How Siraj Raval’s machine learning videos were the victim of their own success
|Oct 11||Public post|| 16|
I first came across Siraj Raval around 2016, when a data scientist friend sent me his video. “This is ridiculous,” he said, and I had to agree that it was. But, it was also hilarious.
Who else was making funny, well-made music videos about machine learning? It was nice to watch, because it’s always nice to see a humorous subculture develop around your discipline. At the time, there was not a lot of funny data science content out there (unless you count my awful tweets, and those came much later).
My interest was piqued. At the time, Siraj had a couple of other videos on machine learning topics like Python and deep learning, and, thanks to YouTube’s ceaseless recsys pusher, I watched them.
They seemed easy to get into, and really well presented video-wise, but very surface-level as far as concepts went. In the sea of data science content I was watching, I’d see his videos pop up from time to time as recommended related views, but I didn’t think about it any more than that.
Then, all of a sudden, a couple weeks ago, this news item came out, seemingly out of nowhere:
The AI hysteria has led to a rash of budding engineers hoping to land a cushy job somewhere in Silicon Valley.
So it's no wonder that thousands flocked to an online course titled Make Money with Machine Learning fronted by Siraj Raval, a self-proclaimed AI educator, rapper, and entertainer with nearly 700,000 subscribers on YouTube.
But his reputation as a rising pop science star has been called into question after hundreds of students enrolled on his 10-week online course began demanding refunds. For $199, machine-learning enthusiasts are fed weekly video lectures, online quizzes and reading assignments to learn how they can make money by applying AI in various industries like finance, agriculture and healthcare.
Ray Phan, a senior computer-vision and deep-learning engineer at Hover, a 3D software startup based in San Francisco, told The Register that the lectures were confusing and contained mistakes. The online projects were too simplistic to be useful in the real world, and Raval was absent and unsupportive.
Even if he hadn't taken on so many students, the course was still received pretty poorly. Many claimed Raval had the habit of copying and pasting code taken from other people's GitHub repositories, or linking to some of his old content, and taking reading material publicly available on sources like Medium blogs.
Sven Niederberger, a mechanical-engineering student known as embersarc on GitHub, told El Reg that his code, simulating how to land a rocket with reinforcement-learning algorithms, was used in one of Raval's YouTube videos.
I was surprised. My first impression of Raval was that he may have been a little light on content, but definitely, judging by the quality of his video production and how much code he had on GitHub, someone with a background in machine learning and a serious data professional.
That’s what many others seemed to think, as well, because he got endorsements from many students, as well as famous people in the AI community:
Raval began his internet fame by uploading videos describing how to build a neural network in five minutes, how logistic regression or sentiment analysis works, and how to build Bitcoin or healthcare startups. His work has been praised by revered figures in the machine-learning community like DeepMind CEO Demis Hassabis, and shared by Tesla and SpaceX boss Elon Musk too. So far, so good.
The Register does a really good job describing Raval’s videos, the courses, the subsequent confusion and refund process but what they don’t get into was: how did this happen? How did someone who set out to make machine learning accessible to everyone end up upsetting hundreds of students, machine learning content creators, and machine learning practitioners online?
And, more importantly, why did people in the industry, including the people at the top, promote this? The Register article doesn’t get into that much. So, I decided to dig a little deeper. And by digging a little deeper, I mean just conducted a bunch of Google searches to see what had really happened.
If you search for Raval, all of his own personal branding comes up first. And his personal branding, as the kids say these days, is lit. He has a really slick website. But, as with seemingly everything about Raval’s machine learning content, there is a lot of very well-executed flash and not a lot of content. What I was really looking for, was to understand his background.
That was a bit harder to find. Most of the podcasts and blogs that conduct interviews with him describe him as a “data scientist, best-selling author, and YouTube star”. But where did he work, doing what? How did he get into making machine learning videos?
It turns out that that he grew up in Houston and went to Columbia University, where he first majored in finance. After a semester-long suspension for stealing a laptop, he decided to “try to do something positive” and majored in computer science to become a software developer.
Based on Twitter, it doesn’t look like he ever finished his degree.
According to his LinkedIn, he spent about one year each at a robotics firm in New York, Meetup.com, CSB Interactive, and Twilio. His initial website has links to his work in iOS development and titles him as a “software developer.”
It’s important to note at this juncture that his LinkedIn pegs his time at Columbia as ending in 2012, 2013 at the latest. In 2016, a link to his YouTube channel, Sirajology, appears. This would have been at Twilio, where he was a “developer educator.” So, when he started down this path, of creating data science content, then, he had been out of college for three years. He was working as a developer, and didn’t have any data science experience.
Which is ok! Lots of people write tutorials for the sake of learning something and teaching others (Myself included. For example, when I wrote this Python type hints post, I had never used them in a project before). In fact, I think more people closer to the beginning of their careers should write tutorials, because they remember beginner problems more clearly than senior people.
However, that’s not the way Siraj framed it from the beginning. Almost from the get-go, he positioned himself as someone who would teach you, rather than someone who was learning along with you. He never gave his credentials. He projected an authoritative air.
And podcasters and others who interviewed him as he became more popular never asked to clarify. And, more importantly, he never corrected them.
Many beginners were drawn in by the engaging content. Learning how to program, and data science in particular, is so popular these days that many people want to participate, to become a data scientist no matter what. Anything that makes entry into the market easier is fantastic. Data science and machine learning are so full of dense terminology, software packages, arcane instructions, and academic language, that anything that makes it easier will instantly become popular.
And as more people watched these eye-candy videos, YouTube’s merciless recommender algorithm, which would show one video from Siraj, showed more and more. Since beginners were likely to look for videos from stars in the field like Andrew Ng, and also clicked on Siraj’s viral content, more and more people were exposed to his content and became fans. His current channel is at almost 700k subscribers.
The thing grew and grew, until people like Elon Musk were retweeting his content, lending credibility, a fact that was proudly added to his site in 2017.
But the Machine Learning community on Reddit suspected something was amiss. In a thread from July of 2017 called, appropriately, “What do you guys think about Siraj Raval's videos on YouTube?”, the original poster asks,
I was watching a few lectures on YouTube about backpropagation in neural networks, because I've forgotten how exactly it was done in detail and wanted to brush up on it. Anyway, YouTube continued playing videos of Siraj Raval for some reason. So, I watched a few of his videos. I must admit I was rather bemused by his approach. He condenses ML subject matters down to things like "TensorFlow in 5 Minutes" or "How to Predict Stock Prices Easily". He talks to the audience like they are supposed to be beginners, but then proceeds to steamroll them with advanced topic information. He glosses over things and is very handwavy. Everything I've heard so far is correct, but I don't understand what or to whom it is useful for. I could follow his videos fine, but I've got plenty of education in this field, I just can't imagine being introduced to the subject mat[t]er like this.
Idk. Maybe I'm being too harsh. He is definitely knowledgeable and entertaining, and a cool guy. But how educational is his channel really?
Other commenters chime in that his shorter content is better, but that his longer content is not helpful, and that he takes from a lot of other tutorials without crediting them.
And then, there were curious comments like this one:
He's good for beginners but doesn’t really have any insight of his own. One video that I watched from him was him literally copying code from a blog tutorial and he just recites the intuition off a secondary screen that you cant see off-screen.
Finally, there was this one, which Siraj, replies to himself, with his own reddit handle (no longer active), “Yup, getting better at the math, thanks for the feedback.”
There were also signs in the public interviews he was giving, like this one, for The Startup, “Medium's largest active publication, followed by +511K people”:
The criticism mainly stems from the approach Siraj takes in his videos. His work is as much about educational content as it is about conveying the sheer excitement he has about the topic at hand. More often than not, it’s hard to pinpoint exactly where one ends and the other starts, giving the material a superficial feel. As a result, many shun him for being too superficial and lacking in technical depth. This is arguably true, but beside the point.
This all is apparent from the approach he takes in creating his content. In a candid interview with Lex Fridman, Siraj mentioned that his ambition is to increase his subscriber count to a million, so as to reach more people. To get there, he meticulously monitors his analytics, mainly audience retention metrics, in order to pinpoint exactly what works and what doesn’t in his videos. One of his main findings is the importance of the relevancy of pop culture references, including jokes and memes. As such, he keeps an eye on mainstream channels like the previously mentioned PewDiePie and others, trying to relate AI to the timely themes that the youth find entertaining.
As for the future, Siraj is thinking about publishing books, coming out with an album, and starring in a Netflix show. Even fashion isn’t out of the question, he says. To him, these disparate mediums are just like different keys in the piano, different instruments in the orchestra of education. And just like many of the great composers in history, he, too, is ahead of his time.
As with the videos, all this press snowballed on itself, and he gained more and more followers, and with them, a larger, critical audience of people potentially with more experience in machine learning than he had.
When he was just starting out and small, no one was telling him that the math and the code were wrong. It seems like he was doing it, yes, to gain followers, but also to learn. You can learn at 50 YouTube followers. You can’t learn when you have 700k. Which is maybe where he started copying and pasting code, to keep up.
The constant stream of ego boosts, likes, favorites, and video shares can grate on anyone. Twitter, Facebook, and Instagram have all said at some stages that they’re all thinking about removing likes from their platforms as part of the ongoing process to detoxify social media.
It’s one thing to expose adults to the platform: already bad. It’s another to expose someone fresh out of school to a constant stream of never-ending validation from people much less experienced in a field than he was. Put those two things together, and you get where Raval is now.
The bigger he got, the bigger he felt the need to become, particularly with so little working experience and so many people telling him he was amazing. It can’t have been easy to have been working in the warped echo chamber of the internet, particularly fed by retweets of Elon Musk and friends. All of this culminated in him creating his “School of AI” and then announcing his most recent course course, which brings us back to today.
What does it all mean? Was Siraj wrong? Or, was he just trying to create AI content for the masses? Was he, as he notes on Twitter, trying to give back?
Or, was he exploiting students, convinced that data science was their ticket to riches (If you haven’t yet, please see my post on how it’s not)? Or was he just very shrewdly taking advantage of today’s internet, which values both hype around topics like machine learning, little fact-checking of technical work, popularity in favor of substance, and five-minute videos? Or maybe, he was just young, unprepared for the (very heavy!) responsibilities to teach hundreds of people in topics he himself was maybe not fully up on.
A little of all of the above, I think, and it’s this nuance that the Register story, and to some extent, the Reddit threads, don’t tease out. It’s fully possible that Siraj was both cognizant that he was scamming students, but also not aware to the extent that he was hurting people, but also maybe not meaning to hurt people, and wanting to keep up with the reputation which had so far very much preceded him. It’s maybe all of this stress that led to the code copying.
So, where are we now with all of this?
So far, Siraj has said the following on his Twitter:
He ends with,
“I’m feeling pretty low right now, but I’m in this for the long-haul. I understand that it will take time to rebuild trust, but I’ll work for it.”
The Register article notes that he’s in the process of issuing refunds,
At $199 a pop, having 1,200 students means a total of $238,800. That pile of cash has dwindled since Raval began refunding students and at least 250 of them have got their money back so far.
Which brings us to another point: Was this also the internet’s fault, just as much as it was his?
I’ve written previously about many things that are wrong with today’s internet: faulty recommendation systems that optimize for clickviews, the attention economy, and systems that need to constantly scale, thereby pushing for content that optimizes for excess rather than focused attention and learning. Siraj was, very smartly, able to capitalize on all of them, until he got to a point where he was in over his head.
He had all the skills necessary to command attention on the internet. But he didn’t yet have the underlying technical skills to actually teach the content, or organization skills to run a course for learners. All of this caught up to him in the end.
In a fair internet, someone in the data science community would have taken him aside quietly and told him that he should maybe reconsider, before he got to the point where he made the tech news. He would have had a maybe a physical or online community to bounce ideas off of.
But, as he notes in an interview,
I wake up at 7:30am, then I ride my bike to my co-working space. When I’m working, I like to be completely alone, I like to be in my own space and then I go home after like 6hours of working and then I will spend the rest of the day just kinda like alone, thinking of ideas so clearly, usually, I’m kinda in my own space but on weekends, I try to be more social.
It was just him, alone with his ideas, with the internet constantly expanding their reach without any recourse. Everyone (myself included) was just on the other end of the line, consuming the memes.
On top of all of this was the layer of AI and ML hype that is so pervasive in the industry today, that makes CEOs and students alike salivate, making new learners, particularly in emerging economies, desperate to shell over cash that grants them entry into what they perceive is the exclusive club of elite data scientists running super-cool algorithms.
And not only newbies. Just as recently as a couple of days ago, he was interviewing Vinod Khosla, the legendary VC, on AI. (As an aside, please read this fascinating and depressing story about Khosla and his battle to try to ban the public from walking by his house to get to the beach.)
Once the heady brew of hype is applied to technologies that are complicated to understand, even reasonable people can lose reason. In the same way that Elizabeth Holmes was young, without people around to ground her in reality and tell her “no”, and able to rise to the top by scamming high-level people who didn’t fact-check her complicated science behind her healthcare startup, Raval was able to get much farther than he probably would have in a reasonable ecosystem.
What’s the takeaway here, for ML people, for people in tech, for any reasonable person armed with today’s internet?
We need to provide people with the right tools and content to evaluate what they’re watching education-wise for tech topics
We need to end the hype cycle around AI and ML pronto. I’ve been pretty vocal about this myself, and one of the reasons I started this newsletter - to moderate the gushing fountain of both tech enthusiasm and tech negativity from the mainstream media.
We need to be critical of what we watch and read (easier said than done!)
We need to help people around us if we see they need technical boost, and to make ourselves availalable. We need to learn to offer constructive criticism outside the parameters of a sarcastic tweet or a one-off YouTube comment. In today’s internet, ultra-engineered for attention and outrage to push all our worst buttons as humans, it can be really hard on a daily basis, but we have to try.
I’m hopeful that Raval’s story gives more people pause before consuming things at face value, particularly in the data science realm.
And, I’m also hopeful that Raval can recover from this and learn the important lessons: First, don’t plagiarize code! Second, don’t overpromise and underdeliver. And third, don’t use Tensorflow. Just kidding. Third is, don’t wade in over your head. Ask for help. Form a network and find mentors.
Siraj has a Q+A on YouTube today at 11 Pacific. I’m hopeful that he address some of this, and it helps him move forward, both for the sake of his own reputation and peace of mind, and for the industry and internet, as a whole.
Artwork: Athens Is Burning! The School of Athens and the Fire in the Borgo, Dali, 1979
What I’m reading lately:
The Rihanna Vogue interview.
The replies to this:
About the Author and Newsletter
I’m a data scientist in Philadelphia. This newsletter is about issues in tech that I’m not seeing covered in the media or blogs and want to read about. Most of my free time is spent wrangling a preschooler and a newborn, reading, and writing bad tweets. I also have longer opinions on things. Find out more here or follow me on Twitter.
If you like this newsletter, forward it to friends!