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.
OpenAI rolled back a recent update and moved to correct model behavior in order to address GPT-4o's excessive sycophancy.
Intel has released AutoRound, a tool for LLMs/VLMs that maintains high accuracy even at low-bit quantization.
PipelineRL has been released, resolving the trade-off between inference throughput and data quality during LLM reinforcement learning.
This introduces a way to implement an ultra-lightweight AI agent in just 50 lines of code using MCP.
A non-developer shared their vibe coding experience of building rule-based HTML tools to reduce LLM uncertainty and automating repetitive tasks using AI agents.
Highlights how the quality of outputs has surged without changes in prompting skills due to advancements in LLM and agent technologies.
Building reliable agentic systems requires understanding the difference between workflows and agents, and controlling the appropriate context.
OpenAI has unveiled its next-generation models o3 and o4-mini, featuring powerful reasoning capabilities and agentic functionality.
FuriosaAI's AI accelerator **RNGD** has launched on Microsoft Azure Marketplace, supporting easy AI infrastructure deployment for enterprises.
Cohere's major LLM and multimodal models are now available for serverless inference on the Hugging Face Hub.
OpenAI has launched the GPT-4.1 model series via API, featuring greatly improved coding, instruction-following, and long-context understanding capabilities.
Visual Salamandra, a 7B-parameter multimodal model, has been released with image and video understanding capabilities.
A new benchmark called **BrowseComp** has been released to measure AI agents' ability to deeply explore information.
This introduces how to use an LLM as a judge to evaluate the performance of RAG systems and presents the RAG Triad framework.
Canva is building an integrated design workflow with AI that spans from idea generation to final output production.
Anthropic released Claude Instant 1.2 via API, featuring improved math, coding, and reasoning capabilities along with strengthened safety.
Hugging Face is updating and overhauling its existing NLP course into an LLM-focused, up-to-date curriculum.
OpenAI has released PaperBench, a benchmark that evaluates AI agents' ability to reproduce state-of-the-art AI research.
As agent-centric AI adoption expands, cost-effective leaner models are being introduced to secure economic viability.
Kakao Tech has compiled its major technical blog posts on LLM performance evaluation and AI agents published in Q1 2025.
**Vibe Coding**, in which users instruct AI agents to write code through natural language prompts to build apps, is changing the paradigm of app development.
DeepSeek has released DeepSeek-V3-0324, a model with improved performance and licensing.
Booking.com built an AI solution using OpenAI's LLM to understand users' complex intent and deliver personalized travel experiences.
LG AI Research has unveiled EXAONE Deep, a reasoning-specialized model with dramatically improved math, science, and coding capabilities.