Skewbed

@jon_barron @thomasahle My guess is that it doesn't. The author linked a post that shows an animation of an iterative optimization process for this problem, not a proof that this point set is optimal.

@thomasahle does this have a closed form? Feels like it should.

Voronoi-diagram for optimal Gaussian Quantization (Lloyd-Max-GQ)

@thomasahle Is this approximate, or is there actually a way to derive the global optimum?

@eersnington @theo NPM now stands for “Now Part of Microsoft”

@theo i'm starting to notice a pattern here

Tanner begged NPM to take down a squatted "tanstack" package that was being held ransom against him. 48 days later, it was compromised and shipped malware. There is no excuse. NPM needs to make significant changes.

@massiviola01 @jon_barron Oh got it, I didn’t realize it was just a path over the edges of the cube. I thought it was a full Hilbert curve.

@skewbed @jon_barron It is a composition of two monotonically increasing functions. One is the power transform that maps depth in [0,inf) to [0,1), and the other is the mapping from [0,1] to RGB space along the color path of the figure. You can also see the second as a mapping from [0,1] to [0,7].

New fancy depth colormaps now on PyPI? Hell yeah! With the help of my friend Claude, I (vibe) coded a tiny library that implements the depth-to-RGB mapping from Vision Banana 🍌 and its generalization. pip install hilbertmap

@bitcloud Life imitates art, I guess

@wahab_twts @ash_twtz Probably just because of Solana/crypto honestly

@ash_twtz So rust pays the highest?

Programming Languages and their Developer Salaries • 🦀 Rust — $130K-$230K • ⚙️ Assembly — $120K-$200K • 🐹 Go — $120K-$180K • 🔺 Scala — $120K-$180K • 🐍 Python — $100K-$180K • 🐦 Swift — $110K-$170K • 🅺 Kotlin — $110K-$170K • 🚀 C++ — $110K-$170K • ☕ Java — $100K-$160K • 🍎 Objective-C — $100K-$160K • 💎 Ruby — $100K-$160K • 🧮 Fortran — $100K-$160K • 🟣 C# — $95K-$150K • 🔧 C — $100K-$150K • 🐪 Perl — $100K-$150K • 🎵 Groovy — $95K-$150K • 🟨 JavaScript — $90K-$170K • 🎯 Dart — $90K-$140K • 🐘 PHP — $80K-$140K • 📊 Visual Basic — $80K-$110K • 🖥️ BASIC — $60K-$90K

@VictorWilsonDev @Yampeleg No, humans evolved to survive and reproduce. Prediction is just a useful tool to survive.

@Yampeleg a human being is just a next token predictor. the token is just of a broader scope (behavior)

You realize it's only next-token prediction? That that's ACTUALLY all it does, for real? How is any of this even real.

@Yampeleg Most models are post-trained with RL and are therefore reward maximizers.

@veorq Why is it an even number lol

I factored the number RSA1024-1 using my home-built QPU stack; alarming sign that RSA1024 will soon be broken. I'm choosing Full Disclosure, in the interest of transparency and Science advancement: gist.github.com/veorq/25bee6ef… Non-ZK proof that the correct RSA1024 was used: #RSA-1024" target="_blank" rel="nofollow noopener">en.wikipedia.org/w/index.php?ti… @yuvadm your move

@_ueaj Think of it like the original wandb. It’s also self-hosted, which is nice.

@skewbed Never heard of it

Can someone make an optimized wandb. Why is it so slow

@runsonai @iotcoi Could you send a link to the HF model?

@skewbed @iotcoi There’s a drafter on HF. But Ddtree must have been custom coded somehow. I tried making one but failed. So curious how he did it.

Qwen3.6-27B-FP8 + Dflash + DDTree, 256k context, 10 agents ~200 tokens/sec max decode 136t/s average on a single tiny GB10 GPU at 49W power

@compchemm @iotcoi Yes, that’s what I said. They do not have a speculator model for this model yet.

@skewbed @iotcoi That's 3.5 not 3.6 here: huggingface.co/z-lab/models

@American_Gladio @iotcoi Yeah, this would have a lower token acceptance rate.

@skewbed @iotcoi Probably using the Q3.5 model, which may affect drafting quality afaik

@thepavelbuchta @Dimillian I've always had issues with it progressively getting slower after opening the app, probably a memory leak somewhere.

@Dimillian Because it’s a pretty fucking good text editor

Why would you buy a text editor for $60b?

@chaumian If you give it the tools to do so, it could.

can an LLM kill itself

I guess we'll have to wait and see what name they use for TPU v8.

@cryptodavidw Prover time

I'm a zk guy, scare me with one word

@MainzOnX @botir33751732 What about Trillium and Ironwood? Why does this generation not have a name when the past generations did?

@botir33751732 They need names like that because you have to totally re write your kernels per generation. There are actually versioned numbers within there but a lot get lost. Ie Blackwell 100s got skipped and Blackwell 200 came out. Then grace blackwell 200 and grace Blackwell 300 etc

TPU v8 is 😎