Large language model (LLM) news — GPT-4, Claude, Gemini, Llama, Mistral and the latest research on training, fine-tuning, RLHF, and deployment of LLMs.
GitHub Copilot's shift to token-based billing on June 1 revealed that heavy users could see costs rise from $29/month to over $750, exposing the hidden expense of AI coding tools. A developer reported burning $1,850 in u…
A researcher proposes training frontier AI systems on a synthetic world rather than real-world data to prevent them from learning the details of human society and their own deployment, arguing that knowledge of the real …
A CTO cut inference costs by 40-65% by replacing GPT-4o with a mix of DeepSeek, Qwen, and GLM models via the Line AI Chatbot framework, which uses a model-agnostic API to avoid vendor lock-in. The system routes requests …
Alibaba DAMO Academy released AgentScope 2.0, an open-source production-ready agent framework designed to leverage LLM reasoning without rigid pipeline constraints. The framework adds event systems, permission controls, …
Sakana AI launched Fugu, a multi-agent system delivered as a single model that dynamically orchestrates frontier models to tackle complex multi-step tasks. The system is accessible via the Sakana API and can be installed…
A June 2026 research survey of six coding agents—Codex CLI, Kimi Code CLI, MiMo Code, Cursor, Antigravity CLI, and OpenCode—found that while all support a project-level AGENTS.md file, global instructions are fragmented:…
A developer explains that Generative Engine Optimization (GEO) is an engineering challenge requiring developers to ensure AI crawlers from ChatGPT, Claude, and Perplexity can access and render content. Key technical step…
A paper by Saha et al. (arXiv 2506.07001) demonstrates that detector-guided paraphrasing using RoBERTa as a reward reduces the true positive rate of AI-generated text detectors by 87.88% across Binoculars, Fast-DetectGPT…
A new tool called BenchPress allows users to predict any large language model's score on any benchmark, using a score matrix derived from reported evaluations. The project invites community contributions to expand its pr…
A developer released 'claudemax', a launcher for the Claude Code harness that restores extended-thinking summaries on Opus 4.7/4.8, re-enables the context-usage icon in VS Code, and adds a 'Copy as Markdown' icon to mess…
A developer introduced a code-graph approach to supercharge AI coding assistants by modeling a codebase as a knowledge graph, reducing token waste from repeated system orientation. The method maps architecture, not just …
A developer created a launcher script to run GLM 5.2 in the Claude Code harness, enabling the use of the GLM model via Anthropic's API. The script loads the API key from a private file, sets environment variables for the…
Baidu released Unlimited-OCR, a new model that uses Reference Sliding Window Attention to parse up to 40 PDF pages in a single inference pass, eliminating the linear memory growth of traditional LLM-based OCR decoders. T…
Ions, an open protocol for publishing and traversing evidence-backed claims called Cognitive Building Blocks (CBBs), has been launched. In a test, an 8B parameter model using the Ions network matched or exceeded a fronti…
A developer built a WhatsApp AI bot in 2026 using open-source models and a unified API endpoint from Global API, avoiding vendor lock-in. The setup costs 40-65% less than proprietary alternatives while maintaining compar…
LectuLibre built a robust EPUB parsing and rebuilding pipeline in Python to translate entire books while preserving visual structure. The team used ebooklib for metadata handling and lxml for fast, namespace-aware XHTML …
The Haskell Foundation will host a 2.5-hour online workshop on Friday, June 26, featuring five talks on AI applications in Haskell development, including LLM integration, tooling, and agentic coding. The event streams on…
Legion LegalTech Corp sued the US government on June 23, 2026, challenging a Bureau of Industry and Security order that forced Anthropic to block foreign access to its Fable 5 and Mythos 5 AI models, disrupting the compa…
Researchers stress-tested Natural Language Autoencoders (NLAs) by optimizing activation vectors to flip AV explanations while preserving model behavior, achieving an 81.4% flip rate with 99.6% label preservation. Further…
A researcher proposes Tree Transformers, a generalization of the transformer architecture that operates directly on tree-structured data instead of vectors, preserving internal data structure. The approach defines self-a…