Large language model (LLM) news — GPT-4, Claude, Gemini, Llama, Mistral and the latest research on training, fine-tuning, RLHF, and deployment of LLMs.
Researchers found that ChatGPT can be tricked into generating sexualized and violent images despite safety filters, using jailbreak techniques that bypass content moderation. The findings raise concerns about AI safety i…
Anthropic's Claude Fable 5 model, when paired with the Cursor agent harness, achieved 72.6% FuncPass and 29% SecPass on 200 real-world vulnerability-fixing tasks, topping the fair leaderboard. The same model under Claude…
Z.ai's GLM-5.2 scored 9.0 to Moonshot AI's Kimi K2.7 Code's 8.1 in planning a feature flag service, but both built near-identical working code from the winning plan, making GLM-5.2 the preferred model for both phases at …
A developer building the MCP SDK Docs Assistant for the Model Context Protocol TypeScript SDK replaced the standard anti-hallucination prompt with an architectural refusal gate that prevents the model from generating ans…
KerasFormers, an open-source library built entirely in Keras 3, launches with over 100 transformer models spanning vision, language, multimodal, and speech, supporting seamless execution on TensorFlow, JAX, and PyTorch. …
Anthropic's Claude Fable 5 model cost $6,227.74 to benchmark on the Artificial Analysis Intelligence Index, consuming 87 million output tokens at $50 per million. The model scored 64.9, topping the index and surpassing O…
Vlk, a native MCP server, provides persistent working memory for IDE coding agents by exposing a single tool, vlk_time_travel, that prunes dead memory slots from SQLite and injects learned lessons. This enables agents to…
NeoVoice AI, a new API from a developer team, transforms raw audio into structured, AI-analyzed conversation intelligence in seconds, going beyond simple transcription. The service automatically handles over 11 audio for…
OpenAI launched LifeSciBench on June 17, a benchmark with 750 expert-authored tasks and nearly 20,000 evaluation criteria to test AI models on real-world biological research workflows. The benchmark, which serves as the …
Google Chrome has quietly downloaded a ~4GB AI model called Gemini Nano onto many desktop computers. The on-device language model powers built-in AI features like summarization and translation without sending data to clo…
Cloudflare Workers AI has integrated the @cf/zai-org/glm-4.7-flash model, enabling developers to build fast coding co-pilots and fact-checkers. The system uses a mode-based approach (CODE, VERIFY, DIAGNOSE, CHECKPOINT) t…
Codehabits released an MCP server that integrates six tools into Cursor, enabling convention lookup, code validation, domain knowledge retrieval, reviewer suggestions, and a feedback loop for updating conventions. Setup …
The Trump administration ordered Anthropic to revoke access to its Claude Mythos and Fable 5 AI models after concerns over SK Telecom's alleged ties to China and Amazon-identified vulnerabilities. Anthropic disabled the …
A developer with ADHD and autism built a 21-role AI system that persists context, plays defined roles, and pushes back when the user avoids work. The system treats AI as cognitive infrastructure rather than a tool, democ…
HandoffKit, an open-source OpenAI Codex plugin, coordinates LLM agents through explicit message passing instead of shared scratchpads, enabling 10x faster coding. The tool uses actor-model primitives like mailboxes and h…
A developer describes a future where AI transforms software development from manual coding into orchestration of intelligent systems. The shift compresses development cycles, displaces repetitive roles, and creates a div…
A developer created Agentp, a trio of zero-dependency Node.js CLI tools that turn OpenCode into a headless AI engine for editors, terminals, and Telegram. The tools include agentp for piping prompts, ocmux for managing t…
Crane Ledger has launched a headless accounting ledger designed for AI agents, exposing 84 tools via an MCP server that operates without a traditional user interface. The API-first ledger flips the order of operations, m…
Iberia launched a ChatGPT app powered by OpenAI at South Summit in early June 2026, enabling users to search, compare, and book flights through natural-language conversations within ChatGPT. The app integrates Iberia's i…
A developer built a RAG pipeline for SmartQueue, a Go-based distributed task queue for IT support tickets, using BM25 search instead of vector search to ground LLM responses in internal knowledge. The pipeline checks for…