Why we joined Scale Synthetic

by Vivek Muppalla and Joel Kronander on February 7th, 2022

Why we joined Scale Synthetic cover

Q&A with Scale's Synthetic team

Why did you join Scale?

Vivek: Choosing a new role is a consequential and high-impact decision. I have been following the fantastic work by the Scale team for a few years now. I was intrigued by their ambition and impact in a relatively short window. Eventually, my brain caught up to my gut. Once I spoke to the team at Scale, joining turned out to be a relatively simple decision.

Joel: I have spent the last decade working at the intersection of computer graphics and computer vision. I have never been more excited about the future. Never before have so many profound technological breakthroughs converged at the same time - unprecedented compute power, large amounts of visual data, powerful generative models, and scalable AI infrastructure. Together, these advancements enable many new AI applications that we could only dream of in the past. I am incredibly excited to join Scale to help accelerate that future.

What aspects of AI and ML are you excited to work on at Scale?

Vivek: We are still very early in the technology arc of transitioning from traditional software systems to AI. Building infrastructure to get there quickly and responsibly has fueled my passion for several years. I believe Scale will play a significant role in shaping that future.

Although we have seen significant investment in model architectures over the past decade, we need a fundamental shift in data practices to unleash AI’s full power. Model code and data form the foundation of an AI system. Scale’s leading end-to-end solutions for the ML lifecycle based on real-world data will continue to set the bar for the data-centric movement.

Joel: When I completed my PhD in 2015, the deep learning revolution was just getting going. My thesis focused on physically based rendering of synthetic objects in real environments. Fast forward to 2022, and deep learning has fundamentally changed how most visual computing problems are approached.

The practical success of these models is often attributable to their capacity to scale to large amounts of data - a fact that I came to appreciate first hand during my time at Apple.

What are the biggest challenges to implementing synthetic data?

Vivek: The physical world continues to be a limitation for data collection. Enter: digital twins. Digital twins are virtual clones that recreate the real world. This synthetic world will give developers access to unlimited data quickly. But, more importantly, it will take us a step further in addressing the most pressing ethical issues in the AI world today - privacy, data bias, and the democratization of data. Scale Synthetic is our next addition towards accelerating the AI future.

Joel: Most recently as Head of Machine Learning at Nines, I came to appreciate that often the fastest way to improve the performance of modern ML models is not to tweak the model to work better for less data, but to focus one’s efforts on understanding, collecting and labeling more high quality data.

This insight also presents a challenge to applications where data is scarce, or difficult to collect or label. Instances of this include when there are privacy considerations, dataset bias, or rare events. Hence, the need for synthetic data.

In your view, what sets Scale apart from other AI and ML companies?

Vivek: Scale’s credos resonate with my core beliefs as tenets for successful teams.

Two credos that stand out for me are “ambition shapes reality” and “results speak loudest”. Raising the bar, and regularly and objectively measuring progress, is foundational to great products. Once I spent time with the team, it was evident that the credos were not symbolic, but formed the day-to-day backbone.

Building, and scaling a culture alongside a product is tough. However, reading Alex and Brad’s posts gave me confidence that I am joining a company where the founder and exec team cares immensely about culture. Solid culture is a necessary condition to build a generational company–one which I hope to contribute to and shape with the team.

Joel: The data-centric approach is also what got me excited about Scale. Scale enables any ML developer to build better AI models, faster, by building solutions to manage the entire ML lifecycle. Scale started with high quality data annotation, and is now a clear market leader. Scale has since continued to innovate and has built an impressive suite of leading products around AI operations and infrastructure