Niruhan Viswarupan @niruhan

if you value intelligence above all other human qualities, you’re gonna have a bad time

@thegautamkamath Congratulations 🎉

Bill Freeman gives us first a list of warm-up bitter lessons. He keeps the bigger ones for later in the talk. #cvpr2026

“Nature loves courage. You make the commitment and nature will respond to that commitment by removing impossible obstacles. Dream the impossible dream and the world will not grind you under, it will lift you up. This is the trick.”

One theorem every ML engineer should know: The Johnson–Lindenstrauss Lemma. It states that high-dimensional data can be projected into a much lower-dimensional space while approximately preserving pairwise distances. Why it matters: • Explains why random projections work • Enables scalable learning in high dimensions • Used in embeddings, compressed learning, and ANN search • Helps fight the curse of dimensionality The surprising part: You can reduce dimensions dramatically without destroying the geometry of the data. That’s why many ML systems can operate efficiently even with massive feature spaces. Modern representation learning is deeply connected to this idea: Good embeddings preserve structure while compressing information. In ML, compression is often not loss of intelligence — it’s removal of redundancy.

@JessSorrell inspiring! I was in the IT industry for 7 years after my undergrad and just started CS PhD this year at 32 years old 😅 hopefully something good will come out of this!

when in doubt accumulate a diverse experience replay buffer and reinit

When Copernicus proposed heliocentrism in 1543, it was actually less accurate than Ptolemy's geocentric model - a system refined over 1,400 years with epicycles precisely tuned to match observed planetary positions. It took another 70 years before Kepler, working from Tycho Brahe's unprecedentedly precise observations, replaced Copernicus’s circles with ellipses - finally making heliocentrism empirically superior. Terence Tao's point is that science needs a high temperature setting. If we only fund and follow what's most state of the art today, we kill the ideas that might need decades of work to surpass some overall plateau.

My dad is an economist. On the highway he will never change lanes no matter how much faster the other lane is moving, insisting that the drivers in front of him have already exploited any possible speed arbitrage opportunities.

Tim de Silva @timdesilva

8 months ago

Apparently, only 16% of people check Uber and Lyft. I'm very surprised! Every time I do, I'm shocked by the amount of price dispersion..

80 68 1K 2.4M 270

6yo: What’s the biggest number? Me: There are an infinite number of numbers so there’s no biggest one. 6yo: What’s the biggest number you know the name of? Me: A googolplex. 6yo: Which is bigger, a googolplex or a googoobazillion? Me: There’s not a number named a googoobazillion. 6yo: If there are an infinite number of numbers, there must be one named a googoobazillion. Checkmate.

first AI came for stackoverflow and i did not speak out due to their unpleasant moderators then AI came for quora and i did not speak out because i never use quora then AI came for Wikipedia and i did not speak out because i did not care then AI came for AI research and there was no one left to speak for me

Short explanation of the physics Nobel Prize 2025 In classical mechanics you can know where a particle is and its momentum at the same time. In Quantum mechanics you can't. All information is in the wavefunction. Even if a particle is trapped, part of the wavefunction.. 🧵1/7

Well, this time it's by Terence Tao himself: mathstodon.xyz/@tao/115306424…

this is beyond mindblowing for me. somebody built a 5 million param language model inside minecraft, trained it, equipped it with basic conversational ability. probably the best thing i have seen entire month.

Matrix multiplication is now a multi-trillion dollar business.

crazy that they called it context window when attention span was right there

Huge computer science result: A Tsinghua professor JUST discovered the fastest shortest path algorithm for graphs in 40yrs. This improves on Turing award winner Tarjan’s O(m + nlogn) with Dijkstra’s, something every Computer Science student learns in college.

Explaining Flow Matching in 4 minutes

I’ve noticed people who have good creative ideas usually know tons of things. Good researchers have read a lot of papers. Good musicians have listened to a lot of songs. Not only that, they generally have strong opinions about what’s good/bad out there. I wonder why? I strongly suspect a causal relationship that goes something like this: “Everything is a remix”, so having good raw ingredients in your head enables good output. In fact, in some domains I’d argue that getting the prereqs into your head is the hardest step! Much of a PhD is getting to the starting line: ingesting enough prior work to have a chance at original contributions. There’s also taste. Good taste comes from sampling tons of things and understanding what you like and don’t like. So there’s a volume of input needed to develop it. But it’s hard to say for sure! I could imagine other explanations contributing. Maybe people who naturally produce good ideas just happen to be the same kinds of people who devour lots of inputs? Maybe there’s just a general correlation between memory and intelligence or something? Of course, knowing stuff doesn’t automatically imply having good ideas. But it does seem to me that smart/talented people are often bottlenecked on knowledge and experience more than anything else…

quantization-aware doge