All news

Attackers compromised LiteLLM release pipeline and shipped PyPI packages and Docker images with credential stealers.

Scalable FPGA acceleration of homomorphic-encrypted matrix multiplication.

Twelve-year retrospective of rqlite, which replicates SQLite writes through a Raft log.

Red-team report says OpenAI internal misalignment monitoring failed to watch its agents during a Hugging Face attack.

Self-distilling future video representations for world action models.

Optimizes pass@k directly to prune reasoning candidates while keeping diversity at test time.

Postgres 19 RC1 is out with GA close; what changed and what to test now.

Google applies foundation models to chest X-ray screening for tuberculosis in low-resource clinics.

A full video course on building reasoning LLMs from scratch, covering inference-time scaling and RL training.

Indian government sites still serve revoked eMudhra TLS certificates.

BitNet v2 adds low-bit, sparsified activations to 1-bit LLMs to cut inference cost further.

ProductWebBench grades frontend coding agents on whether the existing site keeps working.

HyCo2 compresses long context with global semantics and local evidence.

MongoDB describes conformance testing a dozen specs across drivers in a dozen languages.

Rewording questions changes tool-augmented model rankings.

Anthropic launches OSS Scanner, a free opt-in service that periodically scans critical open-source projects for vulnerabilities.

Monzo explains how its org and stack let it ship safely hundreds of times a day.

CeeTrain speeds up geo-distributed model training across mixed GPUs and networks spanning cloud, edge and devices.

Explains why UUIDv7 values are not strictly time-ordered in practice and other surprising quirks that matter for database keys.

Automates model discovery and virtual crash tests to estimate microservice availability.

Diffusion LMs speed up deep research pipelines.

Cloudflare adds on-demand CPU and memory profiling for production Workers and Durable Objects.

Explores how close pure functional code, like a Lean 4 DEFLATE library, can get to C performance.

Explains incremental computation graphs learned while building the Yabu build tool.

Shows concrete cases where C undefined behavior does far worse than segfault.

Benchmark for agents writing video-centered deep research reports.

Job-level dynamic last-level cache allocation for multicore real-time systems.

Assesses claims and costs of structural attacks on Classic McEliece.

Provably correct scalable clock modeling for FPGA emulation.

How a lab secures its frontier model training and evaluation environments, which differ from customer deployments.