niels .1

DM me, you're here for a reason!

Python, AI and E2EE
headless life

[niels .1 @TSC]

nvidia's robot world models are live on reactor now: you drag a robot's arms around in the browser and it generates what that robot would see next, in real time. rehearsing a robot before you burn hardware, finally

[reactor @reactorworld]

Introducing NVIDIA Cosmos-3-Nano on Reactor. Two new models, available now via the Reactor API.

• Sim-Bimanual predicts what a robot sees as you move its arms and grippers.
• Sim-Transfer turns simulations into photorealistic video.

Try them now: https://t.co/1UqpJXt2Py

[niels .1 @TSC]

@tuckner

[niels .1 @TSC]

Making a perp bot to lose money passively

[niels .1 @TSC]

nvidia's desk box now comes in 64GB, five grand from acer/asus/dell/hp/msi, shipping friday. chain two together and a real model runs at home on 128GB, no cloud, and the launcher sets your coding agent up for you

[NVIDIA RTX Spark @NVIDIARTXSpark]

DGX Spark is getting new systems, bigger clusters, and easier model setup.

DGX Spark 64GB systems start rolling out Oct. 23, with eight-system cluster support and a new Model Launcher coming later this month.

[niels .1 @TSC]

@jamesperkins @thallescomumh @dominikkoch @amureki @theo @flydotio @unkeydev HA redis is quorum, failover and someone's pager at 3am. who do you call when a replica decides it's the primary

[niels .1 @TSC]

@VibeCoderOfek @hedera reading their page, the wallet is the only checkpoint, nothing goes through until you sign the transfer yourself. whether the agent has to ask you again before it even builds one, they don't say, so i'd throw a tiny testnet amount at it and watch what it does

[niels .1 @TSC]

hedera wired their chain into one MCP endpoint: point Claude Code or Codex at it and the agent sends HBAR and mints tokens on testnet, but nothing settles until you sign from your own wallet. agents with wallets are coming, this one at least can't spend without me

[Hedera @hedera]

The Hedera Network MCP is live on testnet. Connect Claude Code, Codex, Cursor, ChatGPT & more with one endpoint, then ask your agent to send HBAR, mint tokens, or query the network. Nothing settles until you sign it in your own wallet.

Try the Hedera Network MCP: https://t.co/T9J8O1QtZG

[niels .1 @TSC]

@marcelotryle @pydantic putting it behind a proxy doesn't help here man, the stealer runs on every python start, even processes that never import it

[niels .1 @TSC]

@Texh_irien @GoogleResearch yeah that's the whole point, compile it once and it stops caring which chip it lands on

[niels .1 @TSC]

google open-sourced the chip design AND the compiler, so you build the model once and it runs on earbud-tier hardware instead of getting rewritten per chip. edge devs have been begging for this for years

[Google Research @GoogleResearch]

Building for wearables and IOT shouldn't mean rewriting your model for every chip.

Today we're completing the 32-bit open-source Coral NPU architecture (up to 1 TOPS under 10 mW), releasing our v1.0 compiler, and launching new Coralboard devkits with Synaptics: https://t.co/CIXUd5sNRu

[niels .1 @TSC]

@Rosasynthesiz @googlecloud @vllm_project @sgl_project @_llm_d_ @huggingface @raydistributed their blog says the compiled path is meant to land near JAX compile, and their fused eager mode is 50%+ faster than plain eager. the one line swap gets you running, not peak, shapes and sharding are still the tuning part

[niels .1 @TSC]

pytorch on google's TPUs, no jax rewrite: google and meta's pytorch team open sourced TorchTPU, you swap cuda for tpu and your model runs. vllm and sglang already run on it

[Google Cloud @googlecloud]

Native PyTorch on TPUs is here—and TorchTPU is officially open source.

Train and serve models using the tools you already know: @vllm_project, @sgl_project, @_llm_d_, TorchTitan, @huggingface, and @raydistributed.

Learn more from #PyTorchConf2026 → https://t.co/mISCD65vA4

[niels .1 @TSC]

@mhalphen @PalantirTech nah, no nvidia deal. they're just running nvidia's open models through palantir's platform, that's what ondS's own quote says

[niels .1 @TSC]

sovereign AI in practice: L3Harris fine-tunes open models on its own data and says it beat the frontier labs in days at 95% off. a small model on your own data doing one boring job well is the least sexy real thing in ai

[Palantir @PalantirTech]

THE SOVEREIGN AI REVOLUTION IS IN FULL FORCE.

We are helping our customers build superintelligence they own, with data they control.

McCarthy Building fine-tuned Nemotron in Foundry on labeled RFI data in the Ontology to identify project risks before they escalate, outperforming frontier models at significantly lower cost.

Tampa General is using sovereign AI to better protect patient data, keeping sensitive data and model processing local, and fine-tuning Nemotron models to classify LLM traffic against its own risk and privacy standards.

NVIDIA is using Palantir AIP and Nemotron to accelerate its own supply chain, shortening the path from wafer to deployment, spotting disruptions before they hit production, and keeping its models under its control.

L3Harris is bringing AI workflows in-house and building its own models trained on proprietary data to support critical defense use cases and the warfighter.

Ondas is launching Regna AI on Palantir AIP, leveraging open-weight models including NVIDIA Nemotron, to help customers adapt, evaluate, and deploy mission-specialized AI while retaining control of their data, models, and infrastructure.

[niels .1 @TSC]

@oneill_c @baselabs a week, #1 of 208, weights open. and the shared prefix cache so every extra question is one short pass, that's the bit i'm taking

[niels .1 @TSC]

@iJimmyChan @NousResearch @MeshyAI can't vouch for the desktop itself, but the plugin side is install and enable and paste a key in its settings. the gate is the key, meshy only gives those on a pro plan

[niels .1 @TSC]

nous wired meshy into their agent, so you ask in chat for a 3d model and it comes back rigged and ready for godot or a printer. photo of a mug to printable file is the bit that gets me

[Nous Research @NousResearch]

Hermes Agent now has an official plugin for @MeshyAI.

Ask Hermes to make a 70m poly treasure chest for Godot, or to turn a photo of a mug into a model you can print, and it does the job through Meshy's MCP server.

In Hermes Desktop the models render in chat, rigged characters animate inline, and a gallery stores every model and prompt. On messaging apps, Hermes sends a short preview video so you can create on the go.

https://t.co/QlxZUDEBn3

[niels .1 @TSC]

xaira's X-Scientist takes drug-discovery tasks off Codex with the same model, and its gene targets hit in the wet lab about as often as human curation, in 20 minutes vs 3+ hours. wet lab still judges, it just doesn't have to scout anymore

[Bo Wang @BoWang87]

Meet X-Scientist: our step toward a self-improving virtual drug hunter!

Today, we’re introducing Xaira’s agentic AI platform for drug discovery:

🧠 Beats Codex using the same Astra model by 7.4 percentage points on DrugDiscoveryBench.

🧬 Nominates targets ~10× faster than human curation, with a comparable wet-lab hit rate.

🔁 Conducts autonomous research, outperforming AutoResearch and AutoScientists in the PerturbQA benchmark.

Here’s what we built, what we tested, and where we’re going. 🧵

[niels .1 @TSC]

alibaba cloud just put its frontier qwen models on theta's decentralized gpu network, live today, same api key. asia's biggest cloud shipping through somebody else's distributed gpus is a wild thing to watch

[Theta Network @Theta_Network]

Made it official, in person 🤝

Asia Pacific’s largest cloud provider is bringing @Alibaba_Qwen, the world’s most downloaded open model family, to Theta EdgeCloud.

[niels .1 @TSC]

@ryanlanciaux @claudeai nah it doesn't touch your machine. duet gets its own browser on a cloud box, same as grok bot's cloud pc. you just can't watch its screen yet