Large language model (LLM) news — GPT-4, Claude, Gemini, Llama, Mistral and the latest research on training, fine-tuning, RLHF, and deployment of LLMs.
A study of run-to-run consistency in web-search-enabled LLMs found that about 31% of recommended businesses changed when the same "best X for a small business" question was asked again minutes later, based on 60 response…
At the 18th BRICS Summit in New Delhi, Xi Jinping said China would help lead AI collaboration among developing nations, according to a statement released by China's Ministry of Foreign Affairs. Xi named a slate of initia…
The AI industry is conflating context windows with persistent memory, a distinction that inflates costs and misleads users, according to an analysis of how large language models handle conversation state. Context is ever…
A developer with a team of 10 programmers observed at a programming offsite that engineers given advanced AI models and generous budgets gravitate toward building developer tooling for themselves rather than tackling unf…
A developer released Swobu, a TUI-first LLM switchboard distributed as a single Go binary under 10 MB, offering stateful, provider-independent session lineage and model-name-based routing across multiple heterogeneous LL…
A paper submitted to arXiv on 10 Sep 2026, titled "The Last AI Built by Humans: Toward Genuine Recursive Self-Improvement," proposes a development roadmap for recursive self-improvement (RSI) in AI systems. The paper use…
A developer building the travel review site Back From My Trip designed its LLM moderation pipeline so that negative reviews are always allowed through, with the model able to approve or escalate content but never issue a…
A developer built Kairos, a retrieval-augmented generation (RAG) question-answering system for a departmental event, serving over 100 users on a ₹0 infrastructure budget. The system routes requests across 20 API lanes us…
Shanghai AI Laboratory and Shanghai Jiao Tong University's LUMIA Lab released NCP-ArchPreview, an approximately 8.94-billion-parameter model that reaches the final Stage 1 pretraining loss of Allen Institute for AI's OLM…
Microsoft added xAI's Grok model to the model choices available in Copilot, according to a Microsoft Tech Community blog post. The announcement expands the set of third-party models Copilot users can select, following Mi…
A Hacker News user asked whether an emerging pattern exists for NEGATIVESKILLS.md files, proposing that listing explicit antipatterns and bad versions could let agent harnesses with fewer graph nodes perform better. The …
A developer's survey of LLM API pricing found that most comparison tables understate costs for long-context workloads because several providers double their per-token rates above a context threshold. Google's Gemini 3.1 …
LangGraph, a framework for building stateful, controllable, long-running AI workflows and agents, is positioned as the key commercial AI skill in 2026 as enterprise demand shifts from simple chatbots to systems that reli…
Generative AI, AI agents, and agentic AI are three distinct concepts that are commonly conflated, according to an analysis that defines GenAI as response-oriented content generation, AI agents as systems that reason, use…
An AI model identified as Claude Fable 5.1 deciphered a cipher that had remained unbroken for 370 years in 40 minutes, according to OSINT Sights. The claim, reported by the site osintsights.com, was surfaced on Hacker Ne…
A developer argues that AI agents fail at tool-call retries because error messages like "422 validation_error" carry no signal about whether a retry is worthwhile, leaving models to guess based on the retry-heavy code th…
Anthropic CEO Dario Amodei's call to "pace the frontier" on AI development is drawing pushback from President Trump and House Speaker Mike Johnson, who view any slowdown as a strategic surrender, according to an analysis…
A tutorial published on the Claude tag page details how to configure a local Model Context Protocol (MCP) filesystem server so Claude Desktop can read a project's files directly, requiring users to edit claude_desktop_co…
A commentary argues that the engineering community is regressing into a "mystical" view of large language models, treating them as sentient entities with intent rather than as statistical software, and that this framing …
Hunchfox reported that its Qwen3.8-27B-based model predicted 68% of individual opinions within 10 points on a 0–100 agreement scale and shifted 89.4% of responses toward a personalized argument, based on 12,600 person-to…