18 stories tagged Training
OpenAI introduces the ChatGPT for Small Businesses program, offering entrepreneurs free access to ChatGPT Work and AI training to help automate tasks and grow their business.
A new audit framework reveals that human trainers' emotional states—such as stress or distress—can systematically bias the preference labels used in Reinforcement Learning from Human Feedback (RLHF), leading to unintended shifts in AI behavior. The study proposes methods to detect and correct this 'rater state bias.'
With traditional language model training hitting diminishing returns, researchers are turning to simulated 3D worlds to teach AI agents common sense, planning, and physical reasoning — skills that text alone struggles to impart.
UATC is a closed-loop VRAM control system with dynamic data pruning for LLM training, designed to reduce memory usage during training. While this could improve efficiency, the project appears to be a single open-source contribution with no published results or independent validation yet.
Researchers found that selecting training data can accidentally bias AI models, making them less accurate. This happens when the data used to verify the model is itself incomplete or skewed, leading to a breakdown in performance. This affects how AI systems learn and could impact everything from chatbots to medical diagnostics.
Researchers developed a way to train AI models to handle real-world problems with incomplete or ambiguous information. This could improve AI's ability to make practical decisions in everyday scenarios.
A new position paper argues that AI models should be studied as evolving processes, not just fixed products. The paper calls for a science of AI that focuses on training dynamics to better understand why model behaviors emerge, rather than relying on post-hoc fixes.
Shift, an AI training startup, is offering free home cleaning services in exchange for recording the process to train future robots. The company will use footage of cleaners scrubbing, vacuuming, dusting, tidying, and washing to improve its AI models.
Researchers found that AI models align more closely with human brains when trained on the same language. This suggests language-specific training data shapes how AI understands us, not an inherent property of any language.
Researchers developed a technique called ProxyCoT to help AI models reason better with long texts. This method trains models to use shorter, relevant parts of long documents to solve complex problems.
OpenAI is partnering with Singapore to expand AI use, train local talent, and support businesses. This initiative aims to make advanced AI tools more accessible in the region.
A new technique called verifiable process supervision (VPS) helps AI models produce both accurate answers and sound reasoning. This addresses a common problem where AI might guess correctly but use flawed logic.
AI agents in factories understand technical terms but struggle with real-world relationships between machines, processes, and rules. Researchers have identified this 'semantic training gap' and proposed solutions to bridge it.
Researchers developed a way to generate synthetic factory data for testing AI systems. This could help train AI workers without using real, sensitive factory data.
Researchers have created a framework called Weblica to help train AI agents on real websites. This tool captures and replays web pages to create stable, interactive environments for AI learning. It could make web agents smarter and more adaptable to the constantly changing internet.
OpenAI has introduced MRC, a new networking protocol designed to improve the performance and reliability of large-scale AI training. This could make AI models faster and more efficient to train.
Researchers propose Group Fine-Tuning (GFT), a method that combines imitation and reward learning to improve LLM training. GFT addresses key challenges like single-path dependency and gradient instability.
MegaTrain enables full precision training of large language models on a single GPU. It stores parameters in host memory and uses GPUs as compute engines.