math_dandy

Some speak perfect conversational Hebrew in a standard Israeli accent but switch to an Ashenazi/Yiddish accent in biblical or liturgical contexts. Pretty cool actually.

I know very little about Figma, but if their presentation product can (one day) understand LaTeX math syntax and render it, then I’ll definitely learn it!!

This is certainly an active area of research. When I saw the article, I actually assumed that Google was announcing progress in AI4CO and was surprised that the article stuck to fairly orthodox techniques.

I’m very excited about Mojo as a Rust 2.0 with first-class Python compatibility and support. Who knows if it’ll take off, but I love the project’s ideas and respect its ambition.

Absolutely right. If you keep worrying about whether 1 is actually 1, you’ll never get any math done.

React 19 has full support for web components and it scores perfectly on the “does your framework play nicely with web components” test suite whose name I don’t recall.

PyTorch and JAX have very good — but narrowly scoped — compilation facilities, and this suffices for a lot of really compelling AI applications.

A strong general purpose compilation story for Python is unlikely to materialize, IMO.

Why isn’t this product infused with AI?

Many heuristics in cryptanalysis rely on heuristics widely believed about the distribution of prime numbers. If RH is proved, some of these heuristics would become theorems.

It would not, however, give practicing cryptanalysts new tools since belief in RH (and some even stronger conjectures!) are already baked in to the current toolbox.

It may take many breakthroughs to yield a proof of the Riemann Hypothesis. Progress on a notoriously hard problem through introduction of innovative new ideas can certainly qualify as a breakthrough, imo. Naturally, the importance of the breakthrough will be judged by what can be built upon it.

People need to distinguish between ChapGPT-the-model and ChatGPT-the-service. The latter has memory; the former does not (as far as we know).

We have no definition of reasoning that is sufficiently precise to be useful.

But we do have a bunch of benchmark tasks/datasets that test what we intuitively understand to be aspects of reasoning.

For AI models, "being able to reason" means "performing well on these benchmarks tasks/datasets".

Over time, we'll add more benchmarking tasks and datasets that ostensibly test aspects of "reasoning", and people will develop models that succeed on more and more of these simultaneously.

And these models will become more and more useful. And people will still argue over whether they are truly "reasoning".

Well, yeah. How else do you keep your gizzard topped up?

Contrastively, if the computer has no dedicated copilot button then it’s not at “AI PC”. Makes it simple to avoid the undesirable product lines.

Check out visx. It should really be called D4: Declarative Data Driven Documents. It swaps out d3-select (the DOM manipulation module of D3) in favor of React. That this is possible and works great is a testament to the great design of D3.

This exists, it’s called visx, and it’s great.

https://airbnb.io/visx

If robust, affordable storage technologies that can smooth out the volatility inherent to solar/wind energy become available for a variety of topographies and climate zones, then we’ll be in great shape to reduce GHG from power generation without adding nuclear. Whether this will happen in the next 5-10 years is not clear (at least to me).

For the amount Canada recently spent on the expanding the Trans-Mountain Pipeline ($30 billion) we could probably build a nice nuclear plant.

I'm absolutely not an authority in these matters, and any corrections are welcome and appreciated.)

From what I've read, "green" hydrogen made from electrolysis with energy from non-carbon-emitting sources extremely energy intensive, as is compressing it for transport. Transport is also expensive. To my knowledge, there aren't any pipelines in Western Canada that could transport it, so I guess it's trucks and train cars. But maybe there's potential there because hydrogen wouldn't need to be the backbone of the grid, it would just need to pick up the (substantial?) slack when renewables weren't producing.

I totally believe that this is true on average. But there are a few weeks in the January/February when we're sitting under a high pressure system with clear skies, no wind, and -30C temperatures. Power generation needs to be able to handle annual extremes, not just averages.

Canada doesn't invest in infrastructure until it's too late. And by then, it's too late.

In much of Western Canada, energy use is highest at night in deepest winter when there is no appreciable wind, meaning that the grid is completely reliant on gas-fired power. If we want to reduce gas-fired capacity, we need another source of on-demand power. Meaning either nuclear or a robust storage solution. There seems to be more enthusiasm for the latter, but there’s no buzz about any viable solutions on the horizon.

This notion of derivative of a number depends very weakly on the number itself. It only depends on the multiset of exponents in the number’s factorization into a product of prime powers. In this way it’s like the divisor counting function, the omega and Omega functions, the möbius function, etc. As such, its value distribution might be interesting. Something analytic number theorists might like to play around with.

Investigating and prosecuting anticompetitive behavior simply takes to long to investigate and prosecute. And by the time this happens, the fast-moving nature of the tech industry usually means the issue is not nearly as relevant as it was when the issue arose. Then the company strikes a mea-culpa deal with the appropriate governing body and makes changes that don't actually matter any more.

Since all of these combinatorial optimization problems are, in full generality, are as difficult as possible in the complexity theoretic sense, there is little prospect of AI/ML helping to solve these problems in general. (?)

But people hope that some subclass of "real-life" instances of one of these optimization problems has enough structure that AI/ML may be able to learn useful heuristics for it. (This is not my field. This is just my take-away from reading some AI/ML-for-CO papers.)

I’ve always found it kind of badass that physics students are just expected to pick up the math they need as they study the physics.

If this book has exercises at the end of each chapter, I’d assign it to my calculus classes.

I look forward to my annual eight day matzahpizza orgy.

Hopefully DeepMind will think twice before sending cease-and-desist orders to any Hawaiian AI robotics businesses with aloha in the name!

This means that if you have the multiple instances of the same web component on the page, you need to duplicate the whole style tag in each (right?), which seems unwholesome.