Large language model (LLM) news — GPT-4, Claude, Gemini, Llama, Mistral and the latest research on training, fine-tuning, RLHF, and deployment of LLMs.
Former Google engineering leader Hugh Williams used Anthropic's Claude Code to build Zettair, a search engine indexing 1.5 million Wikipedia articles, while writing zero lines of code himself. The project leverages an in…
PAR Technology Corporation built a multi-tenant LLM analytics system on AWS that enforces row-level security through a three-layer architecture using cryptographic request signing, semantic validation on Amazon Bedrock, …
A developer released AST-guard, an open-source tool that uses deterministic abstract syntax tree analysis to detect reward hacking in AI-generated code, achieving 96.2% recall on a benchmark of reward hacks while remaini…
A new quiz challenges readers to distinguish AI-generated text from human writing, highlighting the growing ability of both people and search engines to detect AI content. The creator's tool aims to rewrite AI text to so…
Tech Mahindra is deploying Perplexity Enterprise Pro across its sales and customer-facing teams to provide real-time, source-backed insights for researching customer priorities and identifying business opportunities, as …
A developer merged Andrej Karpathy's CLAUDE.md principles with a battle-tested system prompt used to build over 100 full-stack apps in under 12 months. The resulting system prompt enforces strict scope lock, requires rea…
A developer argues that retrieval-augmented generation (RAG) for codebases improves context but not verifiability, citing that 30% of failed SWE-agent runs still claimed success. They introduce 'truth', a verification to…
A developer built JobLens, a full-stack job search intelligence platform using Amazon Aurora PostgreSQL on Serverless v2, AWS Bedrock with Amazon Nova Pro, and Vercel. The platform includes a Kanban tracker, JD analyzer,…
AI will worsen biological extinction risks in the near term by accelerating scientific progress, enabling the development of extinction-level pathogens sooner, even if AI models refuse to assist directly. The argument th…
Qwen 3.6 27B, a dense local language model from Alibaba's Qwen team, impresses developers with its general intelligence and practical coding abilities, running efficiently on consumer hardware via llama.cpp. The model pu…
Zhipu AI's GLM-5.2 open-weight model matches Anthropic's Mythos in cybersecurity benchmarks, raising concerns that US export controls have failed to slow China's AI progress. The model's open-weight nature allows unrestr…
Cursor launched a mobile app for iOS that lets users prompt coding agents from their phone, following similar moves by Anthropic and OpenAI. The app integrates with Cursor 2.0's agent-based coding system, reflecting a br…
Claude Code and Codex both use OS-level process sandboxing to prevent AI models from executing harmful shell commands, relying on Linux primitives like bubblewrap for filesystem isolation and seccomp for syscall filterin…
A new benchmark, dspy-security-bench, reveals that DSPy prompt optimization degrades adversarial robustness against harder prompt-injection attacks. Testing with AgentDojo's attack suite, optimizers like BootstrapFewShot…
An open-source coding agent called Relay has been released, supporting non-mainstream and Chinese LLM providers like DeepSeek, Qwen, and GLM. The Electron app offers chat and code workspaces with file editing, command ex…
Cursor launched a native iOS app in public beta, enabling developers to launch and control AI coding agents from their phones. The app supports cloud-based agents and remote control of desktop agents, with push notificat…
GitHub Copilot is rolling out a preview of Claude Opus 4.8 (fast mode), offering faster output token speeds while maintaining the same intelligence as the standard model. The fast mode is available at reduced cost compar…
A developer who switched from running Ollama locally to using MiniMax through OpenClaw reports that the model significantly improved their daily workflow. After three months of use, they found MiniMax-M2.7 excels in code…
A new character-level language model called EntropyBeam, using gradient-free count tables and a Dirichlet prior, achieved a validation loss of 1.596 nats on the Shakespeare character benchmark, outperforming nanoGPT's 2.…
OpenAI's GPT-5.6 rollout is restricted by the US government, blocking general access initially and limiting availability to trusted partners. This follows similar restrictions on Anthropic's models, highlighting a trend …