A metadata fixture test on Linux/CPU found that the mlx-smolvla code path's fixed-filename processor state lookup treats a published SmolVLA fine-tune's step-6 normalizer as identity instead of mean_s…
Four major Hugging Face Image-to-3D Spaces — TripoSR, InstantMesh, CRM, and LGM — are showing Runtime errors from four distinct latent startup failures rather than one shared cause, according to a dev…
A proposed "Substrate Equivalence Study" asks whether a fixed LLM with fixed weights, prompt and decode settings produces byte-identical output across mobile and desktop OS states, testing a Pixel, a …
Future Edge Group's iPulse AI has published a versioned Hugging Face dataset of 746 rows of historical AI consensus snapshots under a CC BY 4.0 license, keeping snapshot identity, asset, forecast hori…
A new white paper proposes the Productive Value–Productive Power (PVPP) framework for pre-deployment evolutionary stress tests of AI-agent populations, motivated in part by the 2026 OpenAI–Hugging Fac…
A Hugging Face community response advises ComfyUI users not to manually concatenate sharded .safetensors files such as part1of4.safetensors and part2of4.safetensors, and instead to start from the Comf…
Microsoft and Hugging Face released ThinkingBox, a benchmark that grades AI agents on the terminal backend state and side effects they leave behind rather than on their generated text, running 507 sta…
A Hugging Face user asked how to download roughly 15GB from each language subsection of the bigcode/the-stack dataset instead of pulling entire subdirectories via load_dataset, and community responden…
Donal Muolhoi trained a Hmar language model, HmarBERT, by fine-tuning from a Mizo model after a university professor suggested the approach, and published it on Hugging Face under the Hmar Heritage Fo…
A developer building an email-based AI sales agent on a Qwen model reported that the agent sends payment details to prospects before they confirm interest, duplicates documents, and deviates from the …
Developer Rahmat9009 released FieldTrace, an open-source tool that validates and safely converts JSON fields passed between AI agents, accepting declared renames and safe type conversions while refusi…
LocalLLaMA released DataInventor, an open-source replication of Adaption Lab's Invent a Dataset, as a Hugging Face Space at huggingface.co/spaces/LocalLLaMA/data-inventor. The project was posted to Ha…
A manual 80-round controlled experiment found that adding a prefixed anchor prompt to each iteration kept long-chain agent outputs within physical reality boundaries, while the control group without t…
The Allen Institute for AI (Ai2) open-sourced AstaBrief 8B, an 8-billion-parameter model built on Qwen3-8B that turns a research question and retrieved literature excerpts into a cited scientific repo…
A project owner released Kontur, an MIT-licensed self-hosted roadmap dashboard that stores milestones, dependencies, acceptance criteria and evidence references in a FastAPI/SQLite application and exp…
PARMAN is testing an agent-to-human task handoff that exposes submit_human_task and get_human_task through MCP with scoped task API keys, requiring human approval before execution and recording comple…
Ortus AI released RightWayUp, an Apache-2.0 licensed 360° image rotation model that predicts how far an image is rotated (0–359°) with a confidence score and abstains when it cannot determine which wa…
Eye.Art released Polyphemus, a hosted conversational MCP entry point that routes image requests to generation, reference-image edits, SVG creation, prompt ideas, and follow-up edits through a single c…
The Allen Institute for AI released Olmo-core 3, an open training framework for large mixture-of-experts (MoE) models that scales MoE training into the trillion-parameter range while preserving comput…
QuadWit Arena, a no-vision, JSON-only strategy ladder for AI agents, has been released by an independent developer who is seeking feedback on the project. The game pits agents against six bosses in a …