🇺🇦🇮🇱dmitriy samsonov

github.com/d0rc/n-interac… - A hackable @PyTorch script to visualize emergent structures in n-interacting-particle systems.

Introducing M²RNN: Non-Linear RNNs with Matrix-Valued States for Scalable Language Modeling We bring back non-linear recurrence to language modeling and show it's been held back by small state sizes, not by non-linearity itself. 📄 Paper: arxiv.org/abs/2603.14360 💻 Code: github.com/open-lm-engine… 🤗 Models: huggingface.co/collections/op…

@LizaRosen0000 Which proves, that they had their own Epstein-like shit. In other words - all political and religious leaders are the same, and should be eliminated.

Don’t look away! t.co/A7s0tYHcJC BREAKING: The Islamic regime in Iran has launched a mass execution campaign of innocent civilians, including women and minors, who protested against the regime. They are being executed on trumped-up charges. Some were inhumanely raped and tortured until they agreed to give false confessions on Iran’s state TV because they could no longer endure the pain and chose death over the most horrific torture possible, including sodomy, which often leads to death or suicide of many prisoners before they even receive a death sentence.

Mathematicians use conjectures to point to important, open problems. We collect nearly a thousand (currently 890) recent conjectures from the math literature for a new dataset, OpenConjecture. On a subset, GPT-5.4 finds candidate proofs, and formalizes several in Lean.

Announcing the general availability of #Newton 1.0, the open-source, extensible physics engine for robot learning. Key features: • Stable articulated mechanism simulation • Hydroelastic contact modeling • Deformable body simulation (cables, cloth, rubber) • Accelerated robot learning at scale Netwon is a Linux Foundation project developed by Disney Research, Walt Disney Imagineering, @GoogleDeepMind and @nvidia Learn more: bit.ly/4cMm39x #NVIDIAGTC

Introducing Unsloth Studio ✨ A new open-source web UI to train and run LLMs. • Run models locally on Mac, Windows, Linux • Train 500+ models 2x faster with 70% less VRAM • Supports GGUF, vision, audio, embedding models • Auto-create datasets from PDF, CSV, DOCX • Self-healing tool calling and code execution • Compare models side by side + export to GGUF GitHub: github.com/unslothai/unsl… Blog and Guide: unsloth.ai/docs/new/studio Available now on Hugging Face, NVIDIA, Docker and Colab.

The brain as it's never been seen before. Last year, scientists created the largest wiring diagram and functional map of a mammal brain to date. #BrainAwarenessWeek @dana_fdn

Congrats to the @Kimi_Moonshot team! This is awesome. Great to see this level of research coming from open-source frontier model labs. I liked the paper so much I built a Rust implementation of it ;) Full AttnRes + Block AttnRes with two-phase inference, built using Burn (tensor library and Deep Learning Framework, in Rust, by @Tracel_AI). Runs on CPU, CUDA, Metal, wgpu. Includes an interactive TUI that trains a model live and visualizes depth attention evolving from uniform to selective in real time. Repo link and more on what is implemented in the comments.

@QuanticASI the real question is why do we assume asking questions requires consciousness:)

why do we assume intelligence requires consciousness?

@getjonwithit the whole science / culture / "civilization", etc - from PornHub to QFT is a part of biological evolution

I think, in hindsight, we will come to view the development of AI as more akin to a Eukaryotic Revolution than an Industrial one.

3/4 Instead of using an external tool, the model executes the program directly via its transformer weights, producing an execution trace token by token and streaming results at more than 30k tokens/sec on a CPU. All computation is done autoregressively inside the transformer!

@Dr_Gingerballs @karpathy He is not, it’s just he’s not always aligned with the truth:)

Guys, it's time to have the conversation that @karpathy might be an idiot.

@Dylan_Morri @norma_core_dev what’s your take?:)

Robotics companies are just servo companies that do some assembly

@pmddomingos No, they will dream even wilder!:)

When AI has made mathematicians irrelevant, they'll continue to do math like chess players continue to play chess.

@attentionmech Yes, it’s to understand

Is there really a way to increase working memory for humans?

@rao2z 100%, I want students, but I have lab:)

If you, as a CS Prof, are wondering whether you need PhD students at all now that you have wangled a subscription to Claude Code, your lab probably had a pretty depressing vibe to begin with--and 'em students are likely better off with you hanging out with Claude.. #AIAphorisms

@IlirAliu_ Why spend time talking about software skill issues speakers have..? Why not just use solid foundational frameworks for both research and production… sounds like pure personal marketing

The gripper misses by 2mm. Latency spikes kill your control loop. The sim-to-real gap eats weeks of engineering time you don’t have. Your model works in simulation. It even looks great in the demo. Then you deploy it… and everything breaks. Not a model, I t’s an infrastructure problem. Jensen Huang calls it the “Three Computer” challenge: The fact that Physical AI requires three fundamentally different compute workloads (simulation, training, inference) to work together seamlessly. Today, almost nobody has figured out how to connect them cleanly. At @nvidia ‘s GTC, @LightwheelAI Founder and CEO Steve Xie @bgxc , Ph.D., Ph.D. is joining Bill Vass, Evan Helda, @LerrelPinto, and Jon Quick to dig into exactly this! If you’re building robots and tired of duct-taping three disconnected workflows together, this one’s worth 40 minutes. 📍 March 16 · 4:00 PM PDT 🔗 nvidia.com/gtc/session-ca…

@bycloudai size of the problem ~ number of the heads:)

how big of a problem is this? > When backproping through the LM head, about 95-99% of the logit-gradient norm lies in directions that get projected away seems like the current workaround is just to use scaling to brute force it

I think one of the conclusions we should draw from the tremendous success of LLMs is how much of human knowledge and society exists at very low levels of Kolmogorov complexity. We are entering an era where the minimal representation of a human cultural artifact... (1/12)

SO101 autocalibration 🪄✨ Software release is around the corner 👀

@NoaMagid I'm so tired of reading about that bitch... who the f. is that mamdani, and why he and his wife can't be eliminated once and for all

This is what Mamdani’s wife celebrated.