Google has released Magenta RealTime (Magenta RT), an open-source AI model for live music creation and control. The model responds to text prompts, audio samples, or both. Magenta RT is built on an ...

OpenAI has rolled out a major update to ChatGPT's integrated search, introducing smarter answers, better handling of long conversations, and a new image search feature.

Researchers at the University of Copenhagen are taking another look at the "Othello world model" hypothesis, asking whether large language models can pick up the rules and board structure of Othello ...

Musk is asking users on X to submit examples of these statements directly under his post. In this new training round, Grok 3.5 will first be used to rewrite data, a version Musk says might also be ...

A new study from Anthropic suggests that large AI models can sometimes behave like disloyal employees, raising real security concerns even if their actions aren't intentional.

Japanese company Sakana AI built an AI agent that can tackle complex optimization problems used in industry. In a live competition, their AI went head-to-head with more than 1,000 human programmers.

Apple executives have been talking internally about potentially buying AI startup Perplexity AI, according to a Bloomberg report. The idea is to grab both the technology and talent for Apple's own ...

Researchers have found an unexpected way for multimodal AI models to learn mathematical reasoning: by playing simple arcade games like Snake and Tetris, rather than training on math datasets.

A new dataset called FineWeb-Edu highlights the importance of high-quality learning content for the performance of large language models.

OpenAI has rolled out an updated version of GPT-4o in ChatGPT, extending its knowledge base through June 2024.

Anthropic has enhanced its AI assistant Claude with two new capabilities: an agent-based research function and Google Workspace integration, both designed to significantly expand the chatbot's ...

A new study reveals that large language models such as GPT-4 perform much worse on counterfactual task variations compared to standard tasks. This suggests that the models often recall memorized ...