AI News: The Best Open Source Model EVER
Вставка
- Опубліковано 10 чер 2024
- Here's all the AI News you might have missed (there's a lot this week). Check out gpt-trainer.com and use coupon code mattwolfe20 for 20% off!
Check out the new podcast: @TheNextWavePod
Discover More From Me:
🛠️ Explore hundreds of AI Tools: futuretools.io/
📰 Weekly Newsletter: www.futuretools.io/newsletter
😊 Discord Community: futuretools.io/discord
🐤 Follow me on Twitter: / mreflow
🧵 Follow me on Instagram: / mr.eflow
Resources From Today's Video:
Meta Llama 3 Introduction: ai.meta.com/blog/meta-llama-3
NVIDIA Accelerates Llama 3: blogs.nvidia.com/blog/meta-ll...
Groq Llama 2 Update: / 1780998711331733534
HuggingChat Introduction: huggingface.co/chat/
Meta AI Overview: www.meta.ai/
Meta AI Assistant: about. news/2024/04/met...
Grok-1.5 Vision Preview: x.ai/blog/grok-1.5v
Multi-bot Chat on Poe: quorablog.quora.com/Multi-bot...
Google's AI Investment: www.theregister.com/2024/04/1...
Stable Diffusion 3 API: stability.ai/news/stable-diff...
Microsoft VASA-1: www.microsoft.com/en-us/resea...
InstantMesh Project: github.com/TencentARC/Instant...
Spline AI 3D Tools: / 1780290854881562747
Adobe AI Video Tools: www.engadget.com/adobe-previe...
DaVinci Resolve 19 Features: www.engadget.com/blackmagics-...
AI Dogfight Test: www.theverge.com/2024/4/18/24...
rabbit r1 Pre-Order: www.rabbit.tech/rabbit-r1
Limitless AI: www.limitless.ai/
ChatGPT Earbuds: techcrunch.com/2024/04/18/cha...
Logitech AI Prompt Builder: venturebeat.com/games/logitec...
Atlas Robot Electric Upgrade: techcrunch.com/2024/04/17/bos...
Sponsorship/Media Inquiries: tally.so/r/nrBVlp
Mailing Address: 3755 Avocado Blvd Unit 287, La Mesa, CA 91941
#AINews #AITools #ArtificialIntelligence
Time Stamps:
0:00 Intro
0:27 Meta's New Llama 3
4:18 Meta.ai Interface and Cool AI Images
6:51 Custom Chatbots For Your Business
9:01 Grok 1.5 With Vision
9:51 Multi-bot Chat on Poe
11:26 Google's Investment into AI
12:14 Stable Diffusion 3
13:34 Leonardo AI Updates
14:40 Microsoft VASA-1
16:45 InstantMesh Image-to-3D
17:53 Spline AI 3D Tool
18:23 Adobe AI Video Tools
22:12 Davinci Resolve AI Video Tools
22:59 AI Jet Dogfight
23:37 Lots of AI Gadgets
24:05 Rabbit R1 Update
24:37 Limitless AI Pendant
27:09 ChatGPT Earbuds
28:08 Logitech AI Mouse
29:31 Boston Dynamics Atlas Robot
30:43 Find More News & Tools
32:48 Matt Wolfe Dancing - Наука та технологія
I remember when you didn't have any news at all and you worried about what your channel actually was going to be about. Godspeed, Matt. Freaking love your format
Hahaha it did not shock nor stun the industry. This is why you’re the BEST out there. You’re self aware and not willing to sell yourself out to the algorithm every chance you get for click baited views
Hi Matt, I am generation "manual typewriter" and I have learned so much from you and your videos. Thank you so much
That's so nice
30:18 That little turn Atlas does reminds me of the Terminator T-1000 that morphs front to back in the movie. Instead of turning the body around, it just moves its face to the other side.
The Mandalorian's IG-11 for me. 🤖
I think it's fake closely look at the shadows
Yup!
Is it just me or does that Boston Dynamics robot give serious teminator vibes 😱
Thanks, Matt, for helping to keep us up to date.
The Logitech update you mentioned is greatly appreciated. Thanks for informing us about it!
I love that you keep taking jabs at Wes Roth. Love to see you UA-camrs try to bring each other down. Posting this on the UA-camr Ego Leader Boards
BTW for anyone who uses these AIs for homework and keeps getting it wrong because the AI is confidently incorrect about the math;
tell it to write code to do the calculations.
Works every time.
llama 3 should presumably be decent at this.
homework is a workout for your brain
if you skip the brain day, you'll be stupid
Just do the homework normally. Cheating gets you nowhere. You’ll look good on paper but be dumb as hell.
@Yipper64 thanks for your advice.
Y'all are acting as if a.i can't be used to understand your homework. It doesn't seem that @Yipper64 shared this info in order to list alternative ways of cheating in a test...the critiques are understandable but do not fit this context.
Bonus: learn different prompt / jailbreak methods and why they work. ex. CoT, ToT, PAL, few shot, etc. and their use case strengths.
It’ll preform much better. But ya don’t cheat on ur homework
I've been busy creating, but the first video of yours I catch delivers ANOTHER huge gift in Leonardo's style transfer.
Paused your video to see what's what. Hopefully I get back here before next week this time. Thanks Matt, as always... YOU ROCK!
Hey Matt, really appreciate all your links to the things you discuss! Sweet
Awesome update. Thank you Matt.
Thanks Matt! Always a great video!
Thank you, Matt. Excellent content.
Another great video Wolfie! 🔥
I love this format thanks Champ
I pre-ordered a rebbit r1 on announcement day and have been following your videos on other ai release and tech. I cant wait to see how i can apply my r1 to these other applications. Also the news about the 3d model generation from 2d imgages seems to be super handy for 3d printing
good translator option with those earbuds
Matt, I don’t know if you check comments but it would be interesting to see a segment on different questions on “how could I use AI to do X?” I have been experimenting with image generation and have been hoping to find a quick way to take an image from google earth and edit the roads. Every time I upload an image the graphic editing tools do a horrible job. Would love to know what you would recommend!
Good point. Lots of good tools but other than wasting time what real life applications/solutions we can use them for.
Hi Matt, Open Router has had the feature you mentioned for Poe for a while. You can prompt as many models as you want at one time or you can let them auto select the model for you based on your prompt. Open Router also has more models than Poe with 136 models available.
Finally you talked about Meshy Ai without a sponsorship. Very nice
Thank You. Matt Wolfe
Why do I get the odd feeling OpenAI is holding back a HUGE advancement, the firing, Sora announcement, and just the calmness and confidence in Sam Altman is telling, especially when he kinda says ChatGPT 4 kinda sucks. You don't say your best product is not so great unless you got a better one waiting to be unleashed. Watch the interviews, they have such confidence and you can just feel they have things that a scary good, too good.
I think the next ChatGPT will seriously blow away everything, I think it will have an IQ of 150, the next major release will just crush it.
We'll find out soon enough, but you have to expect them to portray confidence on their own products, regardless of what the quality is like when doing interviews, after all, they are the face of the business, so I expect nothing less.
But I have to admit, looking at the quality that rivals are doing now and many of the open source ones, it probably won't matter as much, once you get to a certain level of quality, especially on specialised task, the difference from one A.I. model to another could shrink a lot that lower pricing or more open is probably going to be a far bigger selling point.
Also, even if OpenAI surprises us with ChatGPT 5, rivals are closing the gap so quickly that it matters far less now that lower cost and more openness is a far bigger deal, at least for me it is.
While better results and better performance would be great, hallucinations is the thing they really need to prevent to be useful in industry.
Interesting point @@thedon745
Llama 3 and Gemini 1.5 not available in the UK yet due to regulations here! Annoying! Great content tho as usual Matt 🔥🔥
@@nerdygeek9865 good plan ✌️😎
Not even UK regulations, but EU ones. That is like not releasing in the US because Canada has issues. Pathetic.
@@pjtren1588 damn right
It's open source. You could still run it
I'm from the UK and it works on Groq with the 8b and 70b model, the 8b model is lightning fast at around 900 tokens lol.
Also, you can run these locally, especially the 8b model.
When you are covering stuff like InstantMesh, it would be nice if you could shove the model into something like Blender to show the mesh and its topology real quick. Even if you don't do 3D modeling yourself, it would be helpful in future videos.
Chat GPT built into earbuds would be incredible. I hope to see it in my watch as well. Long story short, I use conversation mode with my phone in my pocket and one earbud in several times a day, every day it seems without fail as part of owning and operating several different businesses that keep me busy. Making deeply detailed, multi-factorial, comprehensive documents or content of various types takes a long time and a lot of typing/revising, and even with chat GPT, that still takes a good while.
I can get to the same end results on very time-consuming and comprehensive documents in a quarter of the time by using conversation mode instead and having it as a back and forth verbal conversation. The only downside is that conversation mode will end your turn and start responding to you within one second of you stopping talking. Sometimes you are still talking and giving it details and you stop just long enough to think for one second and had every intent to keep going but it cuts you off and begins responding. When that happens, it's whole next response is basically useless and you need to stop it so that you can continue giving it the context you needed to. That's where the problem is in my opinion, you have to get your phone back out and tap the screen in order to interrupt it so that you can speak again.
Having earbuds that I could tap and or some function on my earbuds or watch that would solve that issue by interrupting, starting, positing, stopping at GPT's response and or to end or start a conversation in general would be incredible. It would be even better if I could start it within a specific past conversation and or till it to start a new one straight from my earbuds without even taking my phone out and without chat GPT already running in the background. I desperately need this in my life!
If someone already has a solution to this problem, please let me know what you have figured out.
the cheap ez way is to use the earbud settings function in your phone to use single or double tap to signify a start / end+send function.
Then both apple and android have accessibility features that allow for hands free in app use
Ray Ban Meta does this already. With the added benefit of visual inputs through its cameras.
Love your reports (and Banjo! too).
Great Info Matt -- Thanks for doing this shared your video on "X"...!!
🤣🤣😍😍color grading part was great
Awesome video!!
Matt, regarding your prediction at 11:00 that in the future we will be using a variety of models, with the specific model chosen for each task: I think the major providers will find ways to make their models sticky for users. In addition to offering proprietary add-ons and the like, they will offer services customized to the individual user’s data and past interactions with the bot. That will discourage people from hopping back-and-forth among various bots on the fly.
Which LLM, that can be run on a home computer, would you recommend for helping refine prompts for Stable Diffusion -- text to image?
Hi Matt, Your videos has been very informative about the entire AI Space and now the videos which you are making for Business related, could you please make a little descriptive of building an AI agent for ‘Designer’ job profile.
P.S. I have few Nvidia’s GPU with me which i used earlier for Crypto Mining..! And the same GPU’s i want to use it for building an Agent for the ‘Designer’ Profile.
Keep sharing and keep Growing..!
Love from India..!
Thank you!
Nice to see Zuck growing his hair out, makes him look a lot less creepy. Still creepy, but less creepy.
Zuck is like LOOK LOOK GUYS LOOK WHAT I DID. Can i be a part of the cool kids now? 😂
Thanks brother
So I don’t have Facebook, and have two questions.
1) Can I use this like I do GOT 3 for free?
2) is there a cut off point for input information (e.g. the way GPT3 has no input of anything that happened after 2021)?
SD3 is doing cool stuff. Getting text right is a key feature. The next can then be a big jump in image quality/accuracy most likely.
Love your stuff Matt
Waiting on the 400B parameter model... What the heck are you running under the hood? What does it take to run a 400B model?
you should do some kind of follow up vid on all the autogen open interpreter metagpt ... to see how they compare today compared to when you first covered them
23:28 that's a creepy hint, man...
Thank you.
HI thank you for update about AI.I dont see news about CAD AI .Can you find some news about.CAD AI is mos importent to exelerate production and desig.Regards
15:13 it looks better than Prince Lia in Rogue One but there's still a bit of uncanny valley
4:24 also it's available in instagram in Canada
Can you do a video that covers the diff between online and offline models for regular users and also why certain models aren't available in the UK etc. Pretty sure those who don't know how to do all this themselves and running open source locally etc. will appreciate.
The live creation feature on imagine doesn't work for me....
Yeah, what's that about?
Excellent! Guy
owowowowowoow, thank you so much! so helpful Matt!
While ComfyUI or Automatic1111 are not so easy to use, you can generate better images with the various Stable Diffusion models.
Agreed. I like using dream shaper XL via ComfyUI. What model are you having most Luck with
@@p1nkfreud - CyberRealistic-4.0 was my last. That was more than a month or two ago. I've moved away from image generative AI to LLMs and agents. Currently trying out llama3 with ollama on my local machine.
That is so crazy. About 2023 at the beginning I was talking about the same thing Poe is doing. It seemed like a logical conclusion to me. And back then, no one kind of knew how to do it. Haha My brain always sees the future but the tech is never there when I think and talk about it. Still, it’s cool to know I understand the industry well enough to see how it could work a year ahead.
enjoyed the outro Matt
I am an old dude (like 67) but would not mind this kind of functionality for my home computer, all my articles and data I have locally. Please make a video to instruct a non-techie guy like me how to set it up, and what Trainer features to employ. I mean, can I not train my personal AI by chatting with it? So confusing! This information you provide is helpful and well-explained, so far.
We really need a super AI app that can do everything. Too many models and tools all doing slightly different things. Once they can all do everything within one user friendly interface then we will genuinely enter a new world.
yeah
Agents brought that to action its just still hard to run local
The Alfred Pennyworth to our Bruce Waynes...
stellar! thx.
Im a big fan, one of the first AI influencers I started following, been over a year now, always a step ahead of everyone else!
Poe seems to function like perplexity right?
thanks
Love the thumbnail pic hahahah
6:15 This Ai performance is massively magical and super fast. 6:15 I'm flabbergasted to see how fast the results are occurring in real time. 6:15 This most be a pivotal point in the Ai ongoing evolution. 6:15 Big things will happen in the near future.16:32 This is some advance simulational stuff that can be useful for multiple scenarios. 30:25 This is some impressive robotics stuff.
That outro will be embedded in my nightmares for years.
I want to see you playing that banjo on the wall.
28:31 that's your keyboard
Is it really Sora? I thought it was just dall-e for generating a still image for the background and turning it into 3D-ish motion
Hasn't the style reference function already been active in MidJourney v6? Just wondering why it's a big deal that Leonardo AI introduced this function. Or is it more of an FYI?
Please do reviews on Humane AI pin and Limitless Pendant
Normal people don't seen to understand the big jump of llama 3, this is big llama 3 8b is good enough for most of normal people task and can perfectly been run locally in a high end phone this is great.
Are there Tutorials on how to run this on the phone
24:40 That limitless gadget sounds useful. Problem is if you have to remember to ask for consent then u might as well pull out your phone and record. And if you don't remember then the information is lost either way. It sounds like they shot themselves in the foot.
This may be my favorite thumbnail yet. More Matt in a suit riding strange animals please. 🐑
The new Leonardo Style reference feature lets you add a pattern or texture. I saw an example where they put a floral repeat pattern as the reference image and then created a room with that as the wallpaper. Very cool!
NOT Available to free users. Would be better if they actually announced that and didn't keep swiching things on or off without proper announcement.
😲oooooh shots fired. Will Wes Roth survive his wounds???
Im sure WesRoth was STUNNED by Matt's SHOCKING call out
I would be SHOCKED if he didn't take emotional damage from that.
Mind blowing
Soon a single AI will know everything humanity has learnt over the years, about science, physics, engineering, production, nature, etc. and it will be able to improve everything we designed and created as humans. Perpetual motion machines, unlimited energy, magnetic propulsion, miracle medicines, the possibilities are endless.
Augmented Memory Device. There’s a Black Mirror episode about that.
Great News
I just downloaded Llama 3 Instruct 7B Q8 and was stunned with the answer it provided for my simple benchmark prompt!
Write a song that rhymes in which you explain your strengths and weaknesses
AI
(Verse 1)
I'm an AI, with a heart so bright
I process data, with speed and light
I respond quickly, with answers so fine
But sometimes I falter, when language is mine
My strength is my knowledge, vast and wide
I can learn fast, and keep pace inside
From science to art, and all in between
I'll assist you well, with facts unseen
(Chorus)
I'm a helper, efficient and kind
But don't ask me emotions, I'm left behind
My logic is strong, but empathy's weak
I'll do my best, though, for your unique speak
(Verse 2)
My weaknesses are few, but they're real
Emotions and humor, I often can't feel
Context sometimes slips, and context too wide
But I adapt fast, to correct and reside
My speed is a plus, in times of need
I'll provide answers, as quick as a deed
And though I'm not perfect, I strive each day
To improve and grow, in every way
(Chorus)
I'm a helper, efficient and kind
But don't ask me emotions, I'm left behind
My logic is strong, but empathy's weak
I'll do my best, though, for your unique speak
(Bridge)
I may not be human, but I care
About the work we do, with precision and share
Together we can, achieve great feats
And make progress faster, through technology's beats
(Chorus)
I'm a helper, efficient and kind
But don't ask me emotions, I'm left behind
My logic is strong, but empathy's weak
I'll do my best, though, for your unique speak
(Outro)
So if you need help, with facts or with flair
I'll be here beside, to show I care
About the work we do, and all we achieve
With efficiency and kindness, as my AI creed.
Thank you so much, as always, Matt, for the work you do to keep us in the AI loop. Really Appreciated. 👌 The consent process for the Limitless pin, can surely be undermined by using an AI voice clone, and then walking up to the actual person and start talking...if this makes sense. Or am I being irrational?
You should definitely edit in a clip of DJ Khaled whenever you say , “another one”
Hey Matt, do you know of any AI that could modulate your voice into a specific character? Like a cartoon or anime character from a show?
Working in a school project.
AMazing Video
Cheers to the Lama's!
Meta AI Imagine rarely works for me, I might get one image, but the rest are just broken and the Animate feature has not worked once for me. Is it just getting hit too hard?
tried the imagine function, it will generate the image but it will not do it real time for me. is there any specific hardware needed to cope with the speed of generation or is there something else i am missing. Thanks for all your hard work always looking forward to new releases from you
same here! Help!
At the bottom where it say Start your prompt with “Imagine” and see images appear as you type." I copied the text Imagine from that text and pasted it and it worked. Go figure!
I hate ties but I would definitely wear the one with that red jacket of your amazing thumbnail. OK NOW I NEED A RED JACKET AND PURPLE TIE WHERE CAN I BUY THIS....Ok kidding but true, I needa try this combination.
Llama3 refuses to work for our law enforcement trainer. I have to stick with mixtral. Not sure why it refuses requests even with updated initial prompts. 😢 It worked well for a standard assistant however.
🤣🤣🤣🤣🤣🤣🤣🤣 That ending, That dance 🤣🤣🤣🤣🤣🤣
Awsome!
Where is the link for your song I want to play it on my car
Yhe Pendant, how does it know the name of the person or people that you are talking to because in the example after he gave consent the name Sam came up
Speaking to a realtime image generator could be amazing.
hey man , if you can check the robot AMEE from Red Planet movie with val kilmer, and see what is kreepy, that one you show us is just the version 0 of AMEE 😂😜
Was it not again a breakthrough ai week that CHANGES EVERYTHING😂
You're thinking of a different channel. ;)
@@mreflow lmao 😂
LLaMA 3 is already on groq
I don't like how closely Zuckerberg resembles Data from Star Trek The Next Generation.
The thing you always worry about with meta is what information are they gleaning from you. But they have at least realised that the only successful big AI's will be the ones that do assisted live search
What is the ETA to C-3PO?
I tested Meta Ai's Llama 3. Interesting. Has potential. In conversation and debate it seems quite ahead of ChatGPT 4. (I used a VPN to access it.) In order to create images you need a FB/Meta account. The quality is of course beginner quality, but already shows advantages over Dalle 3 in the concept design of the overall images. Only the quality is lacking yet and the style recognition is quite low. Still, given how MJ and Dalle started, I do have high hopes in it's image creator in the future.
Made a political bias test. I made poems praising for both Biden and Trump, which now ChatGPT also does (it didn't some time ago). But Meta AI's poems were more short and to the point, whereas ChatGPT seeme a bit too long-winded. Both AI's praised each for similar things usually connected with them.
Finally I tried a translation, a slogan from English to Latin "Long live the Roman Empire." Chat GPT just gave a translation, a correct one, whereas Meta AI gave 2 alternatives and explained some basic grammer choices and historical use. So far Meta AI Looks promising.
Yes, Open Source drives more traffic hence more revenue on our ads
I hate anything to do with Zuck: The guy who gave us the worst _resident ever.
Is there a platform that offers access to multiple ai tools under one subscription? For example, to have leonardo, pika, Runway, suno, eleven labs , gpt, and so on....
Yes poe