NeoSQL, a DB workbench that integrates ERD design, SQL editing, and code generation while supporting AI MCP, has been released.
Dynamic batching boosted throughput by 3.3x when training NLLB-200 600M on an RTX 5090.
Resolved Gemma 4 E2B multimodal OOM in llama.cpp on iOS.
Compared the performance of Qwen 3.6 27B's BF16, Q4_K_M, and Q8_0 quantizations.
Chinese authorities warned that ByteDance's apps and website failed to comply with AI content labeling requirements.
GitHub Copilot is changing from the existing per-request billing method to a token-based AI Credit billing system.
Microsoft has open-sourced its VibeVoice ASR and Realtime models.
Anthropic is investigating unauthorized access to Claude Mythos Preview in a third-party vendor environment.
OpenAI has reportedly missed its internal targets for new users and revenue.
DeepSeek has teased the release of a new vision model.
A survey of 400+ papers organizing agentic world models into a three-tier taxonomy and four domains has been published.
Meta's acquisition of an AI startup has been blocked, increasing uncertainty around cross-border tech deals involving China.
GitHub apologized for recent outages and unveiled improvement plans to prepare for 30x scale.
A llama.cpp benchmark comparing ROCm 6.4.2 and Vulkan on the 6900 XT was published.
Running Qwen 35B-A3B always-on on a 16GB Mac M4 hit an SSD bottleneck before a RAM one.
A 13B vintage model called Talkie, trained on 260B tokens of pre-1931 English, has been released.
Qwen 3.6 27B was remodeled with Opus CoT and Hermes tool-calling SFT and released in FP8.
An Emory research team uncovered the asymmetric interactions of dusty plasma with **99%** accuracy.
Xiaomi MiMo has open-sourced MiMo V2.5-Pro, a 1.02T MoE model.
Analysis suggests OpenAI is developing a new smartphone interface centered on AI agents rather than apps.
Starting June 1, GitHub Copilot code review will also consume GitHub Actions minutes.
Anthropic has strengthened safeguards to ensure Claude provides accurate and neutral information without political bias during election periods.
The open-source coding agent Dirac scored 65.2% on Terminal-Bench-2 and 8/8.
This is a research paper that directly compares and analyzes the performance and efficiency of MoE and Dense architectures.