Key insights from UphillConf 2024

We were at the UphillConf 2024 - here are our takeaways:

.
The event featured insightful talks and workshops about the transformative power of ML and AI, with an emphasis on practical applications and real-world challenges.

Miquel Àngel Farré: ML Lifecycle: Annotation to Deployment, and Pragmatic Decision-Making
- Datasets are the most important part of any ML project
- When labeling datasets, don’t shy away from using human-in-the-loop annotators to correctly identify edge cases.
- Always opt for the simplest possible solution.

Michael Gygli, Elle O'Brien and David Berenbaum: ML Ops Behind the Scenes
- Tailor ML Ops to specific use cases. Evaluations and deployments don’t look the same for a fraud detection application as for a logo creation service.

Miquel Àngel Farré, Michael Gygli: LLMs in the Real World - What’s Left After the Dust Settles
- Current LLMs lack human-like motivations, limiting AGI potential.
- Despite fewer entry-level roles, LLMs can’t replace the unique perspectives to problem-solving a new intern brings.

🤷 Nico Martin: From ML to LLM: On-Device AI in the Browser
Technologies like WebGPU enable local LLMs in browsers, enabling new web application use cases, specifically focused on privacy.

Building AI Applications: Successes and Lessons from Real-World Projects
A key aspect of building AI applications is to manage user expectations. Don’t build a chatbot that can answer “anything”.

Lewis Tunstall: Building Machine Learning Applications using Hugging Face
Tools like HuggingFace transformers and datasets enable the incredible velocity we see in today’s AI research landscape.

Happy to have sponsored the UphillConf!