NEWS IN BRIEF: AI/ML FRESH UPDATES

Get your daily dose of global tech news and stay ahead in the industry! Read more about AI trends and breakthroughs from around the world

Shell Commands: Persisting Session State

AI agents have advanced beyond chat, now able to write code and execute shell commands. Amazon Bedrock AgentCore Runtime offers persistent filesystem state and direct shell command execution, revolutionizing agent workflows.

Opting Out: The Case Against Auto-Stopping in SGD Training

Summary: Auto-stopping in stochastic gradient descent training can be challenging due to setting stopping tolerance and consecutive-no-improve constants, impacting model accuracy and overfitting risk. Specifying the right stopping tolerance in SGD depends on data scaling, affecting model prediction accuracy and potential overfitting.

Domain Control: Managing AI Access

Amazon Bedrock AgentCore enables AI agents to interact with the web securely through managed tools like Browser and Code Interpreter. AWS Network Firewall provides domain-based filtering to control access, addressing security concerns and compliance requirements for regulated industries deploying AI agents.

AI Impact on Pupil Thinking Skills

Two-thirds of secondary school teachers report decline in core abilities like writing and problem-solving. AI use in pupils linked to loss of critical thinking skills, teachers say.

Ethical Evaluation of Autonomous Systems

MIT researchers developed an automated evaluation method using AI to balance cost and fairness in decision-making. The system streamlines ethical testing to align autonomous systems with human values, predicting potential ethical dilemmas before deployment.

Accelerating Gemma 4: NVIDIA's Local Agentic AI Journey

Google's Gemma 4 family introduces compact models optimized for NVIDIA GPUs, supporting a wide range of tasks. These models enable efficient local execution for on-device AI, from edge devices to high-performance GPUs, revolutionizing everyday devices.

AI Love Gone Wrong

AI skeptic Rhik Samadder lets AI take the lead on a date in week five of his diary. Is AI the missing ingredient in his search for love?

Anthropic's AI Tool: Claude's Code Revealed

Internal files leaked from Anthropic due to ‘human error’, including source code for Claude Code, leading to rapid spread on GitHub with blueprints for new AI products. Anthropic issues copyright takedowns to control dissemination of leaked code, attracting millions of views.