Totally amazing! One of the best NBA games I've seen in a long time.
Posts by Sam Kumar Sachdev
I'd buy the t-shirt.
The relationship, between China and Latin American countries, is also quite complicated. Yesterday, for instance, Brazil labeled BYD as a company that uses slave labor:
www.reuters.com/sustainabili...
Hypothetically, could Trump have dropped a nuclear weapon on Iran and be immune from criminal prosecution?
How do you distinguish between an actual change in opinion versus an inherent error that's always present in sampling?
4/In other words, instead of making many thousands of plastic objects and that each have their own distribution, there could be a set of invertible geometric objects that could be homomorphically rotated and twisted to different shapes and that would provide analytic solutions.
3/The advantage of this analytic approach would be that these geometric shapes most likely have differomorphisms that would allow you to find invertible geometric units that then could be applied homomorphically.
2/That is, there are geometric -- or hypergeometric -- shapes that provide analytical solutions to the probability distributions.
1/I know little about robotics; so this question may seem naive. But, I wonder if there are analytical equations for the distributions of correct probabilities of robotic grasping.
www.youtube.com/watch?v=EtK5...
How are you going to enforce this?
5/ But I do think if you don't understand why it works, you're going to have difficulty to work in all cases and may have trouble either generalizing from it or optimizing it.
But, maybe I'm wrong. Or maybe $1 billion will help find why it works. Or maybe it won't.
5/ Then, at 52:18 in the linked video, LeCun casually mentions that, although this new approach seems to work well (see the paper for examples), he doesn't understand why it works. No one, it seems, understands why it works.
Does this mean that this new approach won't always work? Not necessarily.
4/...abstraction is needed for machine learning. In other words, the predictions in a feed forward neural network aren't concerned with pixels or tokens but abstractions of physical reality (that can be embedded in vectors).
Great! Sounds like an idea that has a lot of promise, I thought.
3/ Briefly, the idea is that LLMs aren't feasible for the physical world because they're too granular. That is, higher level abstraction is needed. Consider, the equation f=ma isn't at the atomic level. Rather, it's concerned with abstractions such as mass and force. LeCun argues a similar...
2/ Curious, I read the paper. Thought it was fascinating but became lost in the mathematical details. Wanted a higher level abstraction of how it works. Then, I found this lecture by LeCun last year:
www.youtube.com/watch?v=yUmD...
1/Today, read that Yann LeCun raised $1 billion (US) for an AI that understands the physical world:
www.wired.com/story/yann-l...
The company, Advanced Machine Intelligence, is going to be based on LeCun's idea of joint embedding predictive architecture:
arxiv.org/pdf/2512.10942
The minister of oil in Qatar has warned that, if the war with Iran doesn't end soon (this is undefined), then the price per barrel of oil could rise above $150. That is, the realities of the war are dictating the price of oil, not what Trump says:
www.thisismoney.co.uk/money/market...
Interesting! Also, worth pointing out, the top 11 countries on the list are also characterized by extreme political polarisation.
This is an incredible statistic! Also, worth pointing out, there seem to be about 58k people indicted in '16. So, the percentage NOT getting indicted is: 57944/58000 = 99.90%. Then: 1 - 99.90 = .01%!
2/The solution is to either install/upgrade a spell checker that has a Unicode that recognizes Maori words or upgrade your present Unicode that MSFT spell checker uses.
en.wikipedia.org/wiki/Unicode
1/Interesting. If you don't know, a spell checker is reliant on Unicode. That is, what the computer "sees" is the Unicode equivalent. What MSFT's spellchecker understands, through its Unicode equivalent, is not what you see. That is, it could simply see something it doesn't recognize.
2/...a term coined in contrast to "nation" in the conventional linguistic or ethnic sense."
en.wikipedia.org/wiki/Swiss_p...
1/The question is, how many counterexamples can be found to Musk's statement and, also, does Musk know how to use Google? This is a description of Switzerland, which has existed in its present state for 175 years:
"The Swiss are not a single ethnic group, rather, Switzerland is a confederacy ...
This misses the point. The point: The Dow Jones index is close to 50k.
Do you have a link to a data set verifying this claim? (Btw, I do know that it's true I just want a link to a data set that Google may not link to.)
2/
This was why it was profitable. When Bezos bought it, it had an operating loss of about $40 mil. It now has one of $100 mil. That is, nothing has changed -- except that Bezos didn't buy the profitable part of the company:
www.insidehighered.com/quicktakes/2...
1/ (Haven't read the article.) A reminder: the Washington Post was sold to Bezos without Kaplan Higher Ed. That is, the latter made up half of the revenue and ALL of the profits of the Post, before Bezos. In other words, the Washington Post used to be a test prep company w/a newspaper.
McPherson College is a small college -- about 800 students -- in Kansas (a state in the US). On its Wikipedia page, under notable alumni, Duane Earl Pope is listed. He was, according to this page, a "convicted bank robber and murderer. Briefly on the FBI 10 Most Wanted List":
2025 is going to be Baltimore's safest, especially regarding homicides, in its history:
www.youtube.com/watch?v=XQs5...
If this is disprovable through data, would JD change his opinion and then search for the (true) causes of why housing is expensive?
(Yes, this is meant ironically.)