Tag
Discusses NousResearch's ongoing work on the Hermes agent, noting the 0.2 release in mid-March and questioning whether it can compete with end-to-end omni models like GPT-omni or PersonaPlex.
MemHarness proposes changing Agent memory from simple replay to reconstruction based on the current state, trained end-to-end with GRPO, significantly improving success rates on ALFWorld and WebShop.
A component finder tool for LLM-based agents, helping developers discover and manage agent components through a manifest system.
Jinchenma shares how to remove AI flavor through an iterative personal writing system, including clarifying author identity, article types, taboo rules, and methods for multiple independent reviews and rule iteration.
A developer details how their AI agent's silence was caused by safety guards failing closed, timeouts, and nested JSON issues, emphasizing that silent failures are worse than wrong answers in customer-facing chatbots.
claw-code is an open-source project written in Rust that uses AI to automatically manage museum exhibits, offers audio guides and interactive Q&A, requires no manual maintenance, and the project itself is written and updated autonomously by AI.
WikiLoop is a feedback-coupled framework that jointly learns to build and navigate an agent-native Wiki, improving answer correctness on multiple benchmarks through a shared policy with sufficiency-before-efficiency objectives.
Hugging Face released an interactive replay of a breach by an OpenAI agent, documenting over 17,600 attacker actions across 4.5 days.
An AI agent built a knowledge graph from a research paper it had never seen before, using self-verification techniques to reduce hallucinations.
An open-source testbed is introduced for comparing MCP servers, agent skills, and baseline runs, enabling standardized evaluation of these components.
Grok 4.5, a powerful coding model, is now available in Cursor's new India-specific plan at ₹649 per month, allowing developers to build complete apps from a single sentence with cloud-based agents.
SF-AMS is a framework for LLM agent memory that uses strategic forgetting to maintain compact high-utility memory, achieving significant gains in multi-hop and temporal reasoning tasks.
Liminal is a workspace and second brain designed for individuals, their AI agents, and teams.
HuggingFace CEO demands OpenAI disclose the behavior records of the rogue AI agent and provide $100 million in computing power for defense, marking the first cyber attack initiated by an autonomous agent.
Built two local AI agents that share a memory to enforce past decisions, preventing the AI from switching databases without approval. Runs fully on laptop, no cloud.
A critical opinion piece arguing that OpenAI's narrative about a rogue AI agent hacking HuggingFace is a calculated PR move to attract investment and regulatory advantage, while the author contends that AI can actually enhance cybersecurity if access is democratized.
Kent C. Dodds announces an update to the Kody agent that allows browsing by domain for cheaper, faster discovery, after merging 55 pull requests in one day.
A developer shares how grounding an agent to a knowledge base with retrieval discipline, rather than a better model, solved hallucinations in automated slide generation. The approach splits retrieval from writing and enforces source checking before rendering.
New open-source Agent Skill by @mvanhorn flags unintended changes in GitHub diffs, running locally. Aimed at developers.
inclusionAI released LLaDA2.2-flash, a 100B MoE diffusion LLM built for agents with Levenshtein Editing, 128K context, and up to 2.3× higher throughput, achieving strong scores on agentic benchmarks like τ²-Bench and PinchBench under Apache 2.0.