This feature really does change things. Gemini had cache, but only for costs — Claude claims 90% less cost *and* 85% less latency for cached tokens. Even if it can’t replace all SFT, it’s so much easier to iterate on you probably want to exhaust caching-based strategies first.
SYSTEMS
-
RAG Systems Advances: Multi-modal Support and Production Monitoring
By
–
In the past year, RAG systems have advanced significantly. It now features: •Multi-modal documents and responses
•LLM routing based on query complexity
•Multi-lingual support
•Improved accuracy and reduced hallucinations
•Feedback and monitoring in production. A recent -
Automating System Recovery and Failure Prevention Strategies
By
–
I don't see this as any different to any other breakage of their machine. I'd seek to understand how they broke it, help them avoid doing that again in the future, and ensure they had a well-automated system to rebuilding any part of their machine they had a habit of breaking.
-
Cloud Infrastructure Scaling Limitations and Fixed Instance Requirements
By
–
No it doesn't scale up smoothly – requires fixed instance sizes.
-
Understanding AI System Cards: Documentation for Model Transparency
By
–
Una "model card" o "system card" es una documentación que suelen acompañar a los nuevos modelos o sistemas de IA para dar una descripción detallada de muchos aspectos relacionados con su funcionamiento, ética en el entrenamiento, seguridad, etc. https://
ai.meta.com/blog/system-ca
rds-a-new-resource-for-understanding-how-ai-systems-work/
… -
Ambiguity in Compute Execution Calculation Methods
By
–
They don't specify how they calculate when the compute "runs"
-
Scaling Limitations Fixed Instance Sizes Infrastructure Challenges
By
–
No it doesn't scale up smoothly – requires fixed instance sizes.
-
Scaling ChatGPT: Behind the Scenes Infrastructure
By
–
Behind the scenes scaling ChatGPT – Evan Morikawa at LeadDev West Coast 2023 https://
bit.ly/45WoUap
#AI #MachineLearning #DeepLearning #LLMs #DataScience