Maxime Chevalier

NoiseCraft: a Browser-Based Visual Programming Language for Sound & Music pointersgonewild.com/2021/12/05/noi…

@Spaceillionare @cyberprince_rwo Not just surveillance... You're talking about a machine that can be remotely controlled. Literally an agent with access to the physical world, in every home and every office...

@cyberprince_rwo Yes, but I suspect the US will ban Chinese made robots. Too risky for surveillance.

Im so fucking bullish humanoid robots but there’s like 100 companies in China going full speed already. Feels like it’s gonna end up being a commodity like a car in the future?

@jsuarez The ladies love your RL work

Going through my DM backlog now, finding missed client inquiries for PufferAI. This is my excuse:

@tlhIngan_qoH @flapjackfrog @vxunderground There are even open LLMs that can process images and video. Can't be that hard to hook them up to a USB device and some relevant tool use.

@flapjackfrog @vxunderground you could add a routine for "lovense" toys and train a model natively

Here's the thing no one wants to tell you about AI: 1. It's the worlds largest Python script 2. It runs on Linux, Windows Defender slows it down 3. It uses lots of numbers (nobody knows why) 4. It requires a bunch of GPUs (for gaming) 5. No, you can't have sex with it

@TheHumanoidHub @PTrubey But at least we know from Ashok that Tesla is looking at world models and other relevant technologies. I don't think they are dead set on going in one specific direction because of what Elon said. At least I hope not.

@TheHumanoidHub @PTrubey Stuff Elon has said about AI is not reassuring, e.g. suggesting the same AI stack as FSD will be used for Optimus. The same chip sure, but it can't be the same model or training, that makes no sense. It's a completely different setup and environment.

Brett: Figure will deploy robots to many different customers, we'll probably announce a lot of these over the next 90 days. Molly: Are you worried about Optimus? Brett: In my mind, this is not a manufacturing problem, this is an intelligence problem.

@TheHumanoidHub @BillyZelsnack I'm invested in TSLA but I would say: there is so much hype in humanoid robots, it doesn't matter how expensive the factory is. If Figure can demonstrate useful robots, they can raise several billions for a factory instantly.

@BillyZelsnack Interesting positioning. Maybe longer-term Brett wants Helix to be more like Android - runs on robots from any OEM. Derisks the high-capex bets on a single hardware approach.

@tekwendell @OwenBrakes Flaws can potentially be introduced directly at the fab, after the chip's design is complete.

@OwenBrakes Imagine an adversary with an unlimited budget paid to introduce subtle flaws like this, too.

The RF world is insane. Researchers recovered AES-128 keys from a Bluetooth chip by listening to its own antenna from 10 meters away. Crypto-engine switching noise couples into the RF chain, rides the 2.4 GHz carrier, and leaks out as radio.

@skywind3000 You could write a blog post about these other tricks

为啥 Quake 的代码翻来覆去就再说这个平方根倒数呢？其他惊艳的东西也不少啊：比如纹理映射做透视矫正时四个点做一次除法，四个点内用线性插值，算到第四个点时之前除法刚好并行算完，比如 span 排序代替昂贵的 zbuffer，比如 PVS/BSP，比如最早的延迟渲染，最早的预测同步机制，学生时候学 quake 代码时这些东西给我的惊艳程度远远超过那个平方根倒数，这么多年却很少见人提过，翻来覆去都是这个平方根倒数被人转来转去。

@DominiqueCAPaul The problem is that there are too many prominent hype salesmen misusing the term.

@DominiqueCAPaul IMO there's a perfectly valid definition of AGI: an AI system that can do any task a human can do in a reasonable amount of time. The key part is that it has to be very general-purpose, and not get stuck in ways a human wouldn't. LLMs don't meet that bar.

AGI/ASI is such a worn-out term. Most companies using it are doing so for lack of creativity to describe what they're doing in plain English.

@NemGee @wholemars @Tesla That actually looks pretty nice

@wholemars @tesla can get permits today. Just need this

Robotaxi is ready to go driverless in San Francisco Flawless zero intervention ride. And very pleasant with my Netflix and Apple Music profiles syncing

Jensen is one the smartest and most far seeing folks the world. "If an AI scientist warns people that AI is going to permeate across radiology and radiologists are going to get wiped out, it might seem helpful but it's hurtful. If we convince everybody not to be radiologists and we now need radiologists, that actually is hurtful to society. "It is hurtful to convince all the young college graduates not to study software engineering because we are going to need more software engineers than ever. That's hurtful." "Scaring people with nonsensical things, which are not going to happen, that this is an existential threat, there's a 20% chance that is is existential, that's ridiculous. "That it's going to wipe out 50% of college level jobs. "That is it going to completely destroy democracy. "These kinds of comments are not helpful. They are made by...CEOS. And you become a CEO, maybe you adopt a God complex and somehow you know everything." Brutal. And right.

Passive dynamic stability. Intelligence in mechanism: no sensors, no feedback control.

@lemire The purpose of language is to be understood 👍

I refer to 64-bit Intel/AMD processors as being from the x64 family. It is commonly used by Microsoft. A sensible alternative (common under Linux) is amd64 since this architecture was developed by AMD. We also see AMD64. Some people just refer to it as x86_64. It suggests that it is a variant of the x86 family. In my view, all of them are proper names.

@unixpickle On a logarithmic scale no less

Taylor Swift net worth over time. Images v2

AMDとIntelが共同提唱、x86の行列演算を一気に16倍へ AMDとIntelが、x86向けの新たな行列演算拡張「ACE」を共同で公開した。 AVX10命令と同じく入力ベクトル2本で、1命令あたり1024回の乗算を吐き出す。 従来比16倍の演算密度を、レジスタ負荷を増やさずに引き出す設計だ。 ノートPCからデータセンターまで、同じ命令でAIが走る世界を狙う。 AMXがサーバー専用に閉じていた制約を、両社の合意で外しにかかった。 長らく競合してきた両社が、同じ仕様書に肩を並べて署名した。 この事実そのものが、x86が今どんな岐路に立っているかを示す。 joho-todai.com/amd-intel-join…

@lisyarus Rust wgpu is a bit verbose but pretty good. It's easy to generate the boilerplate with LLMs nowadays. Gemini was able to nail it in one shot.

#graphics folks, I'm thinking of modernizing my undergrad graphics course and move it from OpenGL. Vulkan is too hardcore for an introductory course, Metal & DX12 are not cross-platform enough. Wdyt, how deranged is it to use native WebGPU? (maybe even in rust via wgpu)

@RidgesPhoto @niccruzpatane Yeah it seems to go against the key Tesla principle of having fewer parts...

@niccruzpatane I don't understand the wheel sizes. For a fleet. You want low cost of ownership and staggered tire sizes mean you can't rotate them.

The Tesla Cybercab uses 21” wheels in the rear and 18” wheels up front. Tire sizes: • 215/60/18 Front • 225/60/21 Rear The interesting part is that Tesla uses an aluminum wheel in the rear and a steel wheel up front. The wheels then get a gold cover which has a rubber lip that extends past the rim, to closer to the edge of the tire. The wheel cover helps give the impression of a bigger wheel, it’s a pretty cool futuristic look.

@ID_AA_Carmack Probably lots of very specific heuristics in the compiler / CUDA libraries. They might have a kernel for exactly 512x512.

GPU library performance can be very notchy -- runtime of batched torch.linalg.solve_ex() went up by over 10x going from 511x511 matrices to 512x512.

@alexpotato Thanks Alex :)

@Love2Code this post was great! pointersgonewild.com/2022/01/19/the…