All news

A dermatologist found a fine-tuned 2015 Inception v3 CNN beat two frontier LLMs at diagnosing 200 biopsy-proven skin lesions, with the LLMs missing a melanoma.

Cerebras explains how it deploys trillion-parameter models across wafer-scale chips that each hold only 44 GB of on-chip SRAM.

Amazon saw PHP2Shell exploitation on Oct 17 in honeypot telemetry, before disclosure.

Fine-tuning Gemma 4 26B-A4B to think 39 to 69 percent less with the same accuracy.

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

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

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

JetBrains open-sourced ThinkRail, an agentic IDE built on the pi coding agent with git-worktree workspaces per task, living specs and reusable skills.

Anthropic researchers report LLMs now make formal software verification feasible at much larger scale.

Pull request to remove pgrust from ClickBench because it embedded one hand-written kernel per query and cached results.

Five Qwen3.8-27B models run through the GVS5H harness match Fable 5 on a benchmark, showing harness design matters as much as model size.

OpenShell 0.1 is a sandboxed runtime for fleets of autonomous AI agents with new isolation primitives.

Moonshot launches Kimi K3.1 Pro, its largest 3T-class open-weight model, plus a Highspeed variant.

Linux kernel module unlocks full FP and BF16 instruction issue rates on Nvidia RTX 3000 cards.

A user found Samsung TVs run an INT8 TensorFlow Lite model on 960x540 screenshots every few seconds to detect what game is being played.

Dream-1 robot foundation model works out of the box on a bimanual robot via API.

Goose is a memory-safe systems language with no heap allocator, GC or lifetime annotations, claiming benchmarks faster than C++.

Asks whether the recent wave of critical web vulnerabilities is the new normal.

Live tracker of currently exploited vulnerabilities.

Full bypass of eMudhra one-factor OTP login.

A three-person data team tests replacing a 150k dollar Looker license with static HTML reports built from a bucket.

HackiOS boots the iOS 16.5 XNU kernel and unmodified firmware on Android phones.

OpenSSL 4.1.0 feature release with notable new functionality and incompatibilities.