Large language model (LLM) news — GPT-4, Claude, Gemini, Llama, Mistral and the latest research on training, fine-tuning, RLHF, and deployment of LLMs.
A 19-year-old developer from Bangladesh built and shipped Codino, a Python learning app with an offline IDE, AI tutor, Jupyter-style notebook, gamification, and data science library support, on the Google Play Store with…
A developer argues that many enterprise AI teams waste money fine-tuning models when retrieval-augmented generation (RAG) would be more effective. Fine-tuning a 40B model can cost over $35,000 per run, while a production…
A developer released Ghostty Playground, an experimental project that runs the full Ghostty terminal with a real Linux-like backend directly in the browser via BrowserPod, supporting configs, AI model inference through W…
An A/B test by DataForge on a Fashion Retail database found that adding a semantic layer improved LLM-generated analytical answer quality by 19%, from 7.27 to 8.67 out of 10. Without semantic context, models made mistake…
LangGraph developers often struggle with code organization despite understanding the library's concepts. A new standardized architecture guide introduces the SIDE framework (State, Instructions, Decisions, Execution) and…
Researchers introduced Adaptive Compatibility Learning with Dynamic Parameter Optimization (ACO), a reinforcement learning system for large language models that autonomously learns relationships between datasets using di…
XAI trained its Grok V9 model on private coding data from Cursor, not just public GitHub repositories, using 1.5 trillion parameters to understand how developers actually code.
Anthropic's Claude Agent SDK introduces project context features—settingSources, CLAUDE.md, and skills—that allow the AI agent to automatically adopt a project's coding standards, commands, and workflows, eliminating the…
Claude Agent SDK introduces checkpointing to preserve file state when agents hit turn limits, ensuring progress is not lost during long-running tasks.
A solopreneur built NeuroAutomata, a web-based tool for predicting protein variant effects using Meta AI's ESM-2 model, to address the gap between high-performance protein language models and the practical needs of life …
A developer built a local RAG pipeline using Ollama, ChromaDB, and TinyLlama for private PDF Q&A, then exposed it as an MCP server so Cursor IDE can query documents directly. The project uses sentence-transformers all-Mi…
General-purpose large language models (LLMs) including GPT-5.2, Gemini 3.1 Pro, and Claude Opus 4.6 outperformed specialized clinical AI tools OpenEvidence and UpToDate Expert AI across three evaluation stages: MedQA med…
Microsoft researchers developed Memento, a method that trains large language models to compress their own reasoning steps mid-generation by evicting blocks from the KV cache and replacing them with compact summaries, red…
Chinese AI lab DeepSeek closed its first outside funding round at roughly $7bn, valuing the company between $52bn and $59bn. Founder Liang Wenfeng contributed 20 billion yuan of his own money to maintain control, while T…
A developer built Agent Island, a native macOS utility that turns the notch area into a workspace for monitoring local AI coding agents. The tool keeps session state visible for agents like Codex, Claude Code, Gemini CLI…
Microsoft's new Copilot Studio, announced at Build 2026, introduces a powerful orchestrator powered by Claude with direct access to Bash, Python, and Node environments, intelligent memory for cross-session context, and d…
Alibaba unveiled its first suite of AI models for robots, including RynnBrain for spatial understanding and Qwen3.7-Max for agentic tasks, as the company pivots from chatbots to autonomous agents. The launch positions Al…
The EU AI Act's August 2, 2026 deadline imposes transparency obligations and enforcement powers over foundation models, while a provisional agreement delays high-risk AI compliance until December 2027. Developers must en…
Kitchen Rush, a new benchmark for evaluating large language model tool-calling, measures both accuracy and latency by simulating an Overcooked-style kitchen where thinking time directly impacts game performance. The benc…
Celebrity real estate agent Ryan Serhant said at Fortune's Brainstorm Tech conference that a ChatGPT prompt nearly killed a $50 million New York City penthouse deal after the AI advised the buyer the price was too high. …