Brian Swift

@DonaldPMitchell Next step, have Claude keep rewriting the shader until DLSS Off quality matches On.

NVIDIA DLSS 5 is getting mocked, but I think this is the future direction for game realism. Basic rendering of the game animation, and then AI enhancement. Decades of SIGGRAPH research out the window? Maybe. This is a completely different path to photorealism.

@DonaldM38768041 Digital SAR processing is a bit more affordable in 2019 than 1983.

If you average a number of those terrain strips, you get an image with less noise. At the top is the published Soviet image from this pass. Below are images I generated (the bottom one is best use of Doppler frequency bands).

While writing a chapter on Venera-15, I decided to process the raw data to get a better understanding of the system. The Soviet radar orbiter emitted a continuous beam with a repeating pattern of modulation. A recording of the reflection was called a "radio hologram".

@DonaldM38768041 @SpaceX Interesting to watch the environment get significantly hotter as the 1st mach diamond rises into the camera's FOV.

@SpaceX High-speed Phantom camera placed near the pad. Water-cooled housing, nitrogen purging prevents internal fogging and protects electronics. Sapphire or quartz window. Enclosure features copper alloy with cooling passages for plume exposure. -- Grok-4

Liftoff of Super Heavy, the most powerful launch vehicle in history, on Starship’s tenth flight test

@ID_AA_Carmack @Jonathan_Blow Eventually we get a "new" OS, when maintaining Linux is comparable to maintaining a COBOL system, and someone gets an un-humanly good response from a code-gen AI to "Reimplement Linux in Rust..."

I had been meaning to comment on @Jonathan_Blow’s “Why can’t we even conceive of writing a new OS today” post. Coincidentally, I just got an email that opened with: ——————————————— Hey John, 2 years ago I pitched you LIBBA - a dedicated OS for smart glasses. You were sceptical, your main concern was that a custom OS rarely justifies itself: cost, shelf life, and developer burden outweigh the benefits. You were right. ——————————————— I deeply love the ideals of clear, efficient programs that do their job without baggage, and I have always been very sympathetic to efforts like Oberon, Plan 9, and even TempleOS. But building a new operating system today doesn’t make any product sense. Meta spent a lot of resources working on a fully custom XROS, over my rather strenuous objections. They had top tier engineering talent, tons of support, and they were producing high quality code and docs. It was a best case scenario from a “new OS” perspective, and, as one of the engineers put it, “If we can’t do it, who could?” I wish I could drop (so many of) my old internal posts publicly, since I don’t really have the incentive to relitigate the arguments today – they were carefully considered and prescient. They also got me reported to HR by the manager of the XROS effort for supposedly making his team members feel bad, but I expect many of them would acknowledge in hindsight that the Meta products would not be in a better place today if the new OS effort had been rammed into them. I can only really see a new general purpose OS arriving due to essentially sacrificing a highly successful product’s optimality to the goal of birthing the new OS, and I wouldn’t do that myself as a stakeholder. To make something really different, and not get drawn into the gravity well of existing solutions, you practically need an isolated monastic order of computer engineers. Which was sort of Plan 9…

@DonaldM38768041 @BarrakiAxonn @docmilanfar Commercial hyper-spectral imagers exist, but tend to be push-broom configuration.

@BarrakiAxonn @docmilanfar Yes, imagine a CCD with a mosaic of filters with a dozen different spectral windows. Maybe someone makes that for scientific applications?

The debate around whether every pixel in a photo from your phone's camera is "real" misses a fundamental fact about how digital cameras have always worked for the last 20 years. The camera sensor only captures ONE color (red, green, or blue) per pixel. The rest are made up 1/4

@NASAJPL Lots of amazing JunoCam images created by citizen scientists at missionjuno.swri.edu/junocam/proces…

Those swirls? They’re storms on Jupiter – and some are larger than Earth! This enhanced color view of Jupiter’s south pole was created by a citizen scientist using raw data from the Juno spacecraft's JunoCam.

@DonaldM38768041 Shhh. Don't want to burst any bubbles. Apparently (from the replay popularity graph) I'm not the only one who found OpenAI trainer's comments about current algo vs human data efficiency interesting. youtube.com/watch?v=6nJZop…

@b_swift They are hungry for training text and electric power. Contrast that with a human learning. To be honest, I think a lot is being invested in an algorithm that everyone will laugh at in the future.

If this is true, how could LLM's be expected to get anything right? But it suggests the question of what would you train an AI with to get "truth"?

@ID_AA_Carmack I'd wondered if FFTs could be applicable to neural net convolutions. Nice to see you and others giving it a try. Next step Fourier-domain optical convolutional neural networks.

It is a trope that TV technobabble usually involves “reversing the frequencies” or “modulating the frequencies”, so I feel like a federation science officer when I declare “I am running the convolutions in frequency space!”

@blghnkrkmz @iquilezles Hmm. So PS4 Dreams developers...

@iquilezles You have eliminated 95% of the potential candidates :)

@DonaldM38768041 Exactly

@b_swift I have worries about AI, but I don't think Skynet or Kurzweil-style transhuman event will happen. I think the danger is that human tyrants will use AI to surveil everyone, detect dissidents, and generate hypnotic irresistible propaganda.

I think there is huge potential in robots, but these kind of demos, Boston Dynamics, Chinese robot videos; I don't think these robots are yet doing anything useful. My test is can a robot strip a bed, wash and dry the sheets, and make the bed again. Let's call it the Linen Test.

ChatGPT naming successive Python script revisions the same dumb way I would

@DonaldM38768041 I remember that MythBusters episode. 😉

@DonaldM38768041 Skimming the 53 pages of free material at science.org/doi/10.1126/sc… I concluded red line wasn't from paper. Then checked 9/19/2024 WAPO article, not there either web.archive.org/web/2024092015…

(1) The Earth is usually more tropical than today (2) Linear regression doesn't mean connecting the first and last data points with a line.

San Jose Giants @ModestoNuts

@MCHammer Hmm. Didn't realize Boeing's market cap was that high.

Jensen Huang says Nvidia is missing out on a Boeing-sized opportunity in China: ‘Not a plane, the entire company’ Hamm400✨✍🏽 marketwatch.com/story/jensen-h…

@DonaldM38768041 I suspect the "what impresses you" goalposts have moved a bit from 5 years ago.

AI is getting there, but I wasn't impressed with Sora's attempt to render "HG Wells Martian tripods in Victorian London"

@DonaldM38768041 What are odds we are just electrochemical models, trained on a lot less sensory inputs, exploring pathways in a high dimensional vector space that nobody really understands. And self-awareness is just a pathfinding optimization. (I don't believe they think either.)

Hmmm. One minute grok amazes me. It will tell me how to solve an obscure UI problem with Photoshop CS4, and I wonder how can it do that? But when I ask probing questions about politics and issues, like "who is pulling the strings in this situation", it cannot answer. It rambled, it makes bullet items about what other people think. It doesn't look at influencewatch or opensecrets, it doesn't track down money trails and map out social networks and power relationship. It's a language model, trained on a lot of text, with pathways in a high dimensional vector space that nobody really understands. Does it think? I'm with Yann LaCun, I don't believe it thinks.

@abakcus Numbers all small enough to easily fit in the 60-bit registers.