The sharp decline in AI inference costs is being driven by software, and local open models are quickly catching up to frontier model performance.
Graph spectral analysis can predict the grokking phenomenon in neural networks much earlier than the loss function can.
Supercut MCP has launched to give AI agents access to video records and metadata.
Warp has expanded Oz into a multi-harness cloud agent control plane.
KPMG has partnered with Anthropic to deploy Claude company-wide and in core operations.
cdesktop, an open-source Claude Code Desktop alternative tool supporting various AI models and providers, has been released.
Bug fixes related to MTP and speculative decoding improvements have been merged into llama.cpp.
Forge, a guardrail framework that improves tool-calling reliability of local LLMs, has been released.
Maximum activation values across 27 open LLM checkpoints varied significantly depending on architecture and training stage.
ByteDance has released Lance, a 3B open-source model that combines image and video understanding, generation, and editing.
Apple leverages Apple Intelligence to introduce new accessibility features for people with visual and physical disabilities.
Sapient Intelligence has launched HRM-Text 1B, a 1B model that achieves high reasoning performance with little data and low cost.
OpenAI has strengthened image provenance verification with C2PA, SynthID, and public verification tools.
Standard Chartered will cut more than 7,000 jobs to improve efficiency through AI adoption.
Through information theory, the existence of collective predictive information within multi-agent LLM systems has been numerically demonstrated.
Marlin, a 2B open-source VLM based on Qwen3.5-VL, has been released.
Qwen has unveiled Qwen3.5-LiveTranslate, a real-time interpretation model that leverages visual context as well.
Cloudflare warned about the possibility of real cyberattacks using Mythos AI.
Ccimi split its architecture between Amazon IVS and in-house infrastructure to achieve 4K low-latency live streaming under 4 seconds.
Yeogieotae used Spring Modulith to enforce module boundaries and event flows in code.
Thinnest AI, a programmable voice AI infrastructure that lets developers build custom AI voice agents, has launched.
An npm package supply chain attack has been discovered that risks hijacking AI coding tools such as Claude Code and Codex.
Samsung researchers proposed a method for online handwriting generation using character-independent point classification and stroke transformation.
Instead of adopting Sentry, we built our own 200-line error tracker to gain operational visibility.