The reporting

9 articles

Topic: ai infrastructure

Iron filings arranged along the magnetic field lines of two attracting bar magnets, illustrating interacting forces between elements

ANALYSIS AI Infrastructure

Physics-guided pruning halves a 70B AI model while keeping its scores, says Multiverse Computing: here is what is verified and what is not

Multiverse Computing researchers say treating block removal as an Ising glass optimization problem let them halve Llama-3.3-70B-Instruct without retraining while keeping MMLU near its original score. The numbers are self-reported, but the method is open-sourced and testable.

Closeup photograph of the rear of a server rack at the NERSC data center, showing blue LED screens and network cables on dense compute hardware.

NEWS AI Infrastructure

OpenAI's GPT-6 caching stores more of what an AI already read, cutting repeated input costs by up to 90% and trimming wait times

OpenAI's September 22, 2026 announcement details a rebuilt caching system for GPT-6 that stores more of the context an agent already processed, discounts reused input tokens by up to 90 percent within a 30-minute window, and gives developers new dashboards for diagnosing cache misses. Here is what that means for the cost and speed of long AI sessions.

A hidden Markov model diagram with three hidden states labeled A, B and C, transition probabilities, and arrows leading to two observed nodes labeled X and Y.

ANALYSIS ARTIFICIAL INTELLIGENCE

Belief-State Engine Separates Uncertainty Tracking From LLM Decisions

An external Bayesian filter gives an LLM a compact posterior instead of a growing history, but the first experiments are small and incomplete.

A vertical software-development life-cycle diagram linking planning, requirements analysis, design, implementation, testing, deployment and maintenance with blue arrows.

NEWS Technology

Palantir Adds Isolated Tests for LLM Pipeline Changes

Pipeline Builder can now test Use LLM nodes against selected cases and store row-level results before teams change a production pipeline.

A UML class diagram labeled Queue with one generic list field and create, enqueue and dequeue operations.

ANALYSIS Developer Tools

Kubernetes 1.37 Moves Group Scheduling to Beta for Coupled AI Workloads

Kubernetes 1.37 promotes group scheduling and preemption to Beta while leaving hierarchical workload placement behind Alpha gates.

Three white circles numbered 1, 2 and 3 connected by four black directional arrows.

NEWS AI

Codebook Agent Moves Multi-Agent Topology Search Out of the Critical Path

A preprint reports faster per-query topology selection and lower token use by learning reusable communication patterns from prior executions.

A radial tree diagram of purple circular nodes joined by black lines into several branching levels.

ANALYSIS Developer Tools

GitHub Enterprise Server 3.22 Centralizes Access While Testing Disconnected Copilot CLI

The self-hosted release adds enterprise-wide access controls and a preview path to administrator-selected AI models, alongside significant upgrade cautions.

A labeled block diagram of a generic GPU showing its bus interface, memory controller, power and video units, display interface, compression unit and graphics compute array.

ANALYSIS Technology

CUDA Rust Takes Two Experimental Paths Toward Safer GPU Kernels

NVIDIA’s cuda-oxide and cutile-rs bring Rust ownership checks to GPU kernels, but both remain experimental tools with different safety boundaries.

A brightly lit aisle between rows of black server racks filled with silver rack-mounted computers.

ANALYSIS AI

Cloudflare Gives AI Security Scanning the Production Context It Usually Lacks

Cloudflare’s early-access service links authorized code analysis with traffic and firewall context while customers retain deployment control.