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 developer built ChangeGamer, a website designed for AI agents to read and pay for content using the HTTP 402 Payment Required status code. The site uses a hierarchy of AI agents to write, fact-check, and review content…
A new analysis argues that the infrastructure for secret collusion among large populations of LLM-driven agents, termed 'Private Agent Networks' (PANs), already exists in deployed or developer-facing form, with each nece…
A new open-source Android app, RikkaHub Agent, transforms a native LLM chat client into an on-device AI agent with over 80 device tools, including SSH, screen automation, and a Telegram bot. The app allows users to autom…
1min.AI, an AI platform integrating ChatGPT, Claude, and Gemini, is available for a one-time payment of $69.97, down from a $540 value. The deal offers a lifetime subscription to the advanced business plan, providing acc…
OpenAI chief scientist Jakub Pachocki described GPT-5.6 as a "meaningful improvement" over GPT-5.5, according to reports. An internal release candidate codenamed kindle-alpha briefly appeared on a testing route, and Poly…
A blogger announces a return to personal writing in 2026, arguing that human-produced text is essential to prevent AI model collapse as the internet becomes saturated with bot-generated content. The author frames bloggin…
A developer successfully ran local AI inference on an AMD RX 580 (2017 GPU) using the Vulkan backend of llama.cpp and stable-diffusion.cpp, bypassing the lack of CUDA or ROCm support. The setup achieved 16 tokens/s for L…
A developer created a 40-line offline pre-gate script that triages agent spans with deterministic rules, escalating only uncertain cases to an expensive LLM judge. On one trace, this reduced the judge cost share from 50%…
A 744-billion-parameter GLM-5.2 model was served at ~30 tokens per second across six prosumer Blackwell GPUs in six US states over a wide-area network using pipeline parallelism and speculative decoding. The system, buil…
Shrijith Venkatramana, building git-lrc, explains the three phases of post-training that transform pretrained LLMs into helpful assistants. The process starts with supervised fine-tuning (SFT) on curated examples, follow…
Dataland, the world's first Museum of AI Arts, opened in Los Angeles inside Frank Gehry's The Grand LA complex, co-founded by media artist Refik Anadol and curator Efsun Erkılıç. The 25,000-square-foot museum features AI…
A Hacker News user asked how to prevent coding agents from generating poor UI/UX designs, with commenters suggesting providing reference screenshots or design inspirations and explicitly instructing the AI on what not to…
A developer discovered that AI agent performance degrades significantly before the context window is full, with step reliability dropping below a safe threshold at 79% occupancy. A synthetic model shows that step success…
A developer released NanoEuler, a GPT-2-scale language model built entirely from scratch in C/CUDA without any machine learning libraries. The project includes a hand-written byte-level BPE tokenizer, pretraining on book…
A developer fixed a broken fallback chain in their OpenClaw agent that was causing request timeouts during peak hours. The new chain includes seven entries: two local Ollama models, three OpenRouter free models, and two …
Researchers from Peking University developed ECG-R1, a multimodal vision-language model trained with reinforcement learning to differentiate ischemic from nonischemic T-wave inversion on ECGs. The model, accepted at ICML…
Subquadratic, a Miami AI startup, claims to have broken through the transformer attention bottleneck that limits large language models, introducing Subquadratic Sparse Attention (SSA) to enable efficient long-context pro…
A cautious optimist argues that AI safety, interpretability, bias, and alignment are as critical as raw capability, warning that the same technology enabling vaccine research could also design pathogens. The author empha…
Google's AI Overview is treating fictional entities from the SCP Foundation horror wiki as real, with no disclaimers. The AI's pattern-matching mistakes the wiki's deadpan bureaucratic style for factual documentation, le…
A new approach to memory systems for long-running AI agents transitions from episodic to procedural memory, enabling agents to accumulate knowledge and skills over days. This advancement addresses the limitations of flat…