Researchers from Fudan University demonstrated that 8 of 10 major AI platforms, including ChatGPT and Doubao, cited fabricated research within a week of it being posted on specific websites.
The open-weight model reduces thinking tokens to roughly half of the base model's usage while maintaining valid tool calls in agent loops.
The new variants achieve a 6.58% increase in SWE performance and use 20% fewer reasoning tokens while compressing the KV cache by 4x.
The event showcased breakthroughs including the AMIE medical AI study, Earth AI for public health, and Project Suncatcher.
The new system generates global 3 km climate data at approximately 1/300th the cost of the X-SHiELD atmospheric model.
Netflix reports consistently achieving >99.95% availability for its GPU nodes and increasing job success rates from ~98% to above 99.5% through layered health checks and automated remediation.
The new 35B-A3B MoE model utilizes Lookahead Expert Scheduling to enable local vision-language processing on smartphones.
The new leaderboard evaluates how well embedded signatures survive fine-tuning, distillation, and attacks while preserving model capability.
The new tool integrates code graph mapping and session memory into a single local MCP server, claiming significant speed improvements over the original Graphify.
The 74B parameter MoE model is packed into 24GB using 2-bit quantization, with a 4GB variant releasing Oct. 23.
The new GDN chunked kernel implementation in llama.cpp delivers performance gains ranging from 0.9% to 20.6% across various Qwen models and hardware configurations.
The fix resolved a rare empty-response bug caused by capability mismatches in shared CDN caches, achieving a 99.99% success rate.
Watermarking provides definitive verification for specific providers like ElevenLabs, while statistical tools offer probabilistic detection across broader sources.
OpenAI Daybreak enables SentinelOne researchers to reverse-engineer sophisticated malware in roughly eight hours, down from four to six weeks.
Amp integrates GPT-6 Astra, GPT-Live-1, and Ultrafast to enable agents for design, marketing, and real-time voice control.
The alpha release supports 100M+ vectors using PRISM indexing and RaBitQ quantization, claiming superior query and build speeds compared to pgvector.
The open-source app uses a hidden WebView and local embeddings to retrieve web content without any backend or search API.
Konnect exposes 228 design tools via Model Context Protocol to AI assistants like Claude and Codex, replacing the legacy Python/TypeScript KiCAD-MCP-Server with a single static binary.
KakaoBank resolved mobile WebView keyboard UI issues by using a VirtualViewport component that synchronizes the visual viewport height.
Introducing GPU time budgets and fair-share allocation instead of priorities significantly improves wait times while maintaining 98% cluster utilization
The open-source library supports non-commutative geometric spaces like hyperbolic and SO(n) rotations to preserve event order in sequence modeling.
The new office is Anthropic's fifth in Asia-Pacific and includes a partnership on AI safety and adoption.
The initiative grants direct access to tools like AlphaFold 3 and WeatherNext to support the EU’s RAISE initiative.
Research demonstrating applications across various domains, including a 31.2% code repair success rate, notes that the 99% GUI success rate is based on policy ranking.
AI Briefing