Automation is faster and cheaper — but comes with downsides, too.
By Jerome Pasquero , Sama’s Director of Machine Learning
I recently took a moment to reflect on the crucial role of human annotations in today’s fast-paced world of model development. While automation is often mentioned as a quick fix, it's important to consider not just where it works, but also where it falls short, and why human expertise remains essential.
I took the ideas to Claudel Rheault, and she and I discussed it so much that we wrote a white paper about it. This was a fun project to dive into, offering a bit of an opinion piece from two people who are deeply immersed in the human-driven process of generating training data at Sama.
TL;DR: We’re not done downloading all human skills into training data just yet. Automation is just one tool among many that can help accelerate model development.
In our industry, where we create training data for cutting-edge AI models, there’s a habit of showcasing flashy demos that make AI automation - using existing models to generate annotations - seem like the ultimate solution. But that’s misleading. If those models were truly adding value on their own without any human intervention, clients would be more interested in the models themselves, not just the data they produce.?
I hope you take a moment to look at the full paper (about a 12-min read), available on Sama’s website here.
What do you think? Leave a comment with your thoughts.
Sama recognized as Best AI Model Validation Solution in 2024 AI Breakthrough Awards
What we’re adding to our Web Summit 2024 agendas
We’re Lisbon-bound in November, and eager to meet up with other AI professionals. Connect with Mark Bakker and Patricia Caruso ahead of the event! Here are the sessions they’re most excited to attend. DM them to meet up.
Keep reading, listening, learning:
??? New How AI Happens Podcast Episode
Intel’s VP & GM of Strategy & Execution Melissa Evers shares the factors that determine whether an open-source community is healthy, and speaks to Intel’s philosophy on innovation and open AI.
Plus:
?? Free Guide: Optimizing Your Model with Supervised Fine-Tuning?
?? Sama Spotlight: Faces of Sama: Pius Nashimala's Poetry of Perseverance?
??? In the news: United States and Eight Companies Launch the Partnership for Global Inclusivity on AI
Image credit: Yutong Liu & Kingston School of Art ?/ Better Images of AI / Talking to AI 2.0 / Liscenced by CC-BY 4.0