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 developer warns that building products on top of LLM APIs creates a two-way pipe where usage data signals the platform provider to absorb the product as a feature. The post traces how categories like AI astrology, stoc…
A new analysis of major AI models reveals systematic political biases, with most models leaning left on economic and social issues. The study, which tested models like Gemini, DeepSeek, Llama, Claude, Grok, and ChatGPT o…
Vercel released AI SDK 7, a major update to its TypeScript SDK for building AI applications, adding production-grade features for agent development, including reasoning control, tool context, runtime context, file and sk…
A developer handed over a month's worth of business bookkeeping to Claude Code, which reconciled 200 transactions and miscategorized 11, achieving a 94.5% accuracy rate. The agent built a reusable pipeline that handled b…
Google query expansion broadens search queries to deliver more accurate results, allowing content to rank for unplanned terms. SEOs can use Google Search Console to identify these opportunities and improve content visibi…
Large language models are transforming software development by making coding accessible and affordable, akin to how IKEA revolutionized furniture assembly. This shift enables individuals to build utility software cheaply…
A developer released Unread, a command-line tool that uses any LLM to turn Telegram chats, YouTube videos, web pages, and files into summarized reports with clickable citations. The tool aims to help users quickly digest…
OpenAI and Azure OpenAI's JSON mode endpoints cannot correctly emit escaped Unicode characters like é, instead producing NUL bytes that break systems such as PostgreSQL. The constraint is undocumented and affects any UTF…
A new paper by Simon Willison, highlighted on Bruce Schneier's blog, reveals that large language models suffer from role confusion, enabling prompt injection attacks. The research shows that models learn to recognize tex…
Mistral launched OCR 4, a document understanding system with bounding boxes and 170-language support at $4/1000 pages. Baidu open-sourced Unlimited-OCR, a 3B-parameter MIT-licensed model for multi-page PDFs. Anthropic in…
LiteLLM launched the Lite-Harness SDK, enabling developers to swap between AI agent harnesses such as Claude Code and Codex without rewriting application code. The SDK provides a unified query interface, supports cost co…
A tutorial walks through building a small but complete language model from scratch using PyTorch, using character-level tokenization and explaining each component's purpose. The guide targets developers familiar with PyT…
A developer is building agentic coding skills and asks the Hacker News community which command-line interfaces (CLIs) they find most useful as tools for LLM agents, citing the Jira CLI and GitLab CLI as examples.
Anthropic's Mythos AI security tool, touted as too dangerous to release, found only one low-severity vulnerability in the heavily audited curl codebase after a privileged deployment. The curl project lead dismissed the h…
A developer built a log monitoring Python script using DeepSeek, but the generated code hallucinated and required extensive manual fixes. The script, intended to tail Nginx error logs and send Slack alerts, referenced mi…
A hybrid inference architecture that decouples reasoning from execution is reducing AI costs by up to 60%, according to data from recent open-source utility deployments. The approach, detailed by Genesis Park, shifts foc…
Z.AI's open-weight GLM-5.2 model outperforms GPT-5.5 on the SWE-bench Pro coding benchmark, scoring 62.1 versus 58.6, while costing $1.40 per million input tokens compared to GPT-5.5's $8.00. Released under an MIT licens…
The launch of Z.ai's GLM 5.2 and DeepSeek V4 Flash has created a 50x price gap between open-weight APIs and closed frontier models, reshaping the build-versus-buy calculus for developers. While open-weight models democra…
A practical guide for benchmarking fine-tuned models recommends starting with a held-out test set matching the actual task rather than relying solely on public benchmarks. The workflow includes defining target behavior, …
Hugging Face's top-voted papers on June 25, 2026, reveal a trend of AI shifting from question-answering to world-acting models. Key papers include Qwen-AgentWorld, which uses a language world model for generalist agents;…