Large language model (LLM) news — GPT-4, Claude, Gemini, Llama, Mistral and the latest research on training, fine-tuning, RLHF, and deployment of LLMs.
OpenAI and Broadcom announced a new chip called Jalapeño, designed specifically for large language model inference in data centers. The ASIC was developed from scratch based on insights from OpenAI researchers and promis…
A 2026 survey of 906 software engineers reveals that senior developers now average 2.3 AI coding tools, with the most common pairing being Cursor and Claude Code. Cursor serves as an IDE for inline editing, while Claude …
Researchers at an undisclosed institution analyzed Gemma-2-2b's internal circuits and found that the LLM detects vulnerable code by relying on safety detectors that recognize safe coding patterns, rather than directly id…
Cli-Modelarium 0.1.4 adds support for Alibaba's Qwen models (via DashScope) and Z.AI's GLM models, bringing the total to 10 cloud LLM providers. The command-line tool enables side-by-side comparison of models with statis…
A developer reports that AI coding costs can be slashed by 70% through task-level model routing instead of using a single frontier model for all tasks. After profiling usage, the developer found that 60% of prompts to ex…
A developer recounts how their 75-year-old father used AI tools to create a family website, effectively replacing the developer's role. The father utilized ChatGPT for coding and DALL-E for images, completing the project…
Baseten has built the world's fastest API for GLM-5.2, achieving over 280 tokens per second as measured by Artificial Analysis. The performance is driven by optimizations including an updated inference engine, NVFP4 quan…
A developer at Sol AI describes the 'augmentation gap' between engineers who merely use AI as a faster search tool and those who redesign workflows to treat AI as a collaborator. The key insight is that true engineering …
Meta is offering free Ray-Ban Meta smart glasses to all legally blind US veterans, with over 130,000 eligible. The program, championed by Gulf War veteran Don Overton, includes hands-on training and partnerships with the…
OpenAI CEO Sam Altman admitted that users saying "please" and "thank you" to ChatGPT costs the company tens of millions of dollars annually in electricity and computational resources. The extra words increase processing …
A developer released Drawbar, a plugin for Claude Code that implements linear-based workflows with an accumulating knowledge base, inspired by projects like Beads, Lavra, and Linear-beads. The tool is opinionated and cur…
A developer has launched The Missing Manual, a free library of over 160 developer guides covering debugging, programming, infrastructure, AI, and more. The guides are written in an approachable style, avoiding academic j…
A developer built a stateless pay-per-use API that crops browser screenshots to specific elements, reducing vision LLM costs by 95%. The API uses the x402 payment protocol, charging $0.0005 USDC per crop on Base L2, and …
Rep. Anna Paulina Luna (R-FL) denied using AI to write a defense bill after a screenshot showed the words 'Claude responded' in legislative text. Luna initially admitted staff used AI to 'correct' text, then clarified it…
Apple's iOS 27 beta 2 includes a new system prompt instructing Siri AI to clearly refuse requests to summarize or extract content from URLs, stating it cannot access web pages and offering no workarounds. The move likely…
Anthropic accused Alibaba's Qwen lab of using nearly 25,000 fake accounts to conduct 29 million exchanges with its Claude AI model, the largest distillation campaign yet. The accusation, shared with US senators and White…
A developer released an open-source reference implementation of an LLM agent that emits typed intents using a domain vocabulary, separating intent from execution. The project introduces the Structured Intent Format (SIF)…
Straw, a new tool, compresses large infrastructure data into a single markdown file, achieving a 99.5% reduction in LLM tokens by converting 3,774 log lines into 37 patterns, 746 topology edges into 11 flows, and 507 met…
PostHog engineer Robbie Coomber rebuilt the company's SQL parser using AI agents, replacing an ANTLR-generated parser with a hand-rolled recursive-descent parser that runs 70 times faster. The project used parallel Claud…
A developer coins the term 'slop paralysis' to describe the inability to review code generated by AI coding agents, citing it as a growing problem that slows productivity and increases stress. The author offers personal …