Large language model (LLM) news — GPT-4, Claude, Gemini, Llama, Mistral and the latest research on training, fine-tuning, RLHF, and deployment of LLMs.
Anthropic accused Alibaba of orchestrating the largest known campaign to clone its Claude AI model, generating over 28.8 million exchanges through 25,000 fraudulent accounts between April and June 2026. The attacks targe…
The softmax function, essential to modern AI systems like GPT, originated from Ludwig Boltzmann's 19th-century work on thermodynamics, where he derived the exponential form to maximize entropy in gas molecules. The same …
Public.com integrated its Model Context Protocol server with ChatGPT, enabling users to analyze portfolios, research stocks, and execute trades across stocks, options, crypto, and bonds through natural language conversat…
A developer compares using Claude Code to an addictive productivity drug, describing how it has replaced most of their manual coding and made them fear losing their skills. They find Opus 4.8 more accurate than GPT 5.5 b…
Sakana AI released Fugu, a language model that acts as a coordinator routing tasks to multiple frontier LLMs from OpenAI, Anthropic, and Google, returning a single answer and bill. The system hides vendor seams and aims …
Google retired two Gemini image preview models, gemini-3.1-flash-image-preview and gemini-3-pro-image-preview, on June 25, 2026, forcing developers to migrate to generally available replacements immediately or face broke…
A developer built a verifiable graph of 469 music engineers and producers by starting with 12 trusted names and using an LLM as a filter, not a source, to expand the dataset without hallucination. The system, hosted at 2…
A developer evaluated six LLM-as-judge tools—DeepEval, Confident AI, Evidently, Braintrust, Promptfoo, and Future AGI—and found that none of them prioritize validating judge outputs against human labels as a default firs…
Maneshwar, building git-lrc, explains that a single agent with a growing toolbox should be maxed out before splitting into multiple agents. When splitting is necessary, two patterns emerge: a manager agent that coordinat…
Anthropic's Claude is gaining paying consumers, with credit card data showing a 75% revenue increase since January 2026, while DataCamp reports Claude is now its most searched term and demand for Claude courses outpaces …
Z.AI's open-weight GLM-5.2 (Max) model climbed to second place on the Code Arena frontend leaderboard by June 16, scoring up to 1,595 points and beating Claude Opus 4.7 (Thinking) by 29 points. The 744B+ parameter Mixtur…
Mariana Souza published a tutorial on quantizing and running Meta's Llama 3.2 3B model on Apple Silicon using llama.cpp with Metal GPU acceleration, achieving local inference with Q4_K_M quantization. The guide covers bu…
Red Alice AI released the first official benchmark of its Version 2 architecture, reporting a 200x performance gain in the RedTensor engine. The upgrade introduces a PyTorch-backed TorchTensor backend with unified flat i…
DeepReinforce released Ornith-1.0, an open-source family of coding models that learn their own reinforcement learning scaffolds, achieving state-of-the-art results among open models. The lineup includes four sizes from 9…
A group of publishers owning nearly 400 local and regional newspapers sued OpenAI and Microsoft for allegedly stealing hundreds of thousands of articles to train ChatGPT and Copilot. Separately, Anthropic accused Chinese…
GitHub is experiencing a surge of AI-generated repositories that lack quality standards, with functions having high cyclomatic complexity, no tests, and high code duplication. Stars and forks are no longer reliable indic…
A German court ruled that Google is liable for false information in its AI-generated search summaries, rejecting defenses that users should not blindly trust AI. The decision holds that the summaries are expressions of G…
AI costs in production systems are driven by wasteful architecture—retry loops, excessive token use, and redundant validation—not by model choice. As agentic systems scale, the hidden economics of distributed AI interact…
Context bloat, the accumulation of low-signal tool-call output in an agent's context window, degrades long-running agent performance. Anthropic's analysis found token usage explains 80% of performance variance on BrowseC…
An engineer analyzing Cursor's usage dashboard found two requests with nearly identical token counts (1.34M and 1.40M) that cost vastly different amounts: $1.13 and $2.96. The cost discrepancy is explained by the model's…