วีดีโอ

Andreesen is brilliant

Will we all be substituted by super intelligent ai?

Hey, regarding the stable diffusion training: is there any code available on how to have inhomogeneous GPU Types and assign different GPU types to different parts of the model?

they've been punching way above their weight. and all they needed was hundreds of billions of dollars in weapons from the u.s. taxpayer.

hello

I only clicked this vid to say: “LASERS”

Great content, thank you!

Where can I read the slides?

He’s a genius

Deep neural networks are deterministic. This is a common misconception and embarrassing to say in this context.

Great interview! Keep coming back to this talk since it released two weeks ago.

Very informative. 🙌🏼💎

bros speaking on chinese or english??

He says a lot. But a lot of bullshit as well. 5 min in, he has already started to pouring out SV bullshit, "led the creation of self driving car..." There is no self driving car, yet.. there are only attempts to create "self driving cars". To anyone who questions me if I know who he is, yeah, I know who he is and I have personally interacted with him.

"Giant Laser"

Deterministic vs Probabilistic computers. Cush!! in the drive train. Air tires made motor sports happen. Springs and shock absorbers gave us traction. If Humanity were wiped out after a bad election - do you think AI would rise to take its place? Would they be kings and diplomats? Would there be - an Architect?

Nice

Came here to hear about Dr Evil

hello, thanks for that video, is it possible to have the link for the code use in the webinar ?

Instacart presentation is fantastic! (51:33 to 1:07:37)

custom speed at 0.85 worked great to make this talk sound normal

th-cam.com/video/E-PIidaqCyU/w-d-xo.htmlsi=sDLdM1RoJHotkSnn

Great video, thank you for sharing, appreciate you

The guy is a tech investor. Of course he is going to say big tech won't do AI well.

*reads comments* *raises eyebrow* *places pinky to lips*

Cease and desist malicious use of AI, energy weapons, satellites: Axis of EVIL / MAGA / Terrorists. My family and I are not your property!

5 minutes in and i' m hearing the excuse of semiconductor not fast enough and I'm already reminded that there was a reason for the slow incremental roll-out of chip improvements. While I'm not a fan of nor promoting stock market - as I find fiat currency abhorrent enough, nevermind the flippant value exchange and gambling on infinite pyramid schemes. The world competition for new I.P. breakthroughs is pushing more intelligence into the hands of more people around the globe and more data and intelligence will come up with better solutions for the world's problems that the mentality of the old fixed donestic protection racquets that seemed to have no ending . Sanctions seemed ignorant and more like closing down intelligence

Mark just affirmed that there are idiot politicians and bureaucrats populating DC. Trump with Elon heading the new Department of Government Efficiency. Problem solved.

Looked at the thumbnail once and I started looking around for Minnie me.

cone head

He is just regurgitating what the experts have said and done. All i see is a business man riding the wave of AI.

7:36 well said

Russian's have used ten's of thousands of drones on Ukrainian tanks, APC's and other vehicles. Russia has also used long and medium range drones en masse, where Ukraine has not.

it means go yourself

He is only one who speaks super fast and super clearly and in a soothing voice. I saw others try to imitate him but end up being very irritating.

1:03:29 >>> RUNWAY presentation Anastassus Germanadis 1:19:00 >>> World Modelling Visual data richer than language "Towards Universal Simulation"

You know you know

Rewrite rhe cell's software

Awesome

This man can be the definition of superficial knowledge. The confidence with which he speaks about things he doesn't understand is amazing.

7:25 How is it possible to compute attentions separately block by block? Softmax (attention weight) is calculated based on all of the previous tokens and then those softmax scores are multiplied with all of the previous tokens' value vectors to calculate the attention score for the new token. So it should use all of the previous tokens on other blocks twice. What am I missing here?

reddit is cancer and should be closed

00:09 AI's future capabilities are inspiring 02:19 Evolution of computer industry towards neural networks 05:19 AI's evolution aided by compute power and data availability 06:54 AI will transform various industries significantly. 09:52 Transition from Discovery to Engineering in Biotech 11:24 AI shaping geopolitics and defense 14:32 Innovative warfare technologies impacting geopolitical dynamics 16:06 The future of warfare shifting towards technology and economic strength 19:02 Technology's increasing political influence 20:33 AI's impact on the movie industry and automation concerns 23:22 Importance of keeping technology open and democratic 24:45 Impact of Regulation on AI Innovation 27:27 The AI Revolution at higher levels vs. Robotics Revolution happening soon 28:56 Rapid advancement in hardware capabilities for human-like robots 31:37 Investing across diverse AI approaches 32:56 Value capture uncertainty in AI industry 35:40 Importance of deep domain expertise in starting tech companies 37:03 Successful startups often come from deep understanding of problems and innovative solutions. Crafted by Merlin AI.

Joe Spisak’s session at Ray Summit 2024 provides an in-depth look at Meta's AI roadmap, focusing on the Llama ecosystem’s transformative impact. He explores practical approaches to building scalable generative AI agents and emphasizes the latest Llama models, their real-world applications, and system-level safety considerations. Spisak's insights offer valuable guidance for developers aiming to harness full-stack AI capabilities, from foundational models to advanced applications, equipping attendees to drive future AI innovation.

5D AI designs for emotion and feeling with love upgraded to compassion wisdomly full of knowledges

If sequences of different sizes can be processed in parallel (say request 1 is generating 11th token and request 2 is generating 3rd token), how come those two operations (Query vector of request 1 - say dimension 1x50 - dot product with previous tokens' key vectors matrix 11x50) and (1x50 dot product 3x50) can be batched together?

Very superficial stuff. I am really disappointed in Andreesen whom I used to admire.

If Dr. Evil and Conehead had a baby...

I used Chat to talk to a French person who couldn't speak english on Discord. He had no idea I couldn't speak French.

TH-cam just added AI to it. I only want ONE AI not AI built into every app.