Researchers at XJDR have developed a new technique called Entropix that aims to improve reasoning in language models by making smarter decisions when the model is uncertain, according to a recent blog post by Thariq Shihipar. The method uses adaptive sampling based on two metrics, entropy and varentropy, which measure the uncertainty in the model’s …
Researchers have discovered that Whisper, an AI-powered transcription tool used in various industries including healthcare, is prone to making up text or entire sentences, known as hallucinations. According to interviews with software engineers, developers, and academic researchers by AP, these hallucinations can include problematic content such as racial commentary, violent rhetoric, and imagined medical treatments. …
Cohere For AI has released Aya Expanse, a family of highly performant multilingual models that, according to Cohere, excel across 23 languages and outperform other leading open-weights models. The models, available in 8 and 32 billion parameters, are part of Cohere’s commitment to multilingual research and expanding high-quality coverage of languages in large language models …
DeepMind researchers have introduced a new agentic framework called Talker-Reasoner, which is inspired by the “two systems” model of human cognition. The framework divides the AI agent into two distinct modules, VentureBeat reports: the Talker, which handles real-time interactions with the user and the environment, and the Reasoner, which performs complex reasoning and planning. The …
Notion, a productivity platform with over 100 million users, has announced the launch of its new email client, Notion Mail, at its first “Make with Notion” user conference. Currently in preview, Notion Mail integrates with Notion AI and Notion Calendar to provide users with a more flexible and organized email experience, TechCrunch reports. The company …
Meta has entered into a multi-year agreement with Reuters to integrate the news agency’s content into its AI chatbot, enabling users to receive real-time answers to queries about news and current events. According to Axios, users of Meta’s AI chatbot feature in the U.S. will have access to Reuters’ journalism when asking questions, with answers …
The U.S. Department of Labor released a list of artificial intelligence best practices for developers and employers this week. Here’s the PDF. The highlights:
OpenAI CEO Sam Altman has responded to an exclusive report by The Verge journalists Kylie Robison and Tom Warren, which claimed that the company plans to launch a new frontier AI model codenamed Orion by December, calling it “fake news out of control” on X (formerly Twitter). The report, citing sources familiar with the plan, …
Anthropic’s latest AI feature “Computer Use” can demonstrate remarkable capabilities in interacting with computers, according to a recent hands-on experience by Thariq Shihipar. While the system is still slow, unreliable, and prone to taking over the user’s computer, it represents an exciting step forward in AI-driven computer interaction, write Shihipar. Claude Computer Use excels at …
Cerebras has announced a major update to its Cerebras Inference platform, which now runs the Llama 3.1-70B language model at an impressive 2,100 tokens per second – a threefold performance increase compared to the previous release. According to James Wang from the official Cerebras blog, this performance is 16 times faster than the fastest GPU …