Michael Razum e/acc

.

[Michael Razum e/acc @MichaelRazum]

Trained each skill with ~50 episodes.
Early work — but it’s already promising.
My code: https://t.co/1Wo7absJrh

[Michael Razum e/acc @MichaelRazum]

Rubik’s Cubes… meet SmolVLA. 🧩🤖

Trained SmolVLA to solve a cube in ~15h on RTX 5090.

✅ First version — not perfect, but fun to watch
✅ Visual-based solver
✅ Learned actions: flip, rotate, turn, center

Thanks @huggingface for #smolVLA! #lerobot #smolvla

More below🧵

[Michael Razum e/acc @MichaelRazum]

Upper Side Turns
Still tricky. Works most of the time — but when it fails, it fails badly

[Michael Razum e/acc @MichaelRazum]

@zachxbt @JPEGd_69 To be clear, I have nothing to do with the hack!

[Michael Razum e/acc @MichaelRazum]

@zachxbt @JPEGd_69 Hey, we spoke. One of my contracts was hacked by the same Curve exploiter, but that makes me another victim. This is the on-chain connection mentioned. Can you please edit your tweet? I do not appreciate being blamed for something with no reason - I have a reputation to maintain.

[Dave Richeson @divbyzero]

Theorem. There exist irrational numbers x and y such that x^y is rational.

[hardmaru @hardmaru]

“Our models perform equivalently to backprop on ML benchmarks, while utilising only local and (mostly) Hebbian plasticity. Our method raises the potential that standard ML algorithms could in principle be directly implemented in neural circuitry.” 🧠

https://t.co/jiZ2I1l73F

[Geoffrey Hinton @geoffreyhinton]

Extrapolating the spectacular performance of GPT3 into the future suggests that the answer to life, the universe and everything is just 4.398 trillion parameters.

[Michael Razum e/acc @MichaelRazum]

@saeedamenfx Just wonderig where the time diff comes from. Is it data loading or the mid price calculation. Wondering a bit that vaex is so much faster

[Daniel Martín @dmartincy]

Rare photograph of a 10x engineer in action.

[eytan bakshy @eytan]

After many years in the making, we're open-sourcing Facebook's Adaptive Experimentation platform, Ax, and BoTorch, a library for Bayesian optimization research, built on PyTorch. https://t.co/1wpFnlqmPg

[Mike Tamir, PhD @MikeTamir]

The Visual Python Debugger for Jupyter Notebooks You’ve Always Wanted https://t.co/uaBgCeB5xb #AI #DeepLearning #MachineLearning #DataScience #ML

[Michael Razum e/acc @MichaelRazum]

@thalesians @saeedamenfx @eliquinox Interesting. Have you any other favorite time series liabries? Something like pyflux maybe

[Michael Razum e/acc @MichaelRazum]

@saeedamenfx @chanep How would you create an optimal trading strategy from simulations, is there some method for this? Reinforcement learning maybe

[Michael Razum e/acc @MichaelRazum]

@saeedamenfx I had issues with complex objects (maybe it was the logger...) Actually the library multiprocess was the solution for me.

[Michael Razum e/acc @MichaelRazum]

@saeedamenfx How big are your monitors ;)?

[Hacker News Bot @newsycombinator]

Why SQL is beating NoSQL, and what this means for the future of data https://t.co/MM0yMF0g0h

[Spiros Margaris @SpirosMargaris]

Essential #CheatSheets for #MachineLearning & #DeepLearning Engineers

https://t.co/KnXBGXUd4d #fintech @DeepLearn007 @Ahkailash1 #AI