A Hacker News user asks the community to share specific tasks from the past month where open-weight models like GLM, DeepSeek, Kimi, or Qwen failed but frontier models like Opus, Fable, or GPT succeed…
Researchers have developed a new approach to overcome the quadratic cost of causal self-attention in long-context transformers, using state update design and structural interventions like sink tokens …
Fine-tuning adapts a pretrained large language model to a specific task using task-specific data, building on the general language understanding learned during pretraining. Pretraining teaches a model…
Crusoe launched serverless fine-tuning and inference services on its Intelligence Foundry platform, enabling enterprise AI teams to customize open-weight models without managing GPU infrastructure. Th…
A new study finds that backdoor triggers in large language models often fail to reach the model due to context pipeline preprocessing, invalidating many robustness evaluations. In about 1 million tria…
Neuronpedia announced the J-space, a global workspace in AI models revealed by Jacobian lens, and added support for 11 more models. The feature allows users to see hidden reasoning in AI, with pre-fit…
A developer outlines a strategy for building profitable micro-SaaS platforms in 2026 using AI coding assistants, local AI models, and automation tools. The approach focuses on utility platforms that s…
An independent researcher found that existing indirect prompt injection benchmarks like BIPIA, InjecAgent, and AgentDojo may produce unreliable scores due to reliance on LLM-judges and evaluation of e…
A developer built a free web tool that visualizes and edits the internal representations of open-source language models, inspired by Anthropic's research on the Jacobian Lens. The tool allows users to…
MemLedger, an open-source AI agent memory framework, provides transparent provenance for every stored fact, enabling users to trace the origin, extraction model, and approval history of memories. It u…
Apple is reportedly interested in PrismML technology for running compressed large language models on-device, including a 27-billion-parameter Qwen 3.6 model on an iPhone 17 Pro, according to a 9to5Mac…
VectorNode highlights the critical importance of AI API request logs for multi-model applications. Without detailed logs tracking latency, token usage, routing decisions, and fallback behavior, teams …
A new benchmarking platform, Agentic Coding Arena, compares 21 frontier AI models from OpenAI, Anthropic, Google, Moonshot AI, Z.ai, and Qwen by having each build the same 52 small apps from identical…
China's Ministry of Commerce is considering sweeping restrictions on AI technology, including bans on public releases of frontier models, criminalizing tech leaks, and limiting foreign investment in l…
A developer building the code map tool Sense tested whether larger context windows and smarter models would eliminate the need for structural indexes in AI coding agents. Benchmarking five frontier mo…
A new benchmark called ImagingBench reveals that leading vision-language models and agentic AI systems, including Gemini, GPT, and Qwen, perform poorly on computational imaging tasks involving physics…
A developer reports experiencing burnout from extensive daily use of large language models (LLMs) for coding and information retrieval, citing repetitive writing patterns, hallucinations, and stylisti…
LLM orchestration coordinates models, providers, and steps to make AI features production-grade, handling routing, failover, caching, guardrails, and observability. A dedicated orchestrator sits betwe…
OpenCode released data on real-world AI model usage, cache ratios, and costs, showing DeepSeek's v4-flash leading with 8.2T tokens, followed by DeepSeek v4-pro and Zhipu's glm-5.2. The data reveals sh…
A developer built a local AI sports agent using a DeepSeek R1 distilled Qwen 32B model on a LazyCat AI Pod, but found that local models lack practical value due to slow responses, outdated knowledge, …