Maxim Raginsky

look at these tall bois

Thank you!

Winter Old Fashioned

2oz rye
.5oz Ancho Reyes liqueur
.5oz cinnamon simple syrup
angostura and chocolate bitters
garnish with orange twist and rosemary

so I tied an onion to my belt etc etc

Aristotelian Intelligence Perception is all you need.

That makes sense. I think some sort of a modern hylomorphism would be a better option, something like what I discussed here: realizable.substack.com/p/aristoteli...

What about Thagard? I think he makes a lot of sense here (disclaimer — I more or less find the whole “stochastic parrot” framing to be silly, but I’m not a functionalist either): www.cambridge.org/core/journal...

Strictly speaking, in their model on the sphere it's not self-attention alone because the orthogonal projection onto the tangent space to the sphere at the current point is essentially layer normalization. (In fact, this was their motivation for working on the sphere in the first place.)

PSA: "Geometry, Topology, and Control System Design," published by American Institute of Mathematical Sciences. It is edited by my colleague Ali Belabbas and is dedicated to the memory of Roger W. Brockett, a pioneer of modern geometric and nonlinear control. www.aimsciences.org/book/AM/volu...

Basically, this is a perfect description of the entire field of AI Safety: a tower of factional definitions upon which they have balanced a single paragraph of (really shoddy) argumentation.

The fun part is that it could be either the stochastic parrots paper or a generic aLiGnMeNt paper.

mmmm, yeah

Win Thanksgiving (and Post-Thanksgiving) with Double Stock A true double stock is a culinary showpiece, an investment of time and ingredients that is worthy of its own spotlight.

www.newyorker.com/culture/kitc...

Smoked turkey carcass and leftovers about to turn into delicious stock, per the recipe from @hels.bsky.social

strong floor, no ceiling

I will take a look.

Engineering as lifeworld Much ink has been spilled about the proverbial “physics envy” allegedly felt by the practitioners of various disciplines that are even the slightest bit quantitative.

yes realizable.substack.com/p/engineering

by Jonathan Chait

How could I.

There’s also the related issue of the trade-off between the complexity of the code and the complexity of encoder/decoder. That’s why comparing genome sizes across different species is meaningless on its own.

I mean, I do happen to know a thing or two about information theory and ML.

maxim.ece.illinois.edu/pubs/lazebni...

papers.nips.cc/paper_files/...

Many such cases.

bsky.app/profile/mrag...

My issue is with amateur takes on information theory showing up in AI discussions. Basically, you have to know a lot more than the definition of entropy and its relevance to data compression.

Right before I left Twitter, technical discussions about AI were already beginning to collapse into Ilya/Sutton/Karpathy/… exegesis and obligatory mentions of information theory, thermodynamics, cybernetics, and Kolmogorov complexity with all the sophistication of a LW post. It’s happening here now.

Although in this case we’re also talking about mad cash from industry.

bsky.app/profile/chri...

🎯

My favorite part is “please report all cases of threats or offers of bribes to you in your capacity as a reviewer.” People have completely lost their minds. These are just *papers* at a *conference*.

cranberry Boulevardier

post-Thanksgiving cocktail and mocktail mise en place