Large Language Models

Today's Stories

1. 1

   Palantir partners with Google Cloud to expand business AI tools

   Palantir Technologies announced a multi-year partnership with Google Cloud on June 7, making its data analysis platform available through Google's marketplace. The deal includes integration with Google's BigQuery database service and Gemini AI models. Palantir also showcased new business deployments with law firm Kirkland & Ellis, Mexican insurer GNP Seguros, and construction company McCarthy Building.

   Businesses using Google Cloud will now have easier access to Palantir's data analysis tools, potentially making it simpler for companies to implement AI-powered decision making in their operations.

2. 2

   ChatGPT adds security mode to prevent data theft from AI conversations

   OpenAI released a new "Lockdown Mode" for ChatGPT on June 7 that disables web access, research features, and agent capabilities to protect sensitive data. This addresses prompt injection attacks (tricks that make AI systems leak private information), though the company acknowledges it doesn't fully solve the problem.

   ChatGPT users working with confidential business information can now enable this mode to reduce the risk of accidentally sharing sensitive data through their AI conversations.

3. 3

   Researchers discover why small AI models fail at uncommon tasks

   A new study published June 7 found that smaller language models (AI systems with fewer parameters) struggle with rare tasks because frequent tasks constantly overwrite what they've learned. Researchers tested models ranging from 4 million to 4 billion parameters and found that increasing how often rare tasks appear in training data can be more effective than just making models bigger.

   This could lead to more efficient AI training methods that don't require massive computing power, potentially making advanced AI capabilities more accessible to smaller companies.

4. 4

   NVIDIA unveils new AI capabilities for robots and self-driving cars

   NVIDIA introduced new physical AI agent skills at the CVPR 2026 conference on June 3, designed to accelerate development of autonomous vehicles, robotics, and vision AI. The company integrated these capabilities with its NVIDIA Cosmos 3 platform to address fragmentation in current AI research.

   This could speed up the development of practical robots and self-driving cars by giving developers better tools to train AI systems for physical world tasks.