Full-text search across 71001 articles. Combine with topic and date filters; results sorted by relevance.
Meta's Muse Spark 1.1 scored 51 on the Artificial Analysis Intelligence Index, improving 8 points over its predecessor in three months. The model is cost and token efficient compared to peers, with gains in scientific re…
Two bugs in the open-source coding agent Hermes highlight silent failures that can compromise security. One fix addresses secret redaction leakage, where secrets were not properly scrubbed from logs and worktree metadata…
Reinforcement learning from verifiable rewards (RLVR) has proven effective for math and code tasks but fails for tool-using agents due to sparse terminal rewards. Researchers propose decomposing rewards into intermediate…
Researchers have developed a model-agnostic framework that integrates privacy-enhancing coded computing with defense mechanisms to protect distributed machine learning systems from privacy breaches and malicious interfer…
Researchers introduced DecompRL, a reinforcement learning algorithm that breaks complex problems into smaller sub-tasks for large language models, reducing GPU token costs by approximately 50 times. The method outperform…
Researchers disclosed Ghostcommit, a proof-of-concept attack that hides prompt-injection instructions inside a PNG image referenced by an AGENTS.md file, causing AI coding agents to leak repository secrets. A survey of 6…
Researchers propose adaptive AI contracts that selectively use detailed evaluations to reduce costs, but practical challenges remain. Empirical tests show promise in question-answering and code-generation tasks, yet scal…
Avi Press, chair of the Haskell Foundation board, moved Scarf's production backend off Haskell after seven years, citing that AI coding agents made slow compilation economically unbearable. Press's post has generated hun…
OpenRouter and Andreessen Horowitz analyzed over 100 trillion tokens of real-world LLM usage, finding that open-weight AI models now account for roughly one-third of all token volume. Chinese open-weight models surged fr…
Researchers introduced CORA (Coherent Orthogonal Rotation Adaptation), a parameter-efficient fine-tuning method that uses orthogonal rotations and diagonal scaling to outperform existing approaches like LoRA with up to 8…
A developer built a WhatsApp CRM system for a Durban salon, reducing the no-show rate from 30% to 8% and automating follow-ups. The project highlighted the importance of local context, such as SA business hours and isiZu…
A developer created CLAUDE.md, a rules file that Claude Code reads automatically at the start of every session, to simplify interactions with the AI. The file can be applied globally or per project, aiming to reduce AI f…
Tripplet, a small independent AI platform, launched a workspace combining chat, code, and research with four specialized models led by Astro 5. The platform offers free access without an account, enabling users to build …
A developer argues that human-in-the-loop (HITL) systems in agentic AI are often a rubber stamp rather than real governance, shifting liability without improving safety. The post proposes replacing raw-action modals with…
Researchers have developed WhisperBench, a benchmark of 108 test cases, and MemGhost, a framework that crafts one-shot email payloads, to demonstrate how persistent personal agents can be compromised via stealth memory i…
Meta faces backlash over a series of controversial AI and privacy decisions, including launching an AI image generator that uses Instagram photos, facing EU fines for addictive platforms, and adding facial recognition co…
A developer outlines rules for delegating tasks to subagents in a Claude-powered coding workflow, emphasizing context economy and reasoning quality over parallelism. The guidelines cover batch sizing, model routing, skil…
OpenAI, Meta, SpaceXAI, and Anthropic released a flurry of new AI products and models over 72 hours, including OpenAI's GPT-5.6 family and GPT-Live voice model, Meta's Muse Image and Muse Spark 1.1, and SpaceXAI's Grok 4…
Researchers at Tel Aviv University, Technion, and Intuit have identified HalluSquatting, a supply-chain attack where AI agents can be tricked into using hallucinated resource names that attackers pre-register. The attack…
Anthropic researchers built a tool called the Jacobian lens to uncover a hidden space inside its LLM Claude, revealing words the model considers before responding. OpenAI launched ChatGPT Work, a 'super app' combining it…