Researchers and enterprise teams are leveraging large language models (LLMs) for a variety of innovative applications, but future improvements will rely on more than just parameters. Check out our roundup of the top AI and MLOps articles for May 2023!
OpenAI’s CEO Says the Age of Giant AI Models Is Already Over
We've hit diminishing returns on LLM size. Future models will rely on architecture improvements or fine tuning rather than more parameters: https://www.wired.com/story/openai-ceo-sam-altman-the-age-of-giant-ai-models-is-already-over
Emergent autonomous scientific research capabilities of LLMs
Carnegie Mellon University chemists have demonstrated LLMs can perform autonomous research. The AI agents synthesized drugs like ibuprofen and aspirin from simple prompts. But without guardrails what would prevent nefarious usage of these novel models? https://arxiv.org/abs/2304.05332
91% of models degrade in performance
Researchers used 32 datasets and 4 model types to run ~2.5 million experiments on model performance. Their results are eye-opening: https://www.nature.com/articles/s41598-022-15245-z
Improve model performance with flexible charts and dashboards
Custom charts and rich dashboards help MLOps teams measure and improve model performance with deeper insights: https://www.fiddler.ai/blog/supercharge-model-performance-with-flexible-charts-and-dashboards
What an MLOps engineer actually does
Interested in becoming an MLOps engineer? Mikiko Bazeley has put together a great guide on what to expect and how to get there: https://medium.com/kitchen-sink-data-science/what-an-mlops-engineer-does-565d4d0adb2b
How Meta measures the management of its AI ecosystem
How does Meta manage their thousands of ML models to handle model governance, security, accountability, AI fairness, model robustness, and efficiency? Here's a comprehensive overview: https://ai.facebook.com/blog/meta-ai-ecosystem-management-metrics
Stay up to date on everything AI and MLOps by signing up for our newsletter below.