The Frontier - SiliconNoon (Page 30)

The Frontier

Sparse weight decomposition — LLM circuits from under 1% of the data

Mechanistic interpretability has always had a cost problem: explaining a model meant training a second model. A new preprint from IQuest Research and collaborators at Oxford, Stanford, Tsinghua, and the Safe AI Forum argues the circuits are already sitting in the weights — you just have to pull them out. A research team led by IQuest Research (至知创新研究院), with collaborators from Oxford, Stanford, Tsinghua, and the Safe AI Forum, is proposing Sparse Weight Decomposition (SWD), a method that extrac

The Frontier

World Labs turns one robot task into thousands of sim worlds

Two research stories this morning: World Labs is betting that simulated worlds can replace expensive hardware time in robot training, and a new decomposition method promises to make model interpretability far cheaper. World Labs, the spatial-intelligence startup founded by Fei-Fei Li, says its Real-to-Sim-to-Real (R2S2R) engine can train robot policies entirely in simulation — with zero real-world training data — and then run them for hours on physical hardware. The engine, built on technology

The Frontier

PerceptionBench: no frontier model breaks 60% on basic vision

Frontier models can hold their own on math and code, but Moonshot AI's new benchmark suggests most of them still can't reliably see — and what gets blamed as a reasoning failure may actually start at the eyeball. Moonshot AI, the lab behind Kimi, released PerceptionBench, a benchmark that isolates visual perception from reasoning — and no frontier model clears 60 percent accuracy on it. The test breaks vision into ten atomic skill domains built from real-world errors: counting, localization, de

The Frontier

Fudan's Whitzard agent ranks No. 2 on CyberGym exploit benchmark

A Chinese university team just proved that an open-weight model plus strong agent scaffolding can out-hack the labs that built the models. Fudan University's Whitzard agent jumped to No. 2 globally — and No. 1 among universities — on CyberGym, the AI security benchmark, cracking 91.2% of the real-world vulnerabilities it faced. The new leaderboard, released this week, puts the team led by Professor Yang Min's lab in the world's first tier of AI attack-and-defense research. CyberGym, initiated b

The Frontier

Mimir v1: 1B model trained on permissible data matches 4B rivals

Two research stories this morning, one about what open models are trained on, one about what self-improving agents remember. Denmark's Foundation Models project says a 1-billion-parameter model built only on legally clean data goes toe-to-toe with models four times its size, and a new study quantifies how easily agents bake unsafe behavior into their persistent skills. Danish researchers say their 1B model, trained only on permissible data, matches rivals four times its size. Mimir v1 comes fro

The Frontier

Detecting benchmark contamination that survives RL post-training

Every leaderboard number carries a quiet question: did the model learn the material, or memorize the exam? A preprint posted to arXiv this week gives auditors a sharper way to tell the difference — one built for the exact regime where today's detectors go blind. A new protocol proposes detecting benchmark contamination from a model's internal activations rather than its outputs — and it is designed to keep working after RL post-training, the stage that erases the signals existing detectors rely

The Frontier

Verifier finds 39.5% of LLM-generated GPU kernels are broken

The case for letting AI write GPU kernels just met its sharpest audit yet: a new verifier argues that the field's standard correctness test is so loose it waves through broken code most of the time. A team of researchers built a contract-grade verifier for LLM-generated GPU kernels and found that 39.5 percent of machine-generated kernels a public system's own harness had accepted as correct are broken beyond any tolerance argument — and 62.1 percent carry at least one violation. The paper's pre

The Frontier

Study: frontier agents fail at open-ended AI research

The "AI does AI research" narrative took a hit from an independent evaluation today, Google quietly made visible AI watermarks optional, and the company pushed encrypted inference a step closer to production. A busy morning for the frontier — and for the people trying to verify its claims. A new "shadow evaluation" from Princeton and the UK AI Security Institute found that frontier agents can do the engineering of AI research but not the research itself — both papers the agent produced were rej

The Frontier

Qwen open-sources Qwen3.8-27B that runs on consumer GPUs

Alibaba's Qwen delivered the open-source model the community kept asking for, and Meta's most expensive hire walked out the door — a Friday that splits cleanly into two stories. Alibaba's Qwen open-sourced Qwen3.8-27B on Friday under the permissive Apache 2.0 license — a 27-billion-parameter dense multimodal model built to run on consumer GPUs, free for anyone to download, deploy, and sell. The company positions 27B as the size the global AI community demanded most, and the release leans hard i

The Frontier

Qwen open-sources 27B model that runs on home GPUs

Two open-weights signals landed within hours of each other: Alibaba pushed the most practical member of its Qwen3.8 family out to the open-source community, and Meta's most expensive hire of the AI talent war walked out the door. Qwen has open-sourced Qwen3.8-27B, a 27-billion-parameter dense multimodal model under Apache 2.0 that is small enough to run on consumer GPUs — free to download, deploy, and use commercially. The model, announced Friday, carries 262K tokens of native context (extendab

The Frontier

Deep Dive — Alibaba's 2.4T open model is a strategy, not a gift

On Wednesday morning Beijing time, Alibaba's Qwen team posted the weights of its most powerful model to Hugging Face on the exact schedule it had signaled two weeks earlier. The download is 4.45 terabytes spread across 213 files — and with it, for the first time, a Max-class frontier model is something people can run rather than rent. It is simultaneously the open camp's biggest win of 2026 and the clearest sign yet that "open" no longer means what it used to. The story looks simple from the he

The Frontier

AI 101 — What are open-weight models?

An open-weight model is an AI model whose trained parameters — the billions of numbers that encode everything it has learned — are released for anyone to download, run, modify, and build on. The weights are the model: once they are public, the technology itself is out in the world, not locked behind a company's servers. That single distinction divides AI into two camps — the closed, API-only frontier and the open ecosystem — and it is the most consequential licensing question in the field right

See all