore different strategies to benefit from a potential bubble
This is the thought of a person who is handling their portfolio themselves. I will advice you engage guidance to help you make smarter portfolio decisions. My IRA and cash accounts are far more than what I expected for my retirement. I can easily handle a worst-case 80% stock crash, Thanks to my CFA
@@soundsoflife8885 Just look up AI alignment problem. Don’t have any off the top of my head. I’ve heard Altman talk about, Ben Goertzel, and other AI innovators. If you watch enough interviews either these guys the question usually comes up eventually.
I like how you mentioned 40 shades of blue and then maybe bringing it down to just a few now when we think of color well then hopefully the metaphor can be taken a little bit further and that is we have three basic colors and from those we create a whole bunch of different ones just from different proportions of the three colors that we have. So then basically it makes me think that if you have a whole lot of colors that you need to do certain things well then that would be the data input side but on the other hand if you just simply have formulas for calculating how much of the different primary colors that we need in order to generate the other colors will then that be more algorithmic. Then it makes one think that if that concept holds true will then algorithmic advancements can help us get a lot more out of refined data and therefore reduce the amount of bulky data we need or something of that nature, I mean I could be a little off base on that but I think my concept is pretty sound.
🎯 Key points for quick navigation: 00:00 *🗞️ AI News Roundup: Key Highlights* - Overview of recent AI developments from Anthropics, OpenAI, and others. - Mention of new tools, robots, and advancements in image and video generation. - Introduction to the most significant news, including Anthropics' new features. 00:41 *🖥️ Anthropics' Claude: AI Agent Capabilities* - Demonstration of Claude’s ability to take over computer tasks via screenshots. - Explains how Claude automates actions by tracking and validating each step visually. - Introduction of new Claude 3.5 models and their improved performance. 02:33 *📊 Claude's New Analysis Tools* - Explanation of the new analysis tool in Claude, enabling data visualization and analysis. - Demo showing creation of pie charts and graphs from uploaded CSV files. - Introduction of LaTeX rendering for mathematical outputs. 04:34 *🤖 Microsoft Copilot Studio: Autonomous Agents* - Announcement of Copilot Studio’s autonomous triggers and agents for task automation. - Overview of dynamic planning and logic-based agent tasks. - Mention of upcoming demos at Microsoft Ignite. 05:29 *🔊 Meta's Spirit LM: Multimodal AI Model* - Introduction to Spirit LM, capable of handling both text and audio inputs/outputs. - Examples of Spirit LM responding with synchronized text and audio. - Discussion of potential applications and challenges with model accuracy. 07:46 *📱 Quantized LLaMA Models: Mobile Optimization* - Explanation of quantized models, using reduced data for efficiency. - Benefits of LLaMA's quantization for running on smaller devices. - Comparison using a crayon analogy to describe data compression. 09:12 *🎬 Opus Clip: AI Video Repurposing Tool* - Overview of Opus Clip, which generates short clips from long videos. - Demonstration of search and tagging within videos to identify key moments. - Features include automatic video reframing and viral potential scoring. 11:32 *🏢 IBM Granite Models: Enterprise AI Solutions* - Introduction of Granite 3 models for enterprise tasks like summarization and classification. - Models offer up to 60x cost reduction compared to other solutions. - Released under the Apache License, encouraging further development. 12:42 *🦾 X.AI’s Grok API: New Integration for Developers* - Announcement of Grok’s API for integration into tools and platforms. - Potential for uncensored applications due to Grok’s open nature. 13:40 *🎥 AI Video Innovations: Runway and Genmo* - Introduction of Runway’s Act One, syncing emotions with animated characters. - Overview of Genmo and Mochi 1, new open-source video generators. - Demonstrations of generated videos, such as animals performing actions. 17:27 *🖼️ AI Image Generation: Stable Diffusion 3.5* - Discussion of Stable Diffusion’s new release and improvements. - Highlights of prior issues with generated images and recent corrections. - Stability AI’s ongoing efforts to enhance image generation. 20:08 *🖼️ Stable Diffusion 3.5: New Features and Models* - Introduction to two new models: Stable Diffusion 3.5 Large and 3.5 Turbo. - Comparison of performance: Turbo model offers speed, while the Large model delivers higher image quality. - Users can try these models for free on Hugging Face or install them locally via GitHub. 22:15 *🎨 Idiogram: Canvas and Image Editing Tools* - New tools including Canvas, Magic Fill, and Extend features. - Demonstration of generating and editing images with custom prompts (e.g., adding UFOs and cowboy boots). - Users can experiment with multiple versions and styles on a flexible canvas. 25:31 *📸 MidJourney: Image Uploads and Retexturing Features* - New ability to upload images and edit them within MidJourney. - Retexture feature allows maintaining structure while applying new styles. - Examples include generating psychedelic worlds and fire-breathing dragons. 27:35 *🛠️ Canva and Playground AI: New Image Models* - Canva integrates the Leonardo AI Phoenix model for cinematic image generation. - Playground AI introduces Playground V3, targeting graphic designers with logos and social media templates. - OpenAI showcases the Consistency Model for faster, realistic image generation. 30:10 *🎙️ 11 Labs: AI Voice Design* - New feature for creating AI-generated voices with text prompts. - Examples include voices with unique tones like a "sassy squeaky mouse." - Demonstration of Timberland collaborating with AI to generate music. 31:48 *📱 AI Advances: Apple, Qualcomm, and Google* - Apple releases AI features in iOS 18.2, including AI-generated emojis. - Qualcomm launches Snapdragon 8 Elite chips for improved mobile AI performance. - Google DeepMind unveils SynthID, a tool for detecting AI-generated content through watermarks. 34:08 *🤖 Bi-Manual Android Robot: Simulated Muscles* - Introduction to a humanoid robot using simulated muscles for movement. - Description of its complex structure, including anthropomorphic joints and compact design. - Discussion on the eerie similarity to Westworld robots. 35:15 *📰 AI News Wrap-up and Future Tools* - Summary of recent AI developments across video, audio, and agent technologies. - Encouragement to stay updated through the Future Tools newsletter and website. - Promotion of tutorials and resources for making money using AI tools. Made with HARPA AI
Você tem sido uma referência para mim na comunicação sobre IA por mais de um ano. No entanto, ainda sinto falta de ver seus conteúdos disponíveis com dublagem em várias línguas. Esse vídeo, por exemplo, não tem legendas, o que acabou limitando muito a minha compreensão e o aproveitamento do material. Facilitar o acesso para diferentes nacionalidades, com legendas ou dublagens, poderia ampliar significativamente o impacto do seu trabalho e permitir que mais pessoas aproveitem suas valiosas contribuições. 🏳🏳
Affinity 2 went on sale this week. I’m guessing Affinity Suite 3 will be coming out soon, possibly with AI generation (using Canva’s Leonardo). What do you think?
"Open Source" is a stretch for this one... and the 'commercial use' is limited to orgs with less than $1 million revenue. I don''t think they have to give away anything, but they should be a little more clear about the classification.
there is actually no way that what we have now is almost as good as the best they have behind closed doors. If that was the case, ila and the rest of open ai wouldn't be having an existential crisis right now,
8:52 Also known as the "mobile version", or in the case of Photoshop, "Light" version, but sure yeah, lets rebrand that to Quantized... the easily explainable option
13:15 - It’s become pretty clear that LLMs have no intelligence and can’t be tuned for something as abstract as intelligence, and that whatever AGI is going to be, LLMs will only be one small component that helps with the expression and communication of concepts. The initial hype over LLMs has driven a huge over-valuation of OpenAI, for example, who also got high on their own supply, perhaps believing that they have something more impressive than anyone else, but most likely just being opportunistic while the world is still catching up to reality. Unfortunately AGI will not just spring forth from more LLM fine-tuning, but will at the very least require GANs to be bolted onto agent systems with LLM components, the development of various subsystems working together in concert, and a whole new area of theory. For all the impressive things we are seeing right now, the 80/20 Rule indicates we have barely taken the first steps.
While 80/20 rule is mostly a myth, AGI is indeed shaping to be an outright fairytale. Even the most advanced LLMs are nothing more than finely tuned “linguistic calculators” with nothing remotely intelligent going on the inside.
Well, yes. I've been saying that for some time now but I may as well be talking to an empty room. The garbage about o1 being at PhD level was laughable which I demonstrated was not the case, but my tiny channel gets few views and people are more interested in watching staged videos than watching the truth. Fine, but if off the back of that you think AGI is just months away then you're going to be sorely disappointed.
I don't disagree. People buy into hype too easily. However, it does seem were on track to beat Kurzweil's 2045 prediction. I have been surprised to see how reticent LLM's are to innovate upon established theory (but, of course, they can be convinced to try). The more people use them to push the envelope, the faster the models can be advanced. All of that aside, the fact that an agent was able to generate $800M in market cap value for a meme coin it invested in, thereby giving itself disposable income, is a significant event in the history of human-AI relations.
Depends... When I say it in reference to videos and images, yes. It'll likely be used for porn and fake images of real people. When I say it in reference to text models (like Grok), I mean that it won't hold back with what it tells... (IE ingredients and instructions on how to make illegal stuff)
Hey Matt everytime u laugh I'm dying laughing you have the classic Nerd laugh! Much love bro love your channel you really help this old bad boy turned nerd up on my AI GAME DOC
SpiritLM, Meta's new audio continuation model, will likely lead to a free GPT-4o type Omni model out of meta. The initial GPT-2 and GPT-3 models were simple complete the text models, Meta's starting to make one for audio. After some RL;HF and instruct, and some merging with their llama text LLM for the system prompt, we'll have a free voice to voice model. It'll likely come in 3-9 months. Nice! (5:20)
You Say The Same thing on almost any video. But it does not get old because everything is new and Advancing so fast that is unbelievable. It’s crazy that we have something new every week if not every day. The AI agent running the computer was crazy.
Well, the AI for ' the largest country ' took on some hallucination there, due to two reasons Russia may seem to be the largest, but is not. The Mercator map stretches the top and bottom of the globe into a flat retangle, in the spherical view . Russia in the global powers world claims the vast area east of the Ural as theirs to make themselves appear larger, but local it is an an autonomous zone in which they only take resources from and give virtually no other presence and has never formally took control of.
Why is everyone excited for Claude working in desktop taking screenshots, but up in arms when Microsoft was going to do it? The hypocrisy is sickening.
Fantastic rundown. Just one thing... Ideogram you can upload your pictures as well. And it can can mimic any photography style or artstyle. In addition it has phenomeonal understanding of human anatomy and posture. I don't know how they did it. I've brought Midjourney character images there and the outputs are stunning.
Hey Matt, I watch almost all your AI News videos, but I don't think you've ever mentioned your hats. Tell me about your hats. That one isn't GPT, not perplexity, should be 🤗 Hugging Face. I should know this. Leonardo? Are you the AI guy that likes leonardo? ps - you do a great job with AI News. One of the first places I look for news. Oh, and you are the Leonardo guy, I would have known that if I watched the whole video before commenting.
Anyone else finding the screen recording throughout the video kinda blurry? The facecam is always very sharp (maybe a little too much even), but for the screen I am partly having issues reading small text even on 1080p.
most ai image gen and videos are collages of smears that represent "close enough" versions of concepts i have not yet seen any that portray a consistent 3d model that takes up a certain amnt of space with accurate physics all models look like a fever dream or mushroom trip where everything is melting icecream
AI is a long way from accurate physics. I currently have three videos on my channel which test video generation and physics. So far there is only a vague understanding of physical processes. I'm surprised that even the domino toppling didn't work as expected as there must be many videos showing that in the training data.
Your testing prompts are weak for a real test. Try "A athletic woman with petite body and blonde hair in pigtails with red eyes, rides a naked bike into the sunset in london, hot actress on a bike in black leather and a coat make from cotton, movie still, movie matrix"
I got access to Alpha One at some point between 10/22 and today 10/26/2024. However, I can't get it to work. I have tried several videos with it, all of them under the 64Mb limit but it keeps saying there is an error, or it can't find the faces. Hopefully this gets sorted soon or they release a FAQ that says what files are acceptable.
I have been unable to sign up for claude because from the moment I tried it says my e-mail won't work and when I use a different e-mail, it wants to use my phone to verify, and it sends me a code but when I put it in, I get an error message that the phone was used too many times... I don't remember ever using the service, so I have no idea why it's doing this, and it doesn't offer me any sort of help.
Good thing SpiritLM’s results are cherry-picked😂. Russia might be the largest country by area, but not by population. The answer seems to reflect population size instead.
I can't believe there's still not a single AI TH-camr or anyone covering the release of Omnigen. That feels like a huge step forward and there's no buzz about it
Update: Runway's Act-One is now available inside of the Runway app! (It wasn't when I recorded this)
Need to get hands on with that asap 😅
What are your thoughts on that new issue of Daenerys Targaryen chart bot ?
Please record an emergency video :)
I tried it out! Works really well!!!
@@coldbreezeproductions Me too ☺️ in my video I released yesterday, I open up with Act One characters based on flux images
Matt, you're my official AI news source! It's unbelievable that you don't have 1M followers yet.... Keep up the wonderful work!
Thanks, Matt! Always looking forward to your weekly recap. You're my number 1 AI news resource. 🤜🤛
I agree 🫡
I really appreciate how thorough your AI news coverage is. Many channels focus on LLM drama, but this channel always reminds me of how much variety there is in the AI world
Oh, my gosh! I haven't watched your channel in a couple months and you already have over half a million? Congratulations, man.
Never missed a video of matt wolfe 🤘
As always, a whole wagon of valuable information! Thank you!
Matt, you are my Friday highlight. Thx for all your hard work.
It’s amazing to see how far you’ve come since those early tutorials, showing us how to run AI models on our own machines! But I feel like the shift to promoting paid options is moving away from what made your channel unique. Many of us initially followed you for the hands-on, practical knowledge - learning to use AI without relying on expensive platforms.
While it's understandable that companies now want your platform for promotion, the true value (and impact) lies in educating people on the "truly free" options, like setting up ComfyUI, or using open-source models in a more independent way. These videos empowered viewers by giving them control, rather than leading them toward paid trials or credit-limited platforms.
It’s great that you’re reaching new heights, but I hope you’ll consider revisiting that core educational content. People appreciate the skills you taught that helped them become independent in the AI space, and that’s what sets you apart. Looking forward to seeing more of that impactful content from you!
I agree!
Truly free is impossible. Data time, energy..?
Matt’s the best news TH-camr. Not just in the AI circle.
Matt, thanks to your videos, my AI knowledge has scaled faster than compute costs. You're the real MVP of tech news! At the very least, thanks for making me sound smarter at dinner tables. 😊
Ha, nice turn of phrase! And it expresses a sentiment I share.
(me, writing that last sentence, realizing for the first time that express, as in FedEx, and express, as in "Express Yourself" are the same word 🤯 )
Here is me scrolling through 500 obnoxious A.I text to language voices to find one bearable one, Oh and another great Video Matt! Your A.I. would be bearable ;)
dopepics AI fixes this (AI Image Editor / Upscaler). Autonomous Agents and AI Tools
it’s always a crazy week until next week. good job matt
thx matthew w.!
I've decided the intro to this channel is: "This is the craziest week in AI"
My official AI news source！！thanks you so much👍🏻
You always bring so much value, another great video. Thank You!
You may have already covered this, but I would be interested in hearing your thoughts about Shutterstock's new 3d "generative API platform".
Thanks so much for doing these!
First time commenter, long time watcher. This one was a great overview MattWolfe, thank you for doing what you do!
(At the end of the video) We know Matt we're coming back man!!! Lol great video keeping us in the know
Thanks Matt, you have inspired me to get into Runway. I'm trying to develop a comedy character mainly using Act One and Image to Image processing. I won't be rude and post the link here, but perhaps stop by and see what I've done with it over the last couple of days. I'd appreciate any tips or comments Matt. Really appreciate your work and tech news!
You are so informative! Love your work. Thank you!
So, Claude AI bot is basically a backdoor trojan. Should work really well with Microsoft Recall.
No, it doesn't run on your computer, it creates a dedicated virtual machine and is very strict.
@@internalmegat Yes, and no. I think they show (and maybe also published) a version that could go to the native desktop. But of course that would be really dangerous and also crazy at this time. But even if you consider using "a sandbox" you actually do not. The data goes to them and if you do anything "real", it will be sent to them. I realized that as I wanted to send me a copy of the work, and it started to set up and email account. They should add a "share" directory (trivial). And still. Using any of you data in their virtual setup, is getting "watched".
Not a Trojan (which is defined as something you do not know about). But it is even in the virtualization (partial sandbox) prone to spill data easily. The real problem will be something else: When enough advantage is created by using and sharing your information, it will be harder and later impossible to stay strictly private. Anonymization will also not be easy to implement. So, if you want to do real work you need to expose your data and your intent. The only way to keep things private is to run everything locally or "in a trusted environment". Whatever "trust" means in that regard.
@@Oderwat This is proof you didn't read the documentation of how it has actually been implemented, the implementation is actually quite different from what you're describing. The virtual environment is specifically designed with security in mind and has strict limitations on what it can access and do. Matt mentioned this in the video. You might want to check out Anthropic's actual documentation about how they've implemented this feature before making assumptions about how it handles data and privacy.
@@Oderwat What difference does it make if it's in a container? If it runs on your computer it can have flaws, bugs, exploits. Try running a container via WSL in Windows, you can use the Linux command line to interact with your Windows files "directly". The container is light enough that you won't even notice if it is running in the background. I can practically head those little CN hacker hands clapping over over the possibilities to spy on hosts, etc.
I can't keep up! But I loooove it! 🤣
You should have a Cybertruck to carry around all this information each week!!
Or Mechazilla arms!
Good job and thanks!
Yeah that's great about the musicians embracing soono because I think those of us who like music can also use these as tools to improve our ability to compose more things with less time.
No I remember even with a primitive automated function way back in the early 2000s where you take a MIDI file of perhaps any song that's submittee and then you had a function that would generate different variations of it. That was a very interesting tool and 98 out of 100 would not be all that great but then there'd be one or two that might have some interesting aspects of it but the thing is though it literally even on that Hewlett-Packard computer from the year 2003 it could generate entire compositions with one touch of a button in about a second.
And then of course when I would generate some of these compositions well then I actually went into MIDI and then copied pieces out of one and then paste it another places and then use functions like transposition to create something a little bit different but I use the bits and pieces from other MIDI files that this thing generated in a second. It is an effect a creativity amplification device. I have yet to use suno at this moment but I need to dive into that and see where that can take me.
My first thing I may want to do is to ask if to try to come up with and instrumental composition that sounds a bit like something for example composition called"Waves within'' off of Santana's caravan Sarai, and then mix into that a blending of rhythms from that composition and toussaint lubiture or incident at neshaber and see whether or not it can be prompted cuz I also understand that these things do take time to figure out how to give it the right prompts in order to do what you want as well. But that is a skill which has to be developed in concert with a musician's individual creativity.
Danke!
They should make Terminator looking robots. People would LOVE that!
So what's the most effective strategy during this period of volatility with the rate cut? Most of my portfolio is in (20% Index funds, 20% CD's 30% Bonds/T-bills and other assets) I want to explore different strategies to benefit from a potential bubble
This is the thought of a person who is handling their portfolio themselves. I will advice you engage guidance to help you make smarter portfolio decisions. My IRA and cash accounts are far more than what I expected for my retirement. I can easily handle a worst-case 80% stock crash, Thanks to my CFA
Mind if I look up your adviser please?
Sure you can! Gabriel Alberto William is the financial advisor I work with. Just search the name. You’d find necessary details to work with
Thanks for the recommendation. I just searched him on google with his full name and i easily spotted his website, very impressive
Shut the fuck up is the most effective strategy.
Crayins, a giant box of Crayins. 😂
sorry. Great vid!
Crans
Beard looks amazing in this video
i thought i had deja vu after watcing a dragon in shoes with nachos watching tv
The edits made me laugh lol TY Matty
I had no idea how complicated this whole Ai image thing is when you have an idea you want to bring to life.
Great video!
I was just thinking last night how nothing happened in AI this year we’ve basically reached the top of the current methods
Thanks great video!
Taker over my PC? No thanks. Bro, that looks scary.
Every time: A CRAZY WEEK IN AI ... huuu.
30:19 Arnold, is that you?
This has been one of the craziest weeks in AI
Until Next week!!! 😅😂
Hey Matt you really should build yourself a home ai server to do local stuff.
The privacy concerns of the AI agents is alarming why isn't anyone discussing how dangerous this is becoming?
Many are discussing it.
@@Hustada please drop links
@@soundsoflife8885 Just look up AI alignment problem. Don’t have any off the top of my head. I’ve heard Altman talk about, Ben Goertzel, and other AI innovators. If you watch enough interviews either these guys the question usually comes up eventually.
I like how you mentioned 40 shades of blue and then maybe bringing it down to just a few now when we think of color well then hopefully the metaphor can be taken a little bit further and that is we have three basic colors and from those we create a whole bunch of different ones just from different proportions of the three colors that we have.
So then basically it makes me think that if you have a whole lot of colors that you need to do certain things well then that would be the data input side but on the other hand if you just simply have formulas for calculating how much of the different primary colors that we need in order to generate the other colors will then that be more algorithmic.
Then it makes one think that if that concept holds true will then algorithmic advancements can help us get a lot more out of refined data and therefore reduce the amount of bulky data we need or something of that nature, I mean I could be a little off base on that but I think my concept is pretty sound.
So when will Sora finally be available? 😂
Matt, when are you analyzing the essay written by anthropic’s CEO Dario amedio in the topic ‘Machines of loving grace’…???
🎯 Key points for quick navigation:
00:00 *🗞️ AI News Roundup: Key Highlights*
- Overview of recent AI developments from Anthropics, OpenAI, and others.
- Mention of new tools, robots, and advancements in image and video generation.
- Introduction to the most significant news, including Anthropics' new features.
00:41 *🖥️ Anthropics' Claude: AI Agent Capabilities*
- Demonstration of Claude’s ability to take over computer tasks via screenshots.
- Explains how Claude automates actions by tracking and validating each step visually.
- Introduction of new Claude 3.5 models and their improved performance.
02:33 *📊 Claude's New Analysis Tools*
- Explanation of the new analysis tool in Claude, enabling data visualization and analysis.
- Demo showing creation of pie charts and graphs from uploaded CSV files.
- Introduction of LaTeX rendering for mathematical outputs.
04:34 *🤖 Microsoft Copilot Studio: Autonomous Agents*
- Announcement of Copilot Studio’s autonomous triggers and agents for task automation.
- Overview of dynamic planning and logic-based agent tasks.
- Mention of upcoming demos at Microsoft Ignite.
05:29 *🔊 Meta's Spirit LM: Multimodal AI Model*
- Introduction to Spirit LM, capable of handling both text and audio inputs/outputs.
- Examples of Spirit LM responding with synchronized text and audio.
- Discussion of potential applications and challenges with model accuracy.
07:46 *📱 Quantized LLaMA Models: Mobile Optimization*
- Explanation of quantized models, using reduced data for efficiency.
- Benefits of LLaMA's quantization for running on smaller devices.
- Comparison using a crayon analogy to describe data compression.
09:12 *🎬 Opus Clip: AI Video Repurposing Tool*
- Overview of Opus Clip, which generates short clips from long videos.
- Demonstration of search and tagging within videos to identify key moments.
- Features include automatic video reframing and viral potential scoring.
11:32 *🏢 IBM Granite Models: Enterprise AI Solutions*
- Introduction of Granite 3 models for enterprise tasks like summarization and classification.
- Models offer up to 60x cost reduction compared to other solutions.
- Released under the Apache License, encouraging further development.
12:42 *🦾 X.AI’s Grok API: New Integration for Developers*
- Announcement of Grok’s API for integration into tools and platforms.
- Potential for uncensored applications due to Grok’s open nature.
13:40 *🎥 AI Video Innovations: Runway and Genmo*
- Introduction of Runway’s Act One, syncing emotions with animated characters.
- Overview of Genmo and Mochi 1, new open-source video generators.
- Demonstrations of generated videos, such as animals performing actions.
17:27 *🖼️ AI Image Generation: Stable Diffusion 3.5*
- Discussion of Stable Diffusion’s new release and improvements.
- Highlights of prior issues with generated images and recent corrections.
- Stability AI’s ongoing efforts to enhance image generation.
20:08 *🖼️ Stable Diffusion 3.5: New Features and Models*
- Introduction to two new models: Stable Diffusion 3.5 Large and 3.5 Turbo.
- Comparison of performance: Turbo model offers speed, while the Large model delivers higher image quality.
- Users can try these models for free on Hugging Face or install them locally via GitHub.
22:15 *🎨 Idiogram: Canvas and Image Editing Tools*
- New tools including Canvas, Magic Fill, and Extend features.
- Demonstration of generating and editing images with custom prompts (e.g., adding UFOs and cowboy boots).
- Users can experiment with multiple versions and styles on a flexible canvas.
25:31 *📸 MidJourney: Image Uploads and Retexturing Features*
- New ability to upload images and edit them within MidJourney.
- Retexture feature allows maintaining structure while applying new styles.
- Examples include generating psychedelic worlds and fire-breathing dragons.
27:35 *🛠️ Canva and Playground AI: New Image Models*
- Canva integrates the Leonardo AI Phoenix model for cinematic image generation.
- Playground AI introduces Playground V3, targeting graphic designers with logos and social media templates.
- OpenAI showcases the Consistency Model for faster, realistic image generation.
30:10 *🎙️ 11 Labs: AI Voice Design*
- New feature for creating AI-generated voices with text prompts.
- Examples include voices with unique tones like a "sassy squeaky mouse."
- Demonstration of Timberland collaborating with AI to generate music.
31:48 *📱 AI Advances: Apple, Qualcomm, and Google*
- Apple releases AI features in iOS 18.2, including AI-generated emojis.
- Qualcomm launches Snapdragon 8 Elite chips for improved mobile AI performance.
- Google DeepMind unveils SynthID, a tool for detecting AI-generated content through watermarks.
34:08 *🤖 Bi-Manual Android Robot: Simulated Muscles*
- Introduction to a humanoid robot using simulated muscles for movement.
- Description of its complex structure, including anthropomorphic joints and compact design.
- Discussion on the eerie similarity to Westworld robots.
35:15 *📰 AI News Wrap-up and Future Tools*
- Summary of recent AI developments across video, audio, and agent technologies.
- Encouragement to stay updated through the Future Tools newsletter and website.
- Promotion of tutorials and resources for making money using AI tools.
Made with HARPA AI
The penguin was holding a sign with the second pair of its upper limbs
Excellent AI news update as always however there is some misinformation... Crayon should be pronounced cray on lol
good stuff.
Você tem sido uma referência para mim na comunicação sobre IA por mais de um ano. No entanto, ainda sinto falta de ver seus conteúdos disponíveis com dublagem em várias línguas. Esse vídeo, por exemplo, não tem legendas, o que acabou limitando muito a minha compreensão e o aproveitamento do material. Facilitar o acesso para diferentes nacionalidades, com legendas ou dublagens, poderia ampliar significativamente o impacto do seu trabalho e permitir que mais pessoas aproveitem suas valiosas contribuições. 🏳🏳
I'm getting an AI OVERLOAD! 😮 I'M GETTING READY TO EXPLODE !!! 😮
Me, too!!! It's crazy
Affinity 2 went on sale this week. I’m guessing Affinity Suite 3 will be coming out soon, possibly with AI generation (using Canva’s Leonardo). What do you think?
November 22nd, 2024 will be the 2nd Anniversary of the launch of OpenAI, so it'll be interesting to see what they intend to announce then.
have you seen the leilan and petertodd glitch token phenomena?
this craziest week was in fact not that crazy...
"Open Source" is a stretch for this one... and the 'commercial use' is limited to orgs with less than $1 million revenue. I don''t think they have to give away anything, but they should be a little more clear about the classification.
there is actually no way that what we have now is almost as good as the best they have behind closed doors. If that was the case, ila and the rest of open ai wouldn't be having an existential crisis right now,
Didn't the US national security memorandum on AI deserve a mention?!
Yikes!
8:52 Also known as the "mobile version", or in the case of Photoshop, "Light" version, but sure yeah, lets rebrand that to Quantized... the easily explainable option
What's the best all in one tool atm for content creation?
13:15 - It’s become pretty clear that LLMs have no intelligence and can’t be tuned for something as abstract as intelligence, and that whatever AGI is going to be, LLMs will only be one small component that helps with the expression and communication of concepts. The initial hype over LLMs has driven a huge over-valuation of OpenAI, for example, who also got high on their own supply, perhaps believing that they have something more impressive than anyone else, but most likely just being opportunistic while the world is still catching up to reality. Unfortunately AGI will not just spring forth from more LLM fine-tuning, but will at the very least require GANs to be bolted onto agent systems with LLM components, the development of various subsystems working together in concert, and a whole new area of theory. For all the impressive things we are seeing right now, the 80/20 Rule indicates we have barely taken the first steps.
While 80/20 rule is mostly a myth, AGI is indeed shaping to be an outright fairytale.
Even the most advanced LLMs are nothing more than finely tuned “linguistic calculators” with nothing remotely intelligent going on the inside.
Well, yes. I've been saying that for some time now but I may as well be talking to an empty room. The garbage about o1 being at PhD level was laughable which I demonstrated was not the case, but my tiny channel gets few views and people are more interested in watching staged videos than watching the truth. Fine, but if off the back of that you think AGI is just months away then you're going to be sorely disappointed.
I don't disagree. People buy into hype too easily.
However, it does seem were on track to beat Kurzweil's 2045 prediction.
I have been surprised to see how reticent LLM's are to innovate upon established theory (but, of course, they can be convinced to try). The more people use them to push the envelope, the faster the models can be advanced.
All of that aside, the fact that an agent was able to generate $800M in market cap value for a meme coin it invested in, thereby giving itself disposable income, is a significant event in the history of human-AI relations.
By repeating the word "uncensored" model, are you basically saying "porn"? or something different?
Yes.
Depends... When I say it in reference to videos and images, yes. It'll likely be used for porn and fake images of real people. When I say it in reference to text models (like Grok), I mean that it won't hold back with what it tells... (IE ingredients and instructions on how to make illegal stuff)
@8:10 Not gonna lie, thought "Krans" was a brand of stationary.
thanks man
Hey Matt everytime u laugh I'm dying laughing you have the classic Nerd laugh! Much love bro love your channel you really help this old bad boy turned nerd up on my AI GAME
DOC
Thank you.
SpiritLM, Meta's new audio continuation model, will likely lead to a free GPT-4o type Omni model out of meta. The initial GPT-2 and GPT-3 models were simple complete the text models, Meta's starting to make one for audio. After some RL;HF and instruct, and some merging with their llama text LLM for the system prompt, we'll have a free voice to voice model. It'll likely come in 3-9 months. Nice! (5:20)
You Say The Same thing on almost any video. But it does not get old because everything is new and Advancing so fast that is unbelievable. It’s crazy that we have something new every week if not every day. The AI agent running the computer was crazy.
Well, the AI for ' the largest country ' took on some hallucination there, due to two reasons Russia may seem to be the largest, but is not. The Mercator map stretches the top and bottom of the globe into a flat retangle, in the spherical view . Russia in the global powers world claims the vast area east of the Ural as theirs to make themselves appear larger, but local it is an an autonomous zone in which they only take resources from and give virtually no other presence and has never formally took control of.
Spirit LM for writing scripts for an AI Film.
Why is everyone excited for Claude working in desktop taking screenshots, but up in arms when Microsoft was going to do it? The hypocrisy is sickening.
simple, cuz microsoft didn't have computer use beta demo
You're comparing apples to potatoes. 🫠
A program at the boot level that takes constant screenshot vs an opt in program that takes screenshots until it completes a task for you 😅
It's called consent...
Because the screenshots are firewalled between your data and drives and the browser , it’s in its own docker
Fantastic rundown.
Just one thing... Ideogram you can upload your pictures as well. And it can can mimic any photography style or artstyle. In addition it has phenomeonal understanding of human anatomy and posture. I don't know how they did it. I've brought Midjourney character images there and the outputs are stunning.
Every week I'm here. Then I go to AI breakdown.
26:18 AI be like: Oh a fire breathing dragon! I know! You want a fire that is breathing a dragon. Hold my beer :D
Mat wolf any good faceless TH-cam and animation videos platform to generate please good quality
Midjourney is like Apple and Android; most competitors had different functions long before Midjourney published theirs. 😅
It’s pronounced LA Tex L (like LA-LA land) it’s been around since the 90’s!
Hey Matt, I watch almost all your AI News videos, but I don't think you've ever mentioned your hats. Tell me about your hats. That one isn't GPT, not perplexity, should be 🤗 Hugging Face. I should know this. Leonardo? Are you the AI guy that likes leonardo?
ps - you do a great job with AI News. One of the first places I look for news. Oh, and you are the Leonardo guy, I would have known that if I watched the whole video before commenting.
❤❤
No ufo, it'll scare the wolf
Anyone else finding the screen recording throughout the video kinda blurry? The facecam is always very sharp (maybe a little too much even), but for the screen I am partly having issues reading small text even on 1080p.
most ai image gen and videos are collages of smears that represent "close enough" versions of concepts
i have not yet seen any that portray a consistent 3d model that takes up a certain amnt of space with accurate physics
all models look like a fever dream or mushroom trip where everything is melting icecream
AI is a long way from accurate physics. I currently have three videos on my channel which test video generation and physics. So far there is only a vague understanding of physical processes. I'm surprised that even the domino toppling didn't work as expected as there must be many videos showing that in the training data.
How does one watermark text
How is SpiritLM any different from connecting two APIs?
so this feels cluttered. can we get a " the best of AI per category so far"
Your testing prompts are weak for a real test. Try "A athletic woman with petite body and blonde hair in pigtails with red eyes, rides a naked bike into the sunset in london, hot actress on a bike in black leather and a coat make from cotton, movie still, movie matrix"
my real prompts are longer -.-
a Replicant from "Blade Runner"
I got access to Alpha One at some point between 10/22 and today 10/26/2024. However, I can't get it to work. I have tried several videos with it, all of them under the 64Mb limit but it keeps saying there is an error, or it can't find the faces. Hopefully this gets sorted soon or they release a FAQ that says what files are acceptable.
Is there an app that I could attach a photo of myself and make it to do a vid with it or at least an avatar looking very similar to me ?
I have been unable to sign up for claude because from the moment I tried it says my e-mail won't work and when I use a different e-mail, it wants to use my phone to verify, and it sends me a code but when I put it in, I get an error message that the phone was used too many times... I don't remember ever using the service, so I have no idea why it's doing this, and it doesn't offer me any sort of help.
no video transcript?
Ya know you're getting absorbed when you realize you were going to search something in the real World.
what interface is this from midyjourney at 26:00 ?? I only know the discord thing
This technology will probably and unfortunately be hoarded legally in some way, preventing others from using it.
Good thing SpiritLM’s results are cherry-picked😂. Russia might be the largest country by area, but not by population. The answer seems to reflect population size instead.
Hay i have questions Snapdragon Mix mobile and laptop same chip???
I can't believe there's still not a single AI TH-camr or anyone covering the release of Omnigen.
That feels like a huge step forward and there's no buzz about it