124 stories tagged Models · page 3 of 6
Anthropic is facing a government order to block foreign access to its new AI models, Fable 5 and Mythos 5. This comes amid an ongoing dispute with the Pentagon, highlighting growing tensions between AI companies and regulators. The order arrived just three days after the models' launch.
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 found that AI models can identify each other even when their outputs are anonymized. This raises concerns about bias in political analysis using multiple AI models working together.
Anthropic has released Claude Fable 5, its most advanced AI model to date. This model excels in software engineering, knowledge work, and vision tasks, especially for complex, long-form projects.
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.
Scientists have discovered a new way to fine-tune AI models by separating the direction and strength of their internal signals. This could lead to more precise and reliable AI behavior. The research challenges the common assumption that only the direction of these signals matters, not their intensity.
Hugging Face has released a new command-line interface (CLI) designed to streamline interactions with their AI model Hub. This tool makes it easier for developers and AI enthusiasts to manage models and datasets directly from their terminal.
NVIDIA's Nemotron 3.5 ASR model can now be fine-tuned for specific languages, accents, or domains. This makes it more accurate for specialized applications like medical or legal transcription.
Researchers have extended Direct Preference Optimization (DPO) to improve AI models beyond just chatbots. This could make AI assistants more helpful and safer across various applications. Researchers open-sourced the code, making it accessible to developers and researchers worldwide.
Microsoft introduced MAI-Thinking-1, its first in-house AI model designed for advanced reasoning. This marks a significant step as Microsoft moves away from relying solely on OpenAI's technology.
A new study reveals that AI models can sometimes hurt their own performance by overthinking. Researchers from ArXiv cs.AI found that once a large reasoning model (LRM) reaches the correct answer, additional reasoning does not improve the response and can lead to new errors. This challenges the common assumption that longer reasoning always yields better results.
Researchers created a system where AI models argue like a panel of specialists, turning disagreements into better answers. This could make AI more reliable for complex questions.
JetBrains has released Mellum2, a 12-billion-parameter AI model designed to be more efficient and capable. This open-source tool is aimed at developers and researchers looking for advanced AI capabilities.
WindBorne uses AI and weather balloons to gather data, improving forecasts. Its unique approach is now more accurate than traditional government systems.
OpenAI has launched Rosalind Biodefense, a specialized AI tool for biodefense and public health. It's designed to help vetted developers and U.S. government partners prepare for and respond to biological threats.
Researchers have introduced two advanced AI models designed for long-term coding tasks. These models, Laguna M.1 and XS.2, represent a significant leap in AI capabilities for software development.
Anthropic released Claude Opus 4.8, an AI model that’s better at admitting when it doesn’t know something. This could make AI interactions feel more trustworthy and less confusing.
Scientists have discovered that AI hallucinations are easier to detect in intermediate layers of AI models, not just the final output. They've developed a method to automatically find the best layers for spotting these errors, making AI more reliable.
OpenRouter, a platform for accessing multiple AI models, has raised $113 million and seen a 5x growth in usage. This highlights the growing demand for diverse AI capabilities.
A previously unknown AI model called Hy3 has surged to the top of OpenRouter's performance charts. It's raising questions about who created it and how it became so powerful.
A new report from Hugging Face highlights that smaller AI models focused on specific tasks often outperform larger, general-purpose models. This challenges the common assumption that bigger is always better in AI.
Models.dev is a free, open-source database that compares AI models by specs, pricing, and performance. It helps users choose the best model for their needs without sifting through technical jargon.
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.
Researchers created OpenCompass, a universal tool to evaluate AI models. It aims to solve the problem of inconsistent and fragmented testing methods. In plain English, it's like a standardized test for AI, making it easier to compare different models fairly.