the-decoder

Anthropic claims context engineering beats prompt engineering when managing AI agents

Anthropic is looking to move beyond prompt engineering with a new approach it calls "context engineering." The idea is to help AI agents use their limited attention more efficiently and maintain ...
the-decoder

Study cautions that monitoring chains of thought soon may no longer ensure genuine AI alignment

A new joint study from OpenAI and Apollo Research examines "scheming" - cases where an AI covertly pursues hidden goals not intended by its developers. The researchers tested new training methods to ...
the-decoder

Deepseek outputs weaker code on Falun Gong, Tibet, and Taiwan queries

A CrowdStrike study found that Chinese AI system Deepseek delivers less secure code when prompts involve politically sensitive topics, raising concerns about political bias in technical outputs.
the-decoder

Apple introduces Manzano, a model for both image understanding and generation

Apple is working on Manzano, a new image model designed to handle both image understanding and image generation. This dual capability is a technical hurdle that has kept most open-source models a step ...
the-decoder

AI artist Xania Monet signs $3 million record deal with Hallwood Media

AI artist Xania Monet has signed a $3 million record deal with Hallwood Media, according to Billboard. The project is led by 31-year-old designer Telisha Jones from Mississippi, who uses the Suno ...
the-decoder

Study claims 78 training examples are enough to build autonomous agents

A new study challenges a core assumption in AI: instead of massive datasets, just 78 carefully chosen training examples may be enough to build superior autonomous agents. Researchers from several ...
the-decoder

ChatGPT's Deep Research mode let attackers steal Gmail data with hidden instructions in emails

Security researchers at Radware have uncovered a serious flaw in ChatGPT's "Deep Research" mode that allows attackers to quietly steal sensitive data such as names and addresses from Gmail accounts, ...
the-decoder

Microsoft's new Large Action Model can perform some tasks in Word

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 ...
the-decoder

Salesforce's CRM benchmark finds AI agents struggle in real-world business scenarios

Salesforce's new CRMArena-Pro benchmark reveals major challenges for AI agents in business contexts. Even top models like Gemini 2.5 Pro manage just a 58 percent success rate on single turns. When the ...
the-decoder

Musk unveils Grok 4 as xAI’s new AI model that beats OpenAI and Google on major benchmarks

Amid backlash over antisemitic content and major leadership shakeups, Elon Musk has introduced Grok 4, the latest flagship AI model from xAI. The launch comes during a period of upheaval for Musk's ...
the-decoder

OpenAI's new multimodal "GPT-4 omni" combines text, vision, and audio in a single model

OpenAI has unveiled GPT-4o, its latest large multimodal model that sets new standards in performance and efficiency by combining text, image, and audio processing in a single neural network. The "o" ...
the-decoder

Adobe Photoshop integrates Google's image AI "Nano Banana"

Adobe is bringing Google's new image AI, known as "Nano Banana" (officially Gemini 2.5 Flash Image), to Photoshop as an optional tool. The model is designed for editing existing images with a high ...