When I was young I learned this cool trick to tie a bowline one-handed. Can’t remember who taught it to me but the idea is if you are hanging from a rope by your left hand, you can actually (with a bit of practise that you hopefully do beforehand) tie a loop around yourself with a bowline using your other hand. Moot in my case because there’s no way I could hold my weight using my left hand even if my life depended on it, but it’s a fun trick.
I like the “everything is a distribution” approach a lot[1]. Looks great. I’m looking forward to actually having a chance to mess around with it.
[1] And feels philosophically like the unification in the underlying maths between discrete and continuous probability that you get when you apply measure theory
Does it still count as a Dirac delta when it’s a discrete distribution? (The distributions in TFA are not continuous - they are things like a roll of 1d6 etc)
Backing a question about definition up with the actual source of the definition isn’t “dumb gatekeeping”. You feel that there is a right way to do RAG and you’ve written a framework that does it that way. That’s great.
I’m not the one who’s saying that certain things are not RAG so I don’t really see how I’m gatekeeping. I was actually trying to be helpful.
Yes you are wrong. RAG means retrieval-augmented generation. If you’re generating something and that generation is augmented by some retrieval you’re doing RAG. The retrieval doesn’t need to be from a vector db or even based on semantic similarity.
Joking aside, (and I speak as someone who has lost 2 close members of my family to dementia), as soon as people start to experience any form of dementia, there is a sort of coping mechanism that kicks in where they stop doing “complex” things as much as they used to, because there is a huge amount of anxiety that is caused by losing short-term memory (you are constantly thrown into situations where you can’t make sense of what’s going on). So my mother-in-law for example, stopped cooking for herself long before some of the other symptoms really became apparent. She told everyone she was “bored of it”, and when cooking for a larger group she wolud constantly check and recheck the recipe. At the time it drove me crazy but then later when she was diagnosed with Alzheimer’s it all started to make sense.
Not really sure what you’re getting at here. Maths has a very good answer (Bayes’ Theorem[1]) about how, in an uncertain world, one should update our belief in a hypothesis in the light of updated information.
One of my favourite of these is the Professor of Mathematical Statistics from U. Waterloo who is a reasonably strong amateur chess player and was engaged by chess.com after Kramnik’s accusations about Hikaru and did a video with the chess.com “chief chess officer” (IM Danny Rensch) explaining how Kramnik’s theories about how unlikely various streak lengths were were completely wrong (based on the most basic beginner errors in misunderstanding probabilities of dependent vs independent events) and long streaks were highly likely to occur especially for players like Hikaru who play very large numbers of blitz and bullet matches.
Kramnik responded with “This guy doesn’t know what he’s talking about”. OK Vlad. You know more about stats than a professor of math stats at uw. Sure…
Kramnik’s behaviour throughout has been despicable. I’m pretty sure it’s grounded in the fact that he is bitter about how chess has changed and he has got older and more stuck in his ways and the modern generation of players (I’m thinking people like Danya, Jospem[1] and Lazavik[2]) play/played a brand of chess that is entirely based on deep engine analysis and is totally alien to guys like him with an old-school Russian chess background. So they must be cheating (as far as Kramnik is concerned). The way he hounded Daniel Naroditsky in particular really disgusted me and I say that as someone who used to be a fan of Kramnik’s chess.
This is why I posted the k-anonymity paper. A dataset is k-anonymous if using the dataset is is not possible to distinguish a given person in the dataset from k other people. Then we set a value of k we feel comfortable with and from that we can back into the areas which intuitively you would think would need to vary in size based on time of day and how many people travel through each area in the dataset.
It’s a lot of work but this is what you need to do to guarantee a given (non-zero) level of privacy. It can be done if people are serious about it.
Yes. This general problem is known as "k-anonymity". It's worth everyone who works with any sort of personally-identifiable data to read the original paper because the framework they identified is still really helpful for thinking about these issues. https://dataprivacylab.org/dataprivacy/projects/kanonymity/p...
Newton didn't use dx/dy. That's Leibniz' notation. Newton's notation for the derivative is just to pot a dot above the letter so ṙ would be Newton's symbol for speed (dr/dt) and two dots would be acceleration (d^2r/dt^2) in Leibniz' notation. Physicists still use Newton's notation but only for derivatives with respect to time these days.
The part of Newton's theory that was troublesome is his fluxions don't have the Archimedian property. It took until the 1960s before Newton's notion of fluxions became rigorously formalized with Non-standard analysis. https://en.wikipedia.org/wiki/Nonstandard_analysis
Yess indeed. Additionally, one of the marketing campaigns (“Get the facts”) seemed deliberately arranged to interfere with the SCO vs IBM and SCO vs RedHat legal cases where SCO tried to claim they owned Linux. For people who weren’t around reading “groklaw” on a daily basis at the time this is a bit of a rabbit hole, so be warned https://en.wikipedia.org/wiki/SCO%E2%80%93Linux_disputes
我不认为自我意识真的是他的事。
[1] https://www.entrepreneur.com/science-technology/read-the-con...