Welcome back to Daily Zaps, your regularly-scheduled dose of AI news ⚡️ 

Here’s what we got for ya today:

• 🎮 “Computer-using agent” the next trend in AI

• 🗒️ NotebookLlama: an open version of Google’s NotebookLM

• 🙊 Whisper AI’s hallucinations problem

• 🤖 DoD uses AI to spot intruders

Let’s get right into it!

Sign Up | Advertise | Tools

BIG TECH

‘Computer-using agent’ AI system

Google may soon unveil an AI agent called Project Jarvis, designed to operate a web browser and automate everyday tasks like research, shopping, and booking flights. Set for a possible December preview, Jarvis captures screenshots of a user's computer screen, interprets them, and performs actions like clicking buttons or typing in fields.

It works specifically with web browsers, particularly Chrome. This development coincides with Google's expanding Gemini AI, which recently gained new language support and integration with apps like Google Meet and Photos. Jarvis follows Anthropic’s similar AI, Claude, which can already use standard software tools.

BIG TECH

NotebookLlama: an open version of Google’s NotebookLM

Meta has introduced NotebookLlama, an open-source alternative to Google’s NotebookLM, designed for interactive data analysis and documentation. NotebookLlama integrates large language models into a notebook interface like Jupyter, allowing users to interact with the AI for tasks like code writing and documentation.

Powered by optimized Llama models, it offers customizable solutions and can be deployed on local or cloud servers, making AI-driven tools more accessible. With a focus on openness and flexibility, it provides full control over data and models, outperforming some proprietary tools in early testing. This release fosters community-driven innovation and democratizes access to AI-powered development.

HEALTHCARE

Whisper AI’s hallucinations problem

OpenAI’s transcription tool, Whisper, lauded for its accuracy, has a significant issue: it can fabricate entire sentences, a phenomenon known as “hallucinations.” These errors range from fabricated racial commentary and violent language to non-existent medical treatments. Researchers report frequent hallucinations, with one study finding them in 80% of cases examined.

Concerns about Whisper’s flaws have led experts to advocate for stricter AI regulations, as these inaccuracies could have severe consequences, especially for vulnerable users like the Deaf and in high-stakes environments like healthcare.

GOVERNMENT

DoD uses AI to spot intruders

The Department of Defense (DOD) recently tested a new AI security tool, Scylla, at the Blue Grass Army Depot in Kentucky. Scylla uses video feeds and drones to spot intruders, weapons, and unusual behavior in real time with over 96% accuracy, reducing false alarms and lessening the burden on security staff. This AI system can identify specific faces and detect suspicious actions, instantly alerting security teams to potential threats.

Developed by the Physical Security Enterprise and Analysis Group (PSEAG), Scylla aims to protect critical U.S. assets, including nuclear facilities, with plans to expand its use in cold and marine environments soon. This project supports the National Defense Strategy’s focus on AI, helping the U.S. stay ahead in security technology. The DOD believes that Scylla is a foundational tool for future security innovations.