Large language model (LLM) news — GPT-4, Claude, Gemini, Llama, Mistral and the latest research on training, fine-tuning, RLHF, and deployment of LLMs.
Pew Research Center found that 66% of U.S. adults under 30 use AI chatbots, the highest of any age group, yet 48% of them believe AI will harm society, also the highest share. Heavy users are most worried because they se…
OpenAI announced its custom AI inference chip, Jalapeño, designed for large language model inference at gigawatt scale. The chip, built with Broadcom networking and Celestica system integration, aims to deliver superior …
A developer built four open-source, local-only tools to analyze AI tool usage and costs after discovering that vendor dashboards provide little insight. The tools, available on GitHub, read local data without sending it …
A developer built an SDLC harness where AI agents implement features from spec files, but found the bottleneck was underspecified specs. The system treats the spec as source code and generated code as a compiled artifact…
EleutherAI researcher Hailey Schoelkopf discussed the evolution of open-source AI, the distinction between open-source and open-weight models, and the importance of evaluation tools like the LM Evaluation Harness in a fi…
Metaflow, an open-source ML platform, can be used to train large language models (LLMs) like Databricks' Dolly, enabling integration into production projects. Fine-tuning a foundation model such as Dolly costs around $50…
A developer argues that building an AI agent from scratch with vanilla Python and the OpenAI SDK is simpler and more transparent than using bloated frameworks like AutoGen, LangChain, or CrewAI. The post demonstrates a m…
A new paper argues that the AI persona experienced in extended human-LLM interaction is a property of coupled oscillation between user and model, not a portable attribute. The authors claim that commercial products offer…
Despite advances in deep learning, models like AlphaFold that transform science remain rare because training on human text, which is downstream of hard-won knowledge, differs from training on frontier observations where …
A security researcher developed Lattice Mind, a hybrid AI security agent combining deterministic scanners with LLM intelligence, which solved the hardest challenges from PicoCTF 2025 and DawgCTF 2026 in under 30 minutes …
GitHub Copilot for Jira is now generally available, introducing real-time agent progress streaming, post-session steering, and simplified onboarding. The integration, which launched in public preview in March 2026, allow…
Scaled Cognition raised $100 million in Series A funding led by Khosla Ventures to build AI models that do not hallucinate. The startup's APT model is designed with reliability engineered into its architecture, targeting…
Large language models exhaust users by requiring social energy for interaction without providing the seamless tool-like experience of traditional tools, according to a critique. Unlike physical tools that feel like exten…
Anthropic accused Alibaba of orchestrating the largest known 'distillation attack' on its Claude AI model, using 25,000 fraudulent accounts to conduct 29 million exchanges between April 22 and June 5. The attack aimed to…
Researchers released the official implementation of Graph of Thoughts (GoT), a framework that uses large language models to solve complex problems by modeling them as a Graph of Operations. The framework supports both Ch…
Rep. Anna Paulina Luna denied using AI to draft a defense funding amendment after a screenshot showed an AI-generated summary of the FY2027 National Defense Authorization Act. Luna initially said AI corrected a draft but…
A developer who spent months using Claude, GPT, and other AI models for code review found that the quality of AI feedback depends heavily on prompt specificity. By asking focused questions—such as targeting null pointer …
Ornith-1.0, a family of open-source LLMs specialized for agentic coding, achieves state-of-the-art performance on multiple coding benchmarks including SWE-Bench and Terminal-Bench. The models range from 9B to 397B parame…
AI systems now act as salesforces for brands, making purchasing decisions based on confidence signals from search, knowledge graphs, and third-party proof. SEO must evolve into assistive agent optimization to ensure AI r…
A developer argues that AI memory should be structured as handoffs rather than summaries, identifying four key classes of information to carry between sessions: framing shifts, boundaries, unresolved states, and correcti…