muragekibicho

Getting your foot in the door is always the first step. Sure, you can be as cracked as you want. You can be mediocre as well. None of that matters if you don't have a foundation to start from.

I think a lot about how Terrence Tao and the usual 30 (already well-funded) math guys will receive this cash while us early-career math students are left holding the bag.

It's akin somewhat to "donating 100 dollars to the node JS foundation, not the packages that appear when you run npm fund".

Forisek & Jancina is a deterministic (not probabilistic) primality test for small integers. It’s super fast and easy to code.

For 32-bit integers, one needs only a 512 byte lookup table and and a few adds and muls to test for primality

Anomalous curves are insecure elliptic curves where the trace of Frobenius equals one, ie the number of points on the curve equals the prime modulo the curve exists. They are constructed super fast by searching for primes where p=12s2+6s+1 is a prime as proven in an old paper from 1976 lol

Bernstein’s obscure factorization algorithm is described as “folklore” by the authors of the CADO-NFS RSA breaking library.

It was finally coded in 2020 and this provided the necessary 25% speedup needed to factorize the 795-bit RSA 240

We’re balls deep in the internals of CADO-NFS. Tale a gander

I must admit, I stopped keeping up with the new AI terms after J-space. Absolute peace of mind.

Selling 105,800$ worth of compute for 39*200=7800$ is similar to theft I've only encountered in the third world.

My boiler's copper water supply lines were vandalized for scrap. Their total value was 1540$ but the thief sold the copper for a grand total of 20$. The police and I went to the local scrap market, asked for anyone with a great deal on copper supply lines, and the perp was caught.

The article's subheadings suggest this is AI slop. Seeing a section titles "The Production Symptom" is a Claudism in itself.

I’ve spent the past 9 months reverse engineering CADO-NFS. I finally got the lattice sieve working correctly!

Julia uses 1-based indexing. It's competes with R and Matlab for the same set of users. Both R and Julia have their core functions written in C++. Absolutely nothing new.

From my experience, grad students use Julia when their PI thinks a new programming language will help differentiate their next NSF proposal among vast funding requests.

Perhaps they're targeting a super specific niche. Like somebody who sees WXET Ephemereal and yelps in triumph "that's my quant!".

Mechahitler strikes again

It's the crypto aspect that's silly. Tech is cool IMO. However, the crypto adjacency makes it all look scammy tbh.

I guess IPFS is an artifact from the crypto craze. It's pretty silly if you think of it. However, I must admit, I got my first ever developer grant from Filecoin while they were flush with ZIRP era money.

btw,I thought it was an Initial Coin Offering, not an actual fundraise.

Incredible counterexample, but oddly relatable. I'd probably have achieved techbro 'post-economic' status earlier if I focused on Android dev instead of the shiny (and new at that time) Xamarin for Windows phones.

It's all so bizarre. They don't want to pay but expect in-depth articles.

It's akin somewhat to the open-source crowd demanding features (and security updates) all without donating to the project.

2 is not enough. 3 verifies the Dean-Sanfilippo correspondence.

SOTA quantization aware training (QAT) techniques summary.

1. Bitnet weights are one of three digits (-1, 0, 1). This is the OG 1.58 bit network.

2. Sherry quantization constrains every 3 out of 4 Bitnet weights to non zero for SIMD optimization. Do the math and you'll see this is a 1.25 bit network.

3. Tequila quantization handles excess zeros due to gradient flow issues.

4. Fairy quantization is Bitnet plus an imaginary number {±i}. So it's a 2-Bit network, not a ternary network.

Taken verbatim "But the hardest lesson wasn't about database design. It was discovering that the real bottleneck wasn’t what we were observing and measuring. "

The Hungarian Assignment algorithm is used by MBAs and Operations Managers to optimally assign tasks to their employees.

It’s rooted in Optimal Transport theory and resembles Sinkhorn iterations in Python. Pretty cool IMO!

Super silly of me to ask but how does Bending Spoons make money in all of this? Is it the "fire all American devs and save costs with cheaper Italians" schtick or is there something else?

I don't see how Airtable makes money on its own.

exe dev is a VC-funded company reselling open source software.

Everyone's making cash from your open source code except you.

They even left the em-dashes. Absolute slop.

" The only problem with AMD is software support — slower kernels and less day-0 support on inference frameworks make serving frontier models on AMD a real engineering effort."

They just raised their series Z or something lol.

Honestly, I shut down my webiste and Medium, then moved to Substack because they solved the "get paid for my writing" problem.

In 2022, Medium was defacto blogging platform and getting into their paid program was close to impossible.

In the first 2 days of setting up Subtack I got a 5 dollar deposit from my first paying reader. I couldn't have done that on either Medium or my personal blog.

The article's awkward opening statement proves it wasn't written by AI.

I have been interested in machine-assisted ways to do and teach mathematics from as far back as 1999, when I started coding several applets in Java 1.0, both for my complex analysis and linear algebra courses, to visualize various mathematical objects I was interested in (such as honeycombs or Besicovitch sets).

For the curious, eigenvalues only exist for square matrices. Singular values are like generalized eigenvalues.

Singular values are like the fundamental frequencies of your matrix. You know how you can define any color with RGB? In a (pretty handwavy) way, singular values are like RGB color codes for us math guys.

Optimizers like Muon and Adam play around with weights' first, or second order singular values to train models.

I'm not trying to be reductive but the article's a lot of words for "We're vibecoding our app now and the glorious (almost almighty) Haskell compiler is too slow for the agent to iterate it's mistakes until it gets it right."

This is 2017 Biden vs Trump for people who know who Godbolt is