Alan Luu

@rohanpaul_ai I mean, that's the way you'd get planet-level data, right? Through crowd-sourcing. And you offer them a compelling game to motivate them to scan the world.

143 mn pokémon go players thought they were catching Pikachu. They were actually training the largest robotics vision dataset on Earth — 30 bn images and the corresponding world mode. Now it’s guiding delivery robots with centimeter precision.

I think we as a society need to have a serious discussion around AI, data collection, and privacy. Right now, there are no clear ground rules about what data can be used for training, when generative AI should be labeled, how AI should replace vs. augment human work, etc. Part of this ambiguity is intentional. Some companies push boundaries, operating in gray areas, skirting user expectations, and moving faster than regulation or public understanding. At scale, this erodes trust and pushes us toward a low-trust digital society. Think about where this leads: artists afraid to use tools for fear their style will be extracted. Generated video blurring the line between truth and fabrication, capable of both faking atrocities and obscuring real ones. Workers hesitant to document their work because it might train their replacement. Technology isn’t inherently good or evil, but it is inherently empowering. And machine learning amplifies those who already have data, platforms, and distribution. That concentration of power is what people are reacting to. The fear isn’t really about the technology itself. It’s about exploitation. It's about what happens when more powerful systems are built in an environment without meaningful constraints, accountability, or shared benefit. Companies think they don't need to worry about tragedy of the commons, but that's only because they don't think collectively and far enough ahead. If people are now afraid of augmented reality, how will they ever allow humanoid robots into their home?

@ManyATrueNerd As someone who worked at one of the 3 big memory companies, it’s because a single fab takes a couple billion dollars and 5+ years of construction. Running these at less than full capacity means financial losses, so demand must be guaranteed. AI might be a bubble so not true here.

Ok, I know almost nothing about manufacturing, so please help me understand why collapsing stock and skyrocketing prices of RAM isn't immediately causing a huge number of companies to start making consumer RAM when there's demand for it and the price is high...?

@cyb3rshi3ld hi

F*CK IT sending more study guides and practice dumps to everyone who says 'hi' 🛡️

@bayeslord It depends on your definition of “laws”, but for lay people definitions, they don’t assume the laws are stationary in cosmology. On a technical level, they separate the laws that don’t change with the laws that do, and abstract it into laws with variable dynamics/fields.

this might sound retarded but what justifies the apparent assumption in physics that the physics of our universe is stationary?

@yacineMTB Are you talking about 3D reconstruction from 2D images at different view angles/distances? That should be NERF/Splatt, but you have to have good coverage over all viewing angles. Otherwise, there are methods using structural priors or unsupervised pretrained models.

I have 10 images of something. I need to fit a model of it with some arbitrary shapes. Does this exist and if it doesn't why doesn't it exist? It seems trivial

@ylecun Yann is making the point that looking at language alone can’t solve this problem. You have to have a world model to know that the distance would be 2xPi in flat space, but in space with positive curvature like the surface of a spheriod, distance will always be greater than 2xPi.

To people who claim that "thinking and reasoning require language", here is a problem: Imagine standing at the North Pole of the Earth. Walk in any direction, in a straight line, for 1 km. Now turn 90 degrees to the left. Walk for as long as it takes to pass your starting point. Have you walked: 1. More than 2xPi km 2. Exactly 2xPi km 3. Less than 2xPi km 4. I never came close to my starting point. Think about how you tried to answer this question and tell us whether it was based on language.

🤔As a former data practitioner from a non-CS background, it always puzzled me: Why were certain tools labeled big data tools for "serious work", while the #Python libraries I knew and loved #pandas, #NumPy were seen as "toy" — but not deemed for real production uses? [2/n]

📣 As of TODAY, Ponder is in Public Beta. 🚀 Now ANYONE can start running Python data workflows (pandas, NumPy) directly in their data warehouse in less than five minutes. Just pip install ponder. Learn more below! 🧵[1/N] ponder.io/ponder-in-publ… #python #pandas #datascience

✨ New major release for Modin 0.13.0! This release includes significant upgrades to Modin's documentation 📖,support for pandas 1.4 🐼 , new algebra and partitioning layer APIs 💠, performance improvements ⏳+ bugfixes 🐞 . (1/n)