Tag: #dataset24 results
Apple's mAceReason-Math provides high-quality multilingual math problems designed for Reinforcement Learning with Verifiable Rewards (RLVR). It addresses the English-centric bias in existing datasets, offering appropriate difficulty for current LLMs. The dataset supports boosting math and logic capabilities in pretrained models.
HumanMCP is the first large-scale dataset of diverse, human-like user queries for evaluating MCP servers' tool retrieval performance. It covers 2800 tools across 308 servers, built on MCP Zero, with multiple unique user personas per tool capturing precise to ambiguous intents. This addresses gaps in existing benchmarks lacking realistic user interactions.
Researchers launched CogARC, a human-adapted ARC subset with 75 visual reasoning tasks tested on 260 participants achieving ~85% accuracy. High-res data captures viewing, edits, and attempts, showing varied strategies on hard problems and convergent solutions even in errors. Findings highlight human generalization under uncertainty for AI benchmarking.
Adaption Labs released a dataset covering Nigeria's manufacturing and oil sectors, featuring 89 machine-level records and a domain-grounded chain-of-thought reasoning layer. This resource aims to improve model performance on sparse, numeric industrial data in African economies.
A new 125k coding dataset across 8 languages has been released on Hugging Face for training and upgrading coding models. Generated for free using the cloaked Hunter Alpha model. Full 450k version coming soon with collaboration open.
38-day dataset logs Gemini's daily stock forecasts, rationales, sentiment, and confidence over 10-day horizons. Time-locked and published on Hugging Face with interactive dashboard and Colab notebook. Studies LLM stability, narrative drift, and calibration under uncertainty.
EduEVAL-DB introduces a dataset of 854 explanations for 139 ScienceQA questions across K-12 subjects, with one human-teacher and six LLM-simulated teacher explanations. It features a pedagogical risk rubric covering factual correctness, depth, focus, appropriateness, and bias, annotated via semi-automatic expert review. Preliminary benchmarks compare Gemini 2.5 Pro against fine-tuned Llama 3.1 8B for risk detection on consumer hardware.
NASA provides a public repository of high-quality space imagery and videos. These assets are largely free to use, serving as a valuable resource for developers and researchers.
A query from the MachineLearning subreddit regarding the current availability and access methods for the Books3 dataset. This dataset is a central point of contention in ongoing AI copyright litigation.
Titanium Media has been selected as a founding member of the Mainstream Value Corpus Ecological Alliance. It will lead the core construction of high-quality datasets in the technology domain. The company plans to contribute premium tech corpus resources and advance quality standards for a sustainable corpus ecosystem.