124 stories tagged Models
Researchers have developed a technique to make text watermarks in open-source LLMs survive model merging, a common post-training modification that previously removed such watermarks. The work, published on arXiv, addresses a key vulnerability in tracking AI-generated text.
Anthropic has released Claude Opus 5, an AI model that closely matches the capabilities of its more advanced sibling, Fable 5. The launch comes amid heightened industry scrutiny following recent AI security incidents and Anthropic's own tensions with the US government.
A new study on the open-weight LLM Qwen3-8B reveals that AI models frequently commit to an answer before reasoning through a problem — even when the answer contradicts the task premise. Using a simple car wash scenario, researchers found the model recommended walking instead of driving in 85–100% of cases, despite driving being the only logical choice.
Lunar, an AI research lab, has released an open-source AI model that runs on standard consumer hardware such as gaming PCs, eliminating the need for expensive servers and making advanced AI accessible to hobbyists and small businesses.
Researchers introduced Cura 1T, a healthcare-specialized AI model trained via a human-gated self-evolution loop to handle patient consultations, clinical reasoning over text and images, interactive diagnosis, and EHR tool use without degrading performance across tasks.
Chinese AI companies Moonshot and Alibaba have unveiled models—Kimi K3 and Qwen—that they claim match the performance of top US models from OpenAI and Anthropic at a fraction of the cost, signaling a narrowing gap in AI leadership.
Charles Azam tested Fable 5 and GPT-5.6 on an NP-hard problem and found that both models performed significantly better when given a specific /goal directive. The results suggest that framing tasks with clear objectives can substantially improve AI problem-solving performance.
Dharma-AI released an open-source model on Hugging Face that rivals proprietary AI systems in performance, democratizing access to advanced AI capabilities for developers and researchers worldwide.
Moonshot is launching Kimi 3, a massive open AI model with 2-3 trillion parameters, expected to rival top Western models like Anthropic’s Opus 4.8. This could democratize access to cutting-edge AI technology globally.
A new arXiv paper shows that point-in-time language models—trained only on data available up to each calendar date—can now perform nearly as well as unrestricted models, eliminating lookahead bias that previously compromised financial backtests and causal inference.
Researchers created CANDI-QA, a new test for AI models to evaluate their ability to answer complex questions in specialized fields like medicine and finance. This helps ensure AI assistants can provide accurate, context-aware answers in critical areas.
Researchers found that compressing AI models to make them faster can subtly change how they think, even when their answers seem correct. This could affect everything from chatbots to AI assistants, making some reasoning errors harder to detect.
Hugging Face's CEO argues that businesses now prefer open AI models over cutting-edge ones due to cost and control. This shift could redefine what matters in AI development.
The Hugging Face blog has published a detailed guide on profiling attention mechanisms in PyTorch, the third in a series on PyTorch profiling. The guide walks through how to use PyTorch's profiling tools to identify performance bottlenecks in transformer models, with a focus on attention layers.
OpenAI has released GPT-5.6, a new AI model family that improves performance across various tasks, including cybersecurity. This update could make online interactions safer and more efficient for everyday users.
Hugging Face has partnered with Microsoft to make its AI models available on Foundry Managed Compute, simplifying deployment for developers and businesses.
A new theoretical study analyzes how AI models learn from their mistakes through a process called in-context search. The research provides a mathematical framework for understanding when this reflection helps AI solve problems more efficiently, and when it doesn't. This could guide the development of smarter, more reliable AI assistants.
Hugging Face introduced a new native-speed backend for vLLM that dramatically accelerates inference for large language models, not training. This could significantly reduce costs and latency for developers deploying AI models.
Researchers have developed AI models that can automatically generate 3D CAD designs from plain-English descriptions. This breakthrough could revolutionize how engineers and hobbyists create mechanical parts.
Vercel's CEO argues that separating AI models from agents is crucial for better performance and cost efficiency. This shift could make AI tools faster and more affordable for businesses and developers.
Mistral AI is a French startup developing advanced AI models with a focus on open-source accessibility. The company has raised significant funding and aims to make cutting-edge AI available to everyone.
Base44, a "vibe-coding" platform that lets users build software through natural language, has released its own proprietary AI model to differentiate itself from competitors. The move highlights how AI startups are racing to build proprietary tech to secure their positions and reduce reliance on third-party providers.
AI models are becoming more specialized to handle specific tasks better. This trend is making AI more useful for everyday applications.
Researchers developed a new way to fairly compare AI models by controlling for accuracy. This helps avoid misleading conclusions when evaluating different systems.