Pinned Tweet
never doubt the man who has
1) autism
2) 170+ IQ
3) eidetic memory
4) yrs of experience & knowledge about almost everything Math/Physics
5) no need for money beyond basic needs
6) passion for his work
7) an unyielding confidence in himself beyond what the common man understands
量子ゼロ retweeted
Each of those bullet points about an offside player interfering with an opponent apply to Fernandez. Affected Martinez + Lammens.
Laughable application of VAR.
Also, subjective call, so surely Oliver should have been sent to the screen? #MUFC
coming soon: MoThRA (Model Throttling Rank Adaptation)
two models that don't fit in RAM served side by side on one machine. expert weights page in from disk on demand and stay pinned while they're hot.
a 112 GB MoE runs in 34 GB of resident memory, next to a second one—no swap.
the results that OpenAI just published are a finite-time blowup construction for 3D Navier-Stokes. we've re-derived its leading structure from first principles: the exponents fall out of a rational linear solve, the viscosity identity closes to exact zero. github.com/tsotchke/eshkol/b…
Moonlab v1.2.1 is released. This version adds simulated quantum annealing, attributable seeded shots, Stim and Sinter interop, VQE quantum geometry, and an updated browser playground.
All-in-one quantum computing stack with C, Python, Rust, & JS bindings. github.com/tsotchke/moonlab
量子ゼロ retweeted
Eshkol v1.3.4-evolve: exact ∂ on every substrate
Eshkol v1.3.4-evolve has now been released.
Eshkol returns derivatives with zero floating-point error.
(define (f x) (+ (* 7 x x x) (* 3 x x) x))
(derivative-n f 1/3 1) => 16/3
(exact? ...) => #t
Not 5.333333333333333. 16/3.
JAX, PyTorch and Zygote can't do this.
量子ゼロ retweeted
Releasing the model weights and technical report of Kimi K3.
Kimi K3 is our most capable model: a 2.8T MoE model with native visual understanding and a 1M-token context window.
New model architecture: 2.5x the intelligence per unit of compute, not just more params.
Alongside Kimi K3, we're opening up more of the stack behind it — high-performance attention kernels, MoE communication library, and infrastructure for running agent environments at scale.
Model weights: huggingface.co/moonshotai/Ki…
Tech report: github.com/MoonshotAI/Kimi-K…
Tech blog: kimi.com/blog/kimi-k3
量子ゼロ retweeted
Don’t know if its going be more hilarious or sad to witness Bob Myers drop nuts and get Jaylen Brown and Bron while his former employer continues to be petrified to make moves…
Maybe both
量子ゼロ retweeted
Bob Myers doing this all in one summer btw should show you he was never washed, he was just handicapped by the hands-on owner in GS.
量子ゼロ retweeted
Lmaoooooooo Bob Myers did more this month than the Warriors Front Office has done in years
量子ゼロ retweeted
BREAKING: LeBron James is signing with the Philadelphia 76ers on a two-year, $8 million deal with a player option, Klutch Sports Group and Klutch CEO and Game Over show host Rich Paul tells ESPN.
RT @tsotchke: Eshkol v1.3-evolve: Where the Language Grew Up
---
What Eshkol is
Eshkol is a Scheme, the same small, elegant Lisp family ta…
This quoted post is unavailable.
Eshkol v1.3.1-evolve is out—differentiate to ANY order, get EXACT answers, run FOREVER. The compiled Scheme with ∂ as a compiler primitive. is now built for always-on programs. Long-running loops proven flat. Full R7RS conformance. Zero-GC arena memory. Native LLVM speed.
Replying to @ProofofMaro
patents just let your competitors steal from you faster


















