How Scale AI works

based in San Francisco California scale AI has raised over $600 million from Top VCS like Excel Founders fund index Ventures and has a valuation of over $7 billion this valuation made its co-founder and CEO Alexander Wang the youngest self-made billionaire at 24 years old back in 2021 today's AI technology doesn't really think for itself rather it produces outputs and an Based on data it's been trained on this means that it's all about the training model and the data that's fed into it this is where a company like scale AI comes in scale is a data labeling and annotation platform that helps companies develop Ai and machine learning models it recently garnered a lot of attention and is seen as one of the hottest Tech startups today but what exactly does scale do in this episode we'll talk about how scale AI actually works the Shady dark side of the company and what the future holds for AI here we go to understand scale AI we have to wind the clocks back to 2016 when the company was founded scale is a YC company that was part of its spring 2016 batch and was started by Lucy guo and Alexander Wang both Geniuses in their own rights Lucy was 21 years old at the time and had already worked at meta which was Facebook back then Kora and Snapchat before starting the company she was a Carnegie melon University Dropout and a teal fellow for those of you that don't know the teal Fellowship is a prestigious 2-year program that provides $100,000 to young people who want to build new things instead of sitting in a classroom to receive the fellowship students must be younger than 22 years old and must drop out of college it was started by Tech billionaire Peter teal and is extremely competitive with an acceptance rate of just .1% about 20 to 25 fellows are selected annually through a competitive annual process okay back to scale Alexander Wang was 19 at the time and dropped out of MIT to start the company what's crazy is that while he was just a high schooler he was already a tech lead at Kora working on the infrastructure team Kora back then was known for having top-notch product design and Engineering teams it was here where Lucy and Alex met and began their entrepreneurial journey together but it would soon end a few years later when Lucy decided to leave the company more on this in a moment many people don't know this but scale actually didn't start off as a data platform for AI or at least not directly it was a simple API for human tasks in fact its name and domain main name weren't even scale AI it was scale API for those of you that don't know API stands for application programming interface it's a type of software that allows two or more computer programs to communicate with each other for example the weather app on your phone communicates with the weather bureau's Software System to be able to show you weather updates on your phone scale provided an ond demand Fleet of human laborers to perform tasks that couldn't be done by algorithm all accessible by one line of code companies could get tasks stunned by these workers for cases like content moderation data extraction and scheduling appointments as scal signed more clients to its business it quickly found a strong use case in artificial intelligence and became a viable solution to a problem plaguing self-driving car companies these companies had millions of miles of on the road driving footage to train their autonomous vehicle AI but not nearly enough people to review and label it scale could fill that Gap it soon signed big clients in automotive industry like Toyota Honda wh and Cruz the founders Lucy and Alex had many arguments about the direction they wanted to take the company in 2018 Lucy decided to leave the company due to differences in product vision and its road map since then she started her own VC firm back in capital and launched her own software startup called passes focused on the Creator economy both Lucy and Alex have farther declined to speak any more about their split scale dropped API from its domain name and changed its legal business name to scale AI what it does now is provide a data labeling and annotation platform to help companies turn their raw data into high quality training data for the development of their own AI applications scale does this by combining machine learning powered pre-labeling and active tooling with varying levels and types of human review Alex sees his company as the pics and shovels in today's AI Gold Rush scale has four core products the first product is called scale data engine which helps machine learning teams build AI with this teams collect curate and annotate data they train models and evaluate them self-driving car startup nuro uses this to help it identify infrequent but meaningful scenarios with its training data for example its data included hundreds of thousands of images of pedestrians in unusual postures various types of animals and other types of vehicles this is essential for safe autonomy the next two products allow machine learning teams to apply AI to their companies applications we're proud to introduce um our two new generative AI platforms scale Donovan and scale EGP generative AI is truly the tech innovation of this generation that this technology will be as impactful as the internet um it's poised to disrupt fundamental Enterprise business models and industries across Financial Services media Insurance retail and more will'll be investing billions of dollars into this technology to adapt to the changes scale generative AI platform is for Enterprises this enables teams to compare test and deploy Foundation models from companies like open Ai and anthropic they can then fine-tune the base models with their own Enterprise data and scales data engine which allows them to build compare and securely deploy their own applications accelerating their AI developments scale Donovan is for the US government and defense to make smarter decisions Donovan ingests defense data such as emails intelligence reports and satellite imagery it understands and organizes this data to make it interactable which allows operators and analysts to ask it questions in a chat interface that results in a course of action or summary report provided by donam it rapidly analyzed over 2,000 square km of Ukraine identifying over 370,000 structures in which it provided data directly to the government scale's last product is called scale Spellbook which enables developers to build compare and deploy their own large language model apps scale has over 600 employees and works with organizations across various Industries organizations like meta Adept Microsoft instacart Fox Toyota etsc the US Army and the US Air Force its robust data platform and big clientele have propelled it to generate $250 million in Revenue in 2022 at a time when many AI startups weren't yet making a scent all of this sounds great for scale however you don't get to a billion- dollar company without some shady things happening behind the scenes scale faced a critical problem in its business ironically given its name the more it scaled the harder it became to keep up with the demand for human labor the company first turned to Outsourcing agencies to fill gaps but costs quickly spiraled up gross margins which hovered around 65% in early 2018 approached a mere 30% by 2019 this is where remote tasks comes in remote tasks is scale's in-house Outsourcing agency scale set up dozens of facilities and lower cost of living areas like southeast Asia and Africa to train thousands of data labelers in fact scale actually employs over 240,000 labelers this helped the company's margins bounce back to a healthy percentage however these facilities face poor working conditions and many labelers were paid less than $1 an hour scale has been careful to position remote tasks as a separate brand its website makes no mention of remote tasks and vice versa employees say this was to make the company's strategy less obvious to competitors and for client confidentiality because scale offshores its labelers to other countries it faces another problem as the company is increasingly expanding its US Government footprint and focus on defense contracts the government is unlikely to share classified data with foreign labelers so what did it do it had to open up us offices and employ us labelers this obviously is more expensive is this sustainable only time will tell Alex believes that quality data to create reliable AI outcomes requires human insight and guidance algorithms need data and data needs humans only humans can understand the context and Nuance to properly annotate data to be fed to algorithms they teach algorithms what to do this is why teaching AI human intentions and values is so important it's through this process that ensures AI will have Fair ethical outcomes in line with human values it's this actual alignment that's very challenging to solve I agree with his point of view here this is the Genesis for scale but what happens when AI becomes smart enough to think for themselves we're not there yet but it could come sooner than we think this is a very important question and concern that's top of mind for AI experts policy makers and the general public in fact Tech leaders like Elon Musk Bill Gates and even Sam Alman have raised concerns about AI being a top Global threat that should be treated just as seriously as pandemics and nuclear war or risk Humanities Extinction running a business is tough for Alex it seems like it's life or death at least for the United States He deeply believes in two things first is that AI is a huge Force for good and second America needs to continue to be in a leadership position with growing threats from China and Russia it's become very clear that technological advancements is how you come out on top remote tasks was a business decision to cut down on costs and keep his company afloat while I don't necessarily agree with its poor working conditions and very low compensation for its workers I don't disagree with scale's decision to create this agency in the first place scale's exponential growth comes down to its ultimate purpose which is to play a key role maintaining America's AI Supremacy that's the end of the video I'm Lauren from Darkman digest where I talk about the most exciting insights in the tech industry so if you enjoy this content hit that like button subscribe to the channel and I'll see you in the next one catch you [Music] later

Channel: Dark Mode Digest

Views: 143,269

Keywords: scale ai, scale ai business model, scale ai alexandr wang, scale ai demo, scale aircraft modelling, scale ai founder, scale ai interview, scale ai ceo, scale ai donovan, scale ai spellbook, lucy guo scale, lucy guo passes, mit dropout, alexandr wang mit, alex wang ted talk, how scale ai works, scale data engine, openai chatgpt, ai applications, scale donovan, san francisco tech startup, Darkmodedigest, Morningbrew, alexandr wang bloomberg, yougnest self made billionaire

Id: HsQ8XkhkGdQ

Length: 11min 52sec (712 seconds)

Published: Wed Jul 12 2023