The Archive

GeoPair: training-free Transformer compression via cross-layer factorization preserving activation geometry.

Solver-level warmstarting to accelerate neural network verification by reusing solutions across specification variants.

CausalLoss-Fin decomposes financial agent failures into decision vs. infrastructure faults using causal intervention, addressing attribution bias in payment exception handling.

Amazon blocks Muse acquisition; analysis argues Amazon's physical infrastructure creates AI competitive advantage.

Study identifies domain gap between simulated multi-speaker extraction training and real conversational speech with imbalanced speaker presence and enrollment mismatch.

Bridge of Ψ's applies Schrödinger bridge generative models to learn quantum circuit optimization directly from examples rather than rule-based rewriting.

Destination Support Restoration repairs underrepresented multimodal trajectory predictions for robot planning via causal post-selection without retraining.

ClusterFewshot improves LLM few-shot prompt optimization by combining semantic task structure with utility scoring, outperforming random and metric-based selection in DSPy.

Text-to-SQL conformal abstention certificates misreport risk due to lenient single-database oracles; Spider-Realistic multi-instance swap reveals 2.73–10.23 point gaps.

Informed Masking prioritizes high-leverage token positions in diffusion LLM RL by identifying upstream/downstream structure rather than uniform random masking.

Counterfactual Tucker Diffusion recovers joint conditional distributions in high-dimensional panel data for causal inference via masked Tucker score learning.

Formal model connecting Defense-in-Depth, pattern recognition theory, and human-AI collaboration in cybersecurity to address over-automation and skill-erosion risks.

Activation-aware mixed-precision LLM quantization with cross-layer refinement proves scalar sensitivity proxies incur multiplicative Hessian distortion up to 10^13.

Xiaomi releases MiMo-V2.6-Pro 1T-A42B open-weights model trained for $3M, claiming top performance among open models.

OpenAI publishes framework for third-party safety assessments of frontier models, emphasizing rigor, security, and independence in evaluation protocols.

Apple Store architect Ron Johnson says Apple's secret sauce has always been its people.

TypeSafe AI unveils Jev, a 'decision model' outputting structured numeric predictions instead of text for classification and confidence scoring.

Cloudflare Python Workers reaches general availability, enabling Python code execution via WebAssembly in their serverless platform.

A simple ClickFix attack is only one way to completely hijack the new agent.

Diogo Almeida (TypeSafe AI) discusses System One models for production deployment vs. theoretical optimization.

The compute and memory demands of generative AI increasingly exceed what a single GPU can provide. NVIDIA TensorRT multi-device inference is a new capability... The compute and memory demands of generative AI increasingly exceed what a single GPU can provide. NVIDIA TensorRT multi-device inference is a new capability that enables a single TensorRT network to execute across multiple GPUs using NCCL-backed distributed collectives while retaining TensorRT inference optimizations. It is fully supported starting with TensorRT 11.0. Source

When you ship an AI agent, the key question is whether it can execute a chain of work across dozens of sequential tool calls against a live environment, and... When you ship an AI agent, the key question is whether it can execute a chain of work across dozens of sequential tool calls against a live environment, and recover when a step fails. Scoring whether the model sounds right tells you almost nothing about whether the work finished. That gap is why agent evaluation has had to evolve from scoring a single function call to scoring an entire task… Source

California Gov. Gavin Newsom has signed seven bills designed to prevent AI data centers from passing utility costs onto residents, as reported earlier by the Los Angeles Times. The package of laws requires the California Public Utilities Commission to introduce a new rate classification for data centers, while forcing them to pay for upgrades to local power grids and water systems. Other bills included in the package will require proposed data centers to disclose their estimated water use to local governments, along with information about energy efficiency and drought planning. They must also...

The group won't be given leeway to slow down or redirect OpenAI's ongoing mathematical research.

Meta’s new AI agent Muse has racked up more downloads and daily active users in the U.S. and Canada than ChatGPT did over the same period after its mobile debut, according to new estimates from Appfigures.

GameHorizon Suite benchmarks multi-horizon gameplay capabilities across diverse models with unified data covering visual understanding, planning, and action control.

Critical-State RL identifies which model calls in multi-turn tool use merit training by isolating task-dependent reward signals from downstream noise.