AI Briefing#llm
The latest AI and developer news about #llm, with the original source and a short summary.
The latest AI and developer news about #llm, with the original source and a short summary.
LangChain has released the Deep Agents SDK, equipped with context compression techniques for AI agents performing complex, long-running tasks.
FuriosaAI has started mass shipments of RNGD, marking its full-scale entry into the enterprise AI inference market.
Netflix is leveraging LLMs to evolve its existing structured query-based Graph Search into natural language search.
Vercel AI Gateway now provides real-time throughput and latency metrics for hundreds of models.
Qwen 3 Max Thinking, a model with enhanced reasoning capabilities, is now available through Vercel's AI Gateway.
Moonshot AI's latest model **Kimi K2.5** is now available through **AI Gateway** without a separate account.
The Trinity Large Preview model can now be used directly through AI Gateway without a separate account.
This summarizes the concept and use of **Agent skills**, which help AI agents accurately perform complex workflows.
This analyzes AI agents, the core of the Software 3.0 era, from the perspective of the existing layered architecture.
OpenAI has revealed how the agent loop, the core logic of Codex CLI, works and its design principles.
In testing the hypothesis that bash is the optimal environment for AI agents, the SQL approach showed overwhelmingly better performance than bash for structured data queries.
Anthropic strengthened Claude's safety and helpfulness through Constitutional AI, which is based on explicit principles.
Anthropic has unveiled a new constitution that defines Claude's values and behavioral principles and will be used in training.
Replit has introduced a guidance approach that provides environmental feedback at decision points to reduce agent errors that occur along long task trajectories.
Since the release of DeepSeek R1, the China-centered open-source AI ecosystem has grown rapidly, reshaping the landscape of the global market.
To overcome the non-deterministic limitations of AI models, Replit proposes a hybrid security model that combines static analysis with LLM reasoning.
The first development log for Kanana-2 summarizes the key decisions made during the pre-training stage.
Agentic capabilities were enhanced without degrading Korean language performance by applying mid-training and a parallel RL pipeline.
Perplexity's web search capability can now be integrated and used across all AI models through Vercel AI Gateway.
Vercel has released a knowledge repository of React and Next.js optimizations, tailored for AI agents and LLMs.
Vercel AI Gateway has introduced Perplexity web search functionality, providing real-time web search capability to all LLM models.
AI services need a separate, dedicated guardrail system to compensate for the limitations of the system prompt approach.
Instead of building custom tools, using the filesystem and bash can reduce agent costs and optimize performance.
Agent evals measure multi-turn behavior to catch regressions and quality changes before deployment.