New Wisdom Report
Editor: Qiao Yang Yongyong
[New Wisdom Summary] Scale AI, founded by Alexandr Wang, is a data annotation platform that provides training data for AI models. Recently, it completed a new round of $1 billion in financing, skyrocketing its valuation to $13.8 billion. The company stated that it will use the new funds to produce rich cutting-edge data to pave the way towards AGI.
Scale AI offers data annotation services for companies looking to train machine learning models and has raised $1 billion in Series F financing from many well-known institutions and investors such as Amazon and Meta.
This round of financing was led by Accel, which previously led Scale AI's Series A financing and participated in subsequent venture capital investments.
This financing round has propelled Scale AI's value. Despite a 20% workforce reduction early last year, the company's current valuation has reached $13.8 billion.
Scale AI co-founder and CEO Alexandr Wang
In addition to Amazon and Meta, Scale AI has attracted a variety of new investors: Cisco, Intel, AMD, and other venture capital departments have participated, and many previously invested companies have also returned, including Nvidia, Coatue, Y Combinator, and more.
Teenage prodigy drops out to found unicorn
Founded in 2016 by Alexandr Wang and Lucy Guo, Scale AI is backed by the renowned startup incubator Y Combinator. The company uses machine learning to annotate and categorize large amounts of data for customers to use in training models.
Scale AI's clients include Meta, Microsoft, Nvidia, OpenAI, Toyota, and Harvard Medical School.
After a $100 million Series C funding round led by Founders Fund, Scale AI achieved unicorn status in 2019, raising a total of $602.6 million from notable investors such as Index Ventures, Coatue, and Tiger Global.
In 2022, Alexandr Wang, holding a 15% stake, became the world's youngest self-made billionaire.
Before starting his entrepreneurial journey, Wang's impressive background from childhood to adulthood stands out.
Born in New Mexico in 1997, both of his parents are physicists at the Los Alamos National Laboratory in New Mexico.
During high school, he began self-learning programming online and participated in world-class programming competitions such as the United States of America Computing Olympiad (USACO).
At 17, he became a full-time coder at the renowned Q&A website Quora.
At 18, he was admitted to the Massachusetts Institute of Technology to study machine learning.
During the summer after his freshman year at MIT, he co-founded Scale with Guo and secured investment from Y Combinator.
Nonchalantly, Wang told his parents, "This is just something I'm casually doing over the summer."
Initially, some people thought Scale AI was a joke as the company only had three employees at the time.
However, with continuous funding and growth, Scale AI rapidly expanded, becoming a unicorn valued at $7.3 billion by 2021, with a workforce of 700 by early 2023.
In an exclusive interview with Fortune magazine, Wang revealed that as enterprise clients increasingly train generative AI models, Scale AI's business in this area is rapidly growing.
In 2023, the company's annual recurring revenue (fees paid by enterprises for data services) doubled, with projections to reach $1.4 billion by the end of 2024.
Due to the remarkable achievements of Scale AI, Alexandr Wang was selected for the Forbes "30 under 30" list in the enterprise technology sector in 2021, and he is also known as the "next Zuckerberg" in Silicon Valley.
The "Data Factory" of AI
The three fundamental pillars recognized in the AI field are data, algorithms, and computing power.
In the realm of algorithms, there are large research institutes like Google and Microsoft, and entities like OpenAI that have introduced models such as Sora and the GPT series; in the computing power domain, there is the global supplier NVIDIA. However, back in 2016 before Scale AI was born, the data domain was still largely unexplored.
At the age of 19, Alexandr Wang saw this gap and made the decision to drop out of school and start his own business, stating, "I founded Scale to address the data problem in artificial intelligence."
Most data is unstructured, making it difficult for AI to learn directly from it. Moreover, annotating large datasets is a resource-intensive task, leading many to consider "data" as the most arduous and humble part of the tech field.
Yet, Scale AI achieved tremendous success in a short period. They offer tailored data services to enterprise clients across various industries.
In the field of autonomous driving, companies like Cruise and Waymo collect vast amounts of data through cameras and sensors. Scale AI combines machine learning with human-in-the-loop supervision to manage and annotate this data.
Their development of the "Autonomous Data Engine" has propelled the advancement of L4-level autonomous driving technology.
In 2019, Scale AI assisted the OpenAI team in training GPT-2 together, conducting the first RLHF experiment, and extending these techniques to other LLMs such as InstructGPT.
CEO Wang, in an interview with Fortune magazine, positioned Scale AI as an infrastructure provider for the entire AI ecosystem, building a "data foundry," rather than just hiring a large number of contract workers for manual annotation in its subsidiary, Remotasks.
Scale AI has started collaborating with experts from various fields, such as doctoral scholars, lawyers, accountants, and writers.
Why would doctoral-level experts participate in rating responses to chatbots?
Wang's response is that there are many reasons: "If you are a doctoral student accustomed to conducting very niche, esoteric research that perhaps only a few people in the world can understand, in this work, you can help improve and build cutting-edge data for these artificial intelligence systems, with the opportunity to make a real societal impact."
Furthermore, Wang believes that the high-quality data these experts can provide is crucial for the future of AI.
He added that data from experts with complex reasoning is essential for the future of artificial intelligence. "You can't just feed old data into an algorithm and expect it to improve on its own."
Traditional data sources, such as scraping data from comments on platforms like Reddit, have limitations. Scale AI has processes in place where models first output content, such as drafting research papers, which human experts can then refine to improve the model's output.
Regarding AI-generated and annotated data, some are optimistic that it can eliminate the need for human annotation, but Wang's view is not so straightforward.
He mentioned that Scale AI invests in both synthetic data and human-created data. "While AI-generated data is crucial, the only way to obtain data of a certain quality and accuracy is through validation by human experts."
The Increasing Importance of Data
Data is the lifeblood of artificial intelligence, so companies in the data management and processing field are currently at the forefront.
Just last week, the Indian data platform Weka announced it raised $140 million at a post-money valuation of $1.6 billion to help companies build data pipelines for their AI applications.
The main issue with artificial intelligence data still persists. The existence of the Scaling Law means that as models grow larger, the demand for data also increases exponentially, leading to concerns that large models may deplete available data.
Alexandr Wang wrote on Scale AI's official website, "Data richness is not the default, but a choice that requires bringing together the best talent in engineering, operations, and AI."
One of Scale AI's visions is "data richness," aiming to expand cutting-edge LLMs to larger scales, paving the way towards AGI. "On the path to GPT-10, we should not be constrained by data."
References:
https://fortune.com/2024/05/21/scale-ai-funding-valuation-ceo-alexandr-wang-profitability/
