BitcoinHammer

Not a financial advisor

[Project Tachyon @TachyonZcash]

Together with @zodl_co, @ZcashFoundation, @ValarGroup and @ShieldedLabs, we're advocating for a network upgrade that would make ZEC's circulating supply auditable, providing additional reassurance that no counterfeiting occurred in the Orchard pool before this week's bugfix.

https://t.co/uPeinopdgf

[tsotchke @tsotchke]

Tsotchke Corporation's innovative learning systems will send the price per token of superintelligence to zero

[tsotchke @tsotchke]

download Eshkol and make your code think https://t.co/TrhFKoWTlt

[zooko🛡🦓🦓🦓 ⓩ @zooko]

https://t.co/v7BiOdzU9E

[chiefofautism @chiefofautism]

octra coded

[tsotchke @tsotchke]

tsotchke × octra incoming

[tsotchke @tsotchke]

all this week we're going to be handing out free demo accounts to our agentic laboratory assistant platform. throughout Tech Week you'll be able to try it out for yourself if you meet our representative at an event.

[tsotchke @tsotchke]

later this month we'll be opening up the official beta test in advance of the public debut of our platform. if you're interested in becoming one of our testers drop me a line.

[tsotchke @tsotchke]

and yes: you will be able to pay for your subscription with $TSOTCHKE for a substantial discount (we love our supporters). in fact, a portion of all transactions done on the site will be tied to transactions for our token, regardless of what kind of payment is done. coming tsoon.

[tsotchke @tsotchke]

your demo account provides an unrestricted 30 minute pass to Selene's terminal interface. here you can generate and execute code natively in your browser, try the adversarial reasoning system, pit agents against each other in complex recursive debates and anything else you dream!

[tsotchke @tsotchke]

Selene is a bench-grade laboratory assistant for multimodal workflows. she'll help plan your projects or experiments from the ground up producing everything from detailed diagrams to complete projects. she works with you across a shared workspace with infinite memory and context.

[tsotchke @tsotchke]

i'll be in NYC again this week for a couple of days if you're around and want to get together

[tsotchke @tsotchke]

im going to keep my DMs open for the week. drop me a line if you're going to Tech Week and feel like meeting up

[tsotchke @tsotchke]

Eshkol v1.2-scale: from compiler to deployable ML runtime

Eshkol v1.2-scale is the release where the project moves from a powerful mathematical compiler into something closer to a deployable machine-learning and runtime stack. the arc is simple. v1.0-foundation established the core: a native Scheme compiler, built-in automatic differentiation, deterministic arena memory, HoTT-inspired gradual typing, macros, modules, tensors, and LLVM code generation.

v1.1-accelerate added performance and execution breadth: XLA, SIMD, Metal/CUDA, parallel primitives, exact arithmetic, the consciousness engine, a production bytecode VM, and browser execution. v1.2-scale is about shipping. the headline feature is model persistence. Eshkol now has a .eshkol-model format with model-save and model-load, plus tensor and knowledge-base persistence. trained networks can be saved with metadata and reloaded elsewhere for inference, moving Eshkol closer to a real production workflow. the release also adds the practical data and integration pieces around that core: Python bindings via pybind11, a stable C ABI, NumPy zero-copy tensor interop, image I/O for PNG/JPEG/BMP workflows, CSV/DataFrame utilities, JSON Schema validation, terminal plotting, deterministic PRNG replay, lazy streams, regex capture groups, and reflection APIs. just as important, v1.2 hardens the system. compiler diagnostics now point to real file, line, and column locations with caret underlines.

the release includes security fixes for subprocess invocation, Python FFI input handling, path traversal, integer overflow, ReDoS, SQL injection, and URL validation. the new edge/security regression suite is wired into CI, including sanitizer coverage. under the hood, v1.2 also cleans up core runtime behavior: per-thread arenas for concurrent allocation, deep-recursion stack support, self-contained WASM output, R7RS-compliant stdlib shadowing, better AD closure handling, stronger bignum correctness, and more reliable module/symbol behavior. the result is a meaningful milestone: Eshkol is no longer only a language for exploring differentiable programming and mathematical compilation. with v1.2-scale, it can save models, load data, embed into Python or C systems, run through a completed VM, target native platforms and the browser, and catch the hard edge cases that appear when real programs start depending on it.

v1.0 built the foundation. v1.1 made it fast. v1.2 makes it shippable.

[tsotchke @tsotchke]

i have just released Eshkol v1.2.3-scale. this version adds many new capabilities to the language such as model and tensor saving/loading, full production byte code VM with self-differentiating neural computable transformers, per-thread arenas & much more.
https://t.co/Olsae4Mijl

[tsotchke @tsotchke]

start using Eshkol today. regular development will continue until we finally iterate towards v2.0-starlight at the end of the year.

[tsotchke @tsotchke]

for more information about how to get started check out our helpful examples. we cover everything from machine learning to differential physics, including web applications. find out for yourself why Eshkol makes all the difference.

https://t.co/lnghEcfCf4
https://t.co/TrhFKoWTlt

[tsotchke @tsotchke]

today we embark on a mission to replace the need for NVIDIA for training on consumer hardware. we endeavor to provide CUDA for Apple, and today we have pushed the code for you to begin to use. within the next couple months we will eclipse the demand for it https://t.co/iC7CjwVesh

[tsotchke @tsotchke]

"CUDA for Apple" coming next month. that's the vibe.