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 user tested over 100 Claude Code skills and identified six that significantly improved their daily workflow, as detailed in a post on Towards AI.
A developer explains why AI-generated code often produces large, monolithic files with duplicated logic and inline styles, prioritizing prompt completion over long-term maintainability. The article contrasts AI's 'path o…
A 99% obedient LLM still produces tens of thousands of malformed outputs daily at scale, causing production outages. The root cause is that prompt engineering and validation loops only persuade, not guarantee, correct ou…
A month-long analysis of 1,263 prompts from agent-driven development reveals that prompt length is irrelevant; the key to efficient coding is clarity of intent. The median prompt was 78 characters, and the most effective…
A study benchmarking six RAG architectures on a financial PDF reveals that Vision RAG achieves the highest answer correctness (0.823) but low faithfulness (0.538), while Graph RAG is the most faithful (0.902) yet least c…
Tencent is internally testing an AI agent called Dayuan for its enterprise communication platform WeCom, which reads users' chats, emails, and calendar data to provide context-aware assistance without requiring repeated …
A game developer seeks a lightweight AI model that fits within 12 GB VRAM for use with Unity 6.5 and MCP, as free daily limits for Cursor and Claude are insufficient. The user asks for recommendations or alternative solu…
A new experiment tested whether weaker AI models can effectively monitor stronger coding agents for malicious behavior, finding that detection rates improve with monitor size but vary by threat type. No monitor caught ba…
A freelance developer cut their AI API bill by 62% by switching from GPT-4o to DeepSeek V4 Flash for most workloads, saving $89.10 per month per client. The developer learned that context window size must match the task:…
A developer built CommitBrief, a local-first CLI tool that runs an LLM code review on git diffs before committing. The tool supports providers like Claude, GPT, Gemini, or local Ollama models, and ensures no data leaves …
Researchers at SPAR argue that language models can report on their own internal activations, a method called self-report, which is underexploited compared to external analysis tools. They test five open-weight models usi…
A toy transformer retains provably predictively defunct belief state data in its residual stream, shedding it only under sufficient capacity pressure. Experiments extending Shai et al. 2024 show that information about a …
A developer proposes routing routine coding tasks to cheaper models while reserving the strongest model for judgment-intensive work, addressing issues of cost, context, and key management in multi-user coding agent workf…
Focused Labs analyzes agent handoffs in multi-agent systems, highlighting that handoffs are runtime state changes rather than routing diagrams. The article contrasts orchestration patterns from OpenAI, Microsoft Agent Fr…
The U.S. government has escalated a national security clampdown on Anthropic, forcing the company to pull its Claude Fable 5 and Mythos 5 models globally after the NSA lost access to the systems. The shutdown, triggered …
The Trump administration is pressing Meta to voluntarily submit its AI models for government review before release, according to Reuters. Meta is the only major U.S. AI developer that has not yet agreed to such an arrang…
Veilus Digital released the cryptographic core of its Phantom Chat messenger as source-available code on GitHub, allowing independent verification of its post-quantum security claims. The package includes implementations…
Google fired a 7-year veteran engineer, Justin Poehnelt, after he built an open-source, agent-ready CLI for Google Workspace that went viral on Hacker News. The tool, now hosted under Google's own GitHub org, dynamically…
AI engineering faces a contradiction between paying premium for frontier models' reasoning capabilities and aggressively compressing context to reduce costs, leading to a 'fallacy of context compaction' that starves agen…
A developer compared API pricing versus self-hosting costs for open-source AI models, finding that for small projects with 1 million tokens per day, using an API is 32 times cheaper than self-hosting. The analysis reveal…