So- I'm sourcing ideas for what 14,500 tokens per second in @claudeai could be used for. Distillation of Chinese models? Generation crazy huge data sets in seconds? I'm open to ideas!

We're upgrading the Cashu crypto to BLS blind signatures with public verifiability. This will make our lives so much easier. github.com/cashubtc/nuts/pul…

distributed some free sats on our experimental network last week client-side proofs, private transactions public access soon 🚀

We made a bit of history at @BTCPrague. The first bitcoin payments using zero-knowledge proofs were sent over Lightning. Discussed by Satoshi 2010, ZK can improve bitcoin privacy. Our experiment was the first real-world demo to showcase how:

60 fps paper-like screens have been achieved internally

While writing linux-insides, especially the early boot chapters, I was always annoyed by how non-trivial it is to debug early kernel code. So I decided to have some fun and build a tiny VM with a built-in debugger: something that can boot a kernel and let me inspect and visualize kernel internals nicely at every stage I would like. For now I "can" boot the kernel! Well, ok, it does not boot yet, rather fails beautifully 😀 But I can already see early debug output from linux kernel setup code inside my little vmm If you like or interested as I in Linux internals, bootloaders, kernels, virtualization, or want to help shape this into executable linux-insides chapters - follow along.

Why does it matter if sandboxes start up in 0.0001 nanoseconds if I then spend the next 20s downloading the git repo?

DwarfStar now supports SSD streaming in the DGX Spark and Strix Halo, not just in Metal. You can run the Q4 quants at decent speed, and even DeepSeek v4 PRO at low speed, or you can run Q2 Flash if you have less than 128GB.

We now support rich formatting for all chatbots. Tables, nested lists, inline media, formulas, headers and more — right in Telegram messages. 🔨 Start building! Docs: core.telegram.org/bots/api#r…

The new beta Claude Code model switcher

56,000 tokens/sec at just 80 MHz. 🤯 I burned a full Transformer with KV cache into a custom chip. Designed gate by gate as a 100% digital integrated circuit. Prototyped on a FPGA. (No GPU. No CPU) Just pure digital silicon running @karpathy microGPT, spelling out names on a tiny LCD. This is GateGPT 👇

Major life hack: DeepSeek in the Claude Code harness can also build and drive workflows, at a fraction of the cost and Opus 4.6/7 quality. I've got it running over 250 subagents in a workflow in adversarial reviews. Pennies on the dollar. Use my tool "Deep-Claude"

Not your models Not your mind

Updating the classics

explaining MIPs (Multi-Prover Interactive Proofs)

OK... my new KianV StealthV 5-stage pipelined SoC now runs Rust. Yes, Rust. 😂 16 KiB 2-way I$ 16 KiB 2-way D$ 2-way ITLB/DTLB 50 MHz RISC-V SoC 133 MHz SDRAM burst R/W Interrupt-driven SPI Ethernet UART TX/RX FIFOs FIFO-based SD card FIFO-based SPI Designed from scratch. Boots Linux. Runs Rust.

Build a Linux keyboard computer based on the Raspberry Pi CM0—the LILYGO T-ModDeck—and pair it with LILYGO T-Glass AR glasses featuring an HDMI interface.

We’ve built coro, a crate that lets developers build rollups using the commonware library in a day, allowing them to achieve: 20ms block times (a 10x speedup compared to alto with simplex) without the infrastructure overhead from expensive / bulky validator sets

A startup is Karpathy-style autoresearch, except the market is the loss function and revenue is the backprop signal. A startup is an agentic research loop where the world grades your hypotheses in dollars, retention, usage, referrals, and urgency.