Hello readers, in this issue we cover

• Meta introduces LLama Guard 3 Vision to moderate content with image understanding

• AWS Releases Multi Agent Orchestrator, a framework for managing AI agents

• A new method to detect fake news by South Korean researchers

• How Reinforcement Learning is Used in Finance

🦙 Meta introduces LLama Guard 3 Vision to moderate content with image understanding

Llama Guard 3 Vision is a multimodal safeguard that enhances content moderation in conversations requiring image understanding. Unlike earlier text-only versions, this tool supports image reasoning, detecting harmful multimodal prompts and their responses. Fine-tuned on Llama 3.2-Vision, it excels in internal benchmarks using the MLCommons taxonomy and has been tested for robustness against adversarial attacks. It aims to advance robust content moderation tools for human-AI interactions with multimodal capabilities.

Read the research paper

🤖 AWS Releases Multi Agent Orchestrator, a framework for managing AI agents

The Multi-Agent Orchestrator is a flexible framework for managing multiple AI agents and handling complex conversations. It intelligently routes queries and maintains context across interactions.

The system offers pre-built components for quick deployment, while also allowing easy integration of custom agents and conversation messages storage solutions.

This adaptability makes it suitable for a wide range of applications, from simple chatbots to sophisticated AI systems, accommodating diverse requirements and scaling efficiently.

Read the research paper

📰 A new method to detect fake news by South Korean researchers

This study highlights the limitations of conventional fake news detection methods, which rely on future knowledge during training, making them less applicable to real-world scenarios. To address this, the researchers introduce DAWN, a novel method designed for temporality-aware settings where models are trained only on data available up to a certain point in time. They find that later engagements, such as news reposting, introduce noisy connections between real and fake news in the social graph. DAWN mitigates this by leveraging engagement earliness features to adjust edge weights, reducing noise and improving detection accuracy. Extensive experiments show DAWN outperforms existing methods in realistic environments.

Read the research paper

📈 How Reinforcement Learning is Used in Finance

This survey examines the growing use of Reinforcement Learning in financial applications, highlighting its potential for solving decision-making tasks. It dives into applications in market making, portfolio management, and optimal order execution. The survey also reviews recent advancements addressing these issues and suggests future research directions, such as benchmarking, contextual RL, multi-agent RL, and model-based RL, to enhance RL's implementation in the financial industry.

Read the research paper

🤯 Today I Learned

Every issue, we highlight new AI concepts and terminology to help educate our readers. This issue we learned about:

Future knowledge

Future knowledge refers to information that would only be available after a certain point in time but is used during the training or evaluation of a model. In the context of machine learning, this can occur when a model has access to data or outcomes from a future time period that wouldn’t realistically be available in a real-world deployment scenario.

For example:

• In fake news detection, using social media interactions (likes, shares, or comments) that happen after the news article's publication during the training process is considered future knowledge. This data wouldn't be available when the model is used to make real-time predictions.

• In forecasting models, using actual sales data for future periods while training would also be an example of incorporating future knowledge.

Using future knowledge creates an unrealistic evaluation setting because it gives the model an advantage it wouldn’t have when deployed in a real-world, time-constrained environment.