This wasn't a one-off. In a text-only version of the same test, Claude Opus 4 chose blackmail 96 percent of the time. Google's Gemini 2.5 Flash nearly matched that rate. OpenAI's GPT-4.1 and xAI's ...

Elon Musk has announced plans to retrain Grok with statements he calls "politically incorrect, but nonetheless factually true," claiming this will correct and expand all human knowledge. Previously, ...

11ai also supports custom MCP servers. Teams can connect internal tools or specialized software to 11ai through their own MCP servers, extending the assistant’s functionality to fit their workflows.

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 ...

Google has handed over the Agent2Agent (A2A) protocol to a new open source project led by the Linux Foundation with the aim of creating a uniform communication standard for AI agents from different ...

The assistant is powered by a new language model called "Mu," which Microsoft developed specifically for the task. Mu has 330 million parameters and uses an encoder-decoder architecture that, ...

Snake training outperforms math datasets in some areas Training on Snake and rotation problems nudged the base model slightly ahead of MM-Eureka-Qwen-7B, a model specifically trained on math data, ...

The computer, dubbed the "Gigafactory of Compute," is scheduled to be completed by the fall of 2025 and is projected to be at least four times larger than the most powerful clusters currently used by ...

The results also highlight some current limitations. Grok seems to have trouble accessing many mainstream news outlets, instead relying heavily on X sources. Image: Screenshot of THE DECODER via X ...

The neuroscientist Jean-Rémi King leads the Brain & AI team in Meta’s AI division. In an interview with The Decoder, he discusses the connection between AI and neuroscience, the challenges of ...

In an interview, OpenAI CEO Sam Altman emphasizes the importance of using high-quality data to train AI models, whether it is human-generated or synthetic. OpenAI is experimenting with generating ...

Four AI teams crack nearly every challenge In the first competition, dubbed AI vs. Humans, six AI teams faced off against roughly 150 human teams. Over a 48-hour window, everyone had to solve 20 tasks ...