The Archive

MASS improves self-distillation for reasoning tasks by aggregating predictions from multiple masked views of solutions.

Many AI workloads need only a fraction of a GPU. Giving each one an entire device can leave capacity idle; sharing GPUs lets platform teams serve more workloads... Many AI workloads need only a fraction of a GPU. Giving each one an entire device can leave capacity idle; sharing GPUs lets platform teams serve more workloads with the infrastructure they already have. NVIDIA GPUs already support several approaches to sharing. Multi-instance GPU (MIG) splits a GPU into hardware-isolated instances that are transparent to the application… Source

Atmospheric chemical mechanisms are commonly distributed in the language of the Kinetic PreProcessor (KPP), which generates efficient Fortran or C code but does not fit into the differentiable programming frameworks in which machine-learning models are developed. We present KPPax, a kinetic preprocessor that generates differentiable, just-in-time compiled, and vectorisable chemistry code in JAX from mechanisms written in the KPP language. KPPax evaluates the rate law with sparse operations in log space, which keeps it differentiable, and integrates the stiff system with the implicit solvers a...

Matched same-class controls isolate audio event class sensitivity from waveform artifacts in audio-language model robustness testing.

MiRL applies error-targeted reinforcement learning to terminal agents solving long-horizon shell tasks by profiling and fixing specific misbehaviors.

webframe, a Go tool by Jeff Lindsay, executes web pages as programs in headless browsers and pipes console output to stdout.

CodeRabbit integrates GPT-6.1 Sol to automate security vulnerability detection and fix recommendations in complex codebases.

OpenAI claims AI models can help migrate legacy COBOL systems at financial institutions and government agencies.

Stratechery commentary on the decline of decentralized internet hosting and shift toward platform consolidation, arguing the future remains optimistic despite centralization trends.

Evaluates transformer generalization for Bengali coreference resolution, finding limited cross-lingual transfer in low-resource settings.

Studies solution-set structure of grokking on Latin square tasks, characterizing which minimum-norm endpoints enable generalization from partial observations.

Reason-or-Recite framework uses dual specialists and LLM judge to resolve tension between evidence-grounded reasoning and reference-matching in traffic anomaly explanation.

Causal ML framework for dynamic tolling policy balancing mobility, equity, emissions; domain application, not AI-specific.

Nvidia CEO demonstrates Microduck toy robots with voice-command capability; 20,000 units sold.

BF16 RL training instability from inference-training stack misalignment solved by aligning rounding locations in RoPE, SwiGLU, and softmax without full stack harmonization.

ZENO: neural operator framework for solving high-dimensional nonlinear filtering via Zakai equation SPDEs, combining Fokker-Planck dynamics with likelihood updates.

ARENA neural architecture predicts solar flares from variable active-region imagery using multiple-instance learning without flare-origin labels.

Study investigates identifiability of hybrid deep generative models combining physics constraints with neural components via meta-learning.

Neural modeling of 3D Navier-Stokes turbulence using 28TB+ time-resolved DNS to evaluate numerical schemes for multi-scale flows.

Tabular foundation models solve power grid state estimation via in-context prediction without updating weights on system data.

OpenAI details containment measures for frontier model training: offline environments, compute isolation, and red-teaming infrastructure.

Canva is diving even deeper into AI with the launch of an image generation model that competes directly with those from OpenAI, Microsoft, and Google. The model, called Canva Craft, uses a two-stage architecture that allows it to plan a design before generating an image, allowing for "greater precision, consistency and control," according to an announcement on Thursday. Canva Craft can generate images at up to 2K resolution and in more than 30 aspect ratios. In addition to generating social media posts and marketing content, it's also supposed to create realistic scenes, digital art, environm...

Introduces Direct Pass@k optimization for pruning candidate reasoning solutions during test-time inference, balancing solution quality and diversity.

CausalCoT enforces causal transparency in chain-of-thought reasoning by decomposing problems into DAGs of sub-questions, preventing post-hoc justifications.

The Nintendo Switch 2 has apparently been jailbroken, and a modder has already shown the console running a custom version of Mario Kart World featuring new playable characters like Cinnamoroll, the red Angry Bird, Peter Griffin, and Steve from Minecraft. i jailbroke the nintendo switch 2 so i modded cinnamoroll, peter griffin, hatsune miku, and more into Mario Kart World, a switch 2 exclusive now watch my friends and i get cooked by bots learn more and get the exploit https://t.co/ykjgyfrFRp pic.twitter.com/BOcFhejRz1 - pepsipu (@pepsipu) October 22, 2026 OpenAI's Codex helped make this "ns2j...

Twin-world credit assignment isolates policy decisions from random sampling noise in episodic reinforcement learning with deterministic components.

Google DeepMind releases global forest-type classification map at 10-meter resolution using satellite imagery and ML.

Using an AI agent to shop for you is a strange experience. Tech companies have marketed them as personal assistants - the democratizing of labor that ordinarily would only be available to the rich and powerful. In practice, at least for buying things, it's more like sending an anxious and apologetic teenager out to the store: They require an inordinate amount of instructions and specifications, run into technical problems regularly, and come back with solutions that are not exactly what you asked for. They did, technically speaking, complete a task, just not the specific one you asked them to...

Study on prompt sensitivity in tool-augmented LLM comparisons, proposing matched protocol to isolate reasoning, tool-calling, and readout steps.

A* search with domain-specific heuristics outperforms SAT/IP solvers on Flood-It puzzle optimization, offering generalizable algorithmic insights.