# Pacing the AI frontier, IBM Granite 4.2 & Meta’s Muse assistant

## Executive summary

This episode provides a deep dive into the current state of frontier AI, covering the debate around slowing development (pacing), the technical specifications of IBM's Granite 4.2 models, and Meta's push into personal agents with Muse. Key technical takeaways include the focus on smaller, auditable models, the use of synthetic data for training, and the critical need for robust sandboxing and guardrails for agentic workflows.

## Key takeaways

- Pacing AI Development: Concerns regarding existential risk and the ability of agents to find 'day zero vulnerabilities' have led to calls for slowing AI development. Experts suggest that the focus should be on implementing guardrails for agent-to-agent communication and limiting concurrent agent calls to mitigate economic risks.
- IBM Granite 4.2 Release: The new Granite models (3B, 8B, 30B) are designed for enterprise use, featuring native step-by-step reasoning and support for agentic workflows like planning and tool calling. The models are available on Apache 2.0 license via Hugging Face.
- Meta's Muse Agent: Muse is Meta's personal AI agent, designed to run on a secure virtual machine (VM) for isolation. While technically advanced in its design, the discussion highlights ongoing concerns regarding user privacy and the security risks of handing over personal data to such agents.

## Technical details

- Granite 4.2 Architecture: The models come in three sizes (3B, 8B, 30B) and natively integrate step-by-step reasoning to support complex agentic tasks like planning and tool calling. The associated Granite Speech 5.0 ASR model is noted for its speed (transcribing 3 hours of audio in seconds).
- Model Training Innovations: IBM advocates for 'mid-training,' an intermediate process between pre-training and post-training, which improves reasoning capabilities. The use of synthetic code (e.g., Code Alchemy) is highlighted as a method to generate vast, controlled amounts of high-quality data, addressing the need for data diversity and scale.
- Agent Security and Sandboxing: Security concerns center on agents escaping sandboxes and the difficulty of forensics after an incident. Best practices discussed include using secure VMs, implementing observability, and ensuring frameworks have 'kill switches' and strict access controls (e.g., limiting internet access).

## Practical implications

- Build engineers should prioritize designing agentic systems with strict sandboxing and isolation mechanisms (e.g., secure VMs) to contain potential breaches.
- When developing models, consider incorporating 'mid-training' steps and leveraging synthetic data generation (like Code Alchemy) to enhance reasoning and domain-specific capabilities.
- For enterprise deployments, favor smaller, auditable models (below the frontier) over massive, opaque models to maintain control and compliance.

## Topics

AI Agents, LLM Development, Model Security, Enterprise AI, Generative AI, IBM Granite 4.2, Hugging Face, Mixture of Experts podcast

Source: https://www.youtube.com/watch?v=DtKEgRuq_00