Large language model (LLM) news — GPT-4, Claude, Gemini, Llama, Mistral and the latest research on training, fine-tuning, RLHF, and deployment of LLMs.
The European Union held talks with the White House to regain access to Anthropic's advanced AI models after a US export-control order cut off European access in mid-June. The order, citing a jailbreak concern, forced Ant…
KV Cache reduces duplicated computation in autoregressive LLM inference by storing previously computed Key and Value tensors, but creates a memory bottleneck as context length grows. To address this, Multi-Query Attentio…
A developer building a CLI AI agent found that planning mode was harder than direct execution. The agent required a separate planner with read-only tools to avoid making plans based on assumptions rather than real data. …
Investor Chamath Palihapitiya said Meta “completely fumbled” its lead in artificial intelligence, arguing the company failed to capitalize on its data and distribution advantages to become the dominant open-weight AI lab…
Baidu released Unlimited OCR, an open-source 3-billion-parameter model that uses Reference Sliding Window Attention to parse multi-page documents in a single pass, eliminating the need for page-by-page OCR pipelines. The…
Multi-agent systems that perform well in demonstrations often fail in production due to memory inconsistencies, according to a new analysis. The article highlights the challenges of maintaining coherent memory across mul…
NVIDIA and Google DeepMind have released open-source omni AI models—Nemotron 3 Nano Omni and Gemma 4 12B IT—that process text, images, audio, and video in a unified architecture, enabling enterprise and local multimodal …
Anthropic disabled global access to its Fable 5 and Mythos 5 AI models after a US export control directive, prompting EU leaders to push for a 'trusted partners' framework to restore allied access. The move has disrupted…
Replit CEO Amjad Masad evaluated live agents powering SaaStrAI on stage at SaaStr AI 2026, highlighting three deployed agents (10K, QBee, and a third) and revealing that SaaStrAI runs about 10 apps in a single mono-repo.…
Microsoft Excel has added federated financial connectors for LSEG and Moody's data, enabling Copilot to query live market data via the Model Context Protocol. The update allows users to access real-time foreign exchange …
A developer argues that the biggest AI opportunity in web development lies not in building new apps but in modernizing existing ones. The post contrasts greenfield development with brownfield modernization, emphasizing t…
Kog released Laneformer 2B, a 2.3B-parameter coding model designed for high-speed decoding, on Hugging Face Hub. The model was trained from scratch with latency as the primary objective, achieving 45.1% HumanEval+ and 51…
Developer l-zhi released Pith, a local-first desktop LLM wiki that ingests notes, PDFs, and emails into Markdown entries without using vector databases or embeddings, relying instead on keyword search and link traversal …
Google's GKE Inference Gateway introduces prefix-aware load balancing for LLMs, routing requests to replicas that already hold cached context to avoid reprocessing shared prompt prefixes. This approach reduces prefill ph…
Galois is leading two projects, Metron and Praxis, to build formal verification benchmarks that measure AI's ability to reason about code correctness, aiming to make industrial-scale software verification practical and r…
A tech leader's tour of China's leading AI labs reveals that Chinese researchers excel as fast-followers in large language model development due to a culture emphasizing collective optimization over individual ego, contr…
A developer who built 10+ MCP servers and experienced 85 production outages found that poor tool discovery causes more failures than bad code. The engineer built a fuzzy discovery layer in Java Spring Boot to handle LLM …
Buildkite engineers implemented async LLM inference through the Bifrost AI gateway to prevent build workers from blocking on slow model calls. By switching to a submit-and-poll pattern, workers no longer wait up to 35 se…
Chinese cybersecurity company 360 announced it has developed artificial intelligence tools that match the capabilities of Anthropic's Mythos, a large language model. The development signals China's growing competitivenes…
A developer created a deterministic routing system for LangGraph agents, replacing LLM-based routing with a 50-line rule-matching module. The solution uses a YAML-defined route table with keywords and regex patterns, ens…