EmbeddingGemma 2: an open, lightweight multimodal embedding model
deepmind.google·6d ago
AI Engineering
Hacker News·Aug 19·5 min read
Ornith-1.5: From Self-Scaffolding to Self-Improvement
TL;DR
Ornith-1.5 introduces a self-improvement framework that allows models to autonomously generate tasks and learning experiences, achieving state-of-the-art performance across various benchmarks. It offers scalable options, from a 397B parameter model to a compact 9B version suitable for mobile deployment.
✦ Why It Matters
Explore deploying Ornith-1.5-9B for mobile applications to enhance user experiences today.
Key Takeaways
- Consider using Ornith-1.5 for advanced AI tasks in your projects.
- Evaluate the 9B model for mobile applications to enhance user experience.
- Leverage the self-improvement framework to automate task generation.
- Benchmark your models against Ornith-1.5 to assess performance.
How It Works
Ornith-1.5 operates through a self-improvement loop that includes three stages: task generation, scaffold construction, and solution rollout. The model autonomously proposes new tasks that challenge its current capabilities, generates specific scaffolds to approach these tasks, and produces solution rollouts that are evaluated for effectiveness.
Rewards from successful rollouts inform future task generation, creating a dynamic learning environment that evolves with the model's capabilities.