Large Language Models

Today's Stories

1. 1

   AMD challenges NVIDIA with cheaper AI computer for running text-generating AI

   AMD released the Ryzen AI Halo computer for $3,999, undercutting NVIDIA's similar $4,679 DGX Spark by $680. The machine comes with 128GB of memory designed to run large language models (AI that understands and writes text) quickly without needing cloud services.

   Companies wanting to run ChatGPT-style AI privately on their own computers now have a more affordable option that doesn't send data to external servers.

2. 2

   Mastercard enables AI agents to make payments using cryptocurrency platforms

   Mastercard partnered with Coinbase and Ripple to let AI agents automatically make payments on behalf of users. The system allows AI assistants to complete transactions without human intervention when programmed to do so.

   AI assistants could eventually pay bills, buy products, or transfer money automatically, though this raises questions about oversight and security.

3. 3

   Google research shows AI safety filters fail more often than expected

   Google DeepMind researchers found that training filters designed to block harmful AI responses work surprisingly poorly. Their study of the Gemini AI model revealed that safety measures often transfer unexpected behaviors from training data rather than preventing problems.

   This suggests current methods for making AI safe may be less reliable than companies claim, potentially affecting how AI assistants respond to sensitive requests.

4. 4

   Bank of America doubles down on AI agent investments with stock upgrade

   Bank of America upgraded its investment rating for companies developing agentic AI (AI that can take actions independently rather than just answering questions). The bank sees this as a major growth opportunity in the AI sector.

   Investment money is flowing toward AI that can perform tasks automatically, which could accelerate the development of AI assistants that handle real work.

5. 5

   Claude Code users report approval fatigue with AI programming assistant

   Users of Anthropic's Claude Code programming tool say they're overwhelmed by constant approval requests for code changes. Many report clicking 'yes' automatically without understanding what they're approving, especially for database changes and system modifications.

   This highlights a key challenge as AI tools become more powerful - people may approve changes they don't understand, potentially causing problems in software and systems.