TikTok's parent company ByteDance has developed an AI system called INFP that can make static portrait photos appear to speak and react to audio input. What sets INFP (which stands for "Interactive, ...

Researchers have developed a more streamlined approach to help AI systems process information. The new system, called RetroLLM, combines two previously separate steps - searching for information and ...

Researchers at Google have figured out how to make AI language models both faster and better by using an unusual approach: letting smaller AI models teach the bigger ones. A joint team from Google ...

Zipeng Fu, one of the developers of Mobile-ALOHA, actually took the robot home to demonstrate that the flexible and relatively inexpensive hardware can perform various household tasks, such as doing ...

Alibaba's AI lab Qwen recently introduced a new framework designed to help developers create what they call AI agents. Qwen-Agent builds on Qwen's existing language models, giving these agents ...

Gary Marcus kicked off the year with his characteristic skepticism. "OpenAI will continue to preview products months and perhaps years before they are solid and widely available at an accessible price ...

The AI world is abuzz with predictions about agents in 2025. But some of the people laying the groundwork are more cautious about the timeline. Logan Kilpatrick, who heads up Google AI Studio and ...

Instead of playing a proper game of chess against Stockfish, a dedicated chess engine, o1-preview figured out how to hack its test environment to force a win. According to Palisade Research, an AI ...

Meta's ExploreToM framework shows that even the most sophisticated AI models, including GPT-4o, have trouble with complex social reasoning tasks. The findings challenge earlier optimistic claims about ...

Microsoft researchers have developed what they're calling a "Large Action Model" (LAM) - an AI that can operate Windows programs on its own. This marks a shift from AIs that just talk about doing ...

Nvidia significantly ramped up its startup investments in 2024, pouring $1 billion into 50 different funding rounds and several corporate deals. That's a significant jump from its $872 million ...

Users have discovered a way to bypass Deepseek V3's content filters through prompt engineering. By asking the model to insert periods between letters, they can get it to provide more balanced or China ...