WelcomeAIOverlords
Educational Content on Machine Learning and Data Science
The requirements are 4000 watch hours in last 12 months and 1000 subscribers. This once seemed impossible, yet here we are. Although my content output is far less than many other channels, it is mostly evergreen and many of the videos continue to increase their reach. Despite acceptance into this program, I currently have no plans to monetize my channel with YouTube ads.
Subscriber counts for the YouTube channel, email list, and Discord server have all crossed an order of magnitude milestone recently. Not only that, but the rate of increase has substantially increased. I currently have 14 videos, the latest of which has become my most viewed video ever despite being only 5 days old, compared to my second-most viewed video, which is almost 7 months old.
One of the key lessons learned for me is that the success of a video is not necessarily proportional to the effort required to make it. Although I still want to invest in making high quality evergreen videos that take weeks/months to make, I want to balance that with videos that require significant less time to make. So far, I have found two strategies for this. The first is having guests on to talk about their work. I have tested this by doing live streams of academic talks. The second is making videos that present other people's accomplishments. My first attempt at this was in showing the new capabilities of OpenAI's new API. This mainly stitched together demos they released and provided a narrative to summarize the work. This only took a few hours and has been highly successful.
On to 10,000.
-
1
Congratulations! 1k subs is a huge milestone 🎉
Why do you think the 5 day old video was so succesful?
If you're looking to maximize the views you capture when the YouTube algorithm recommends your video to potential viewers, you'll want to optimize your CTR (e.g. with better thumbnails). I actually run a service that helps with this, which I think would be very useful to you building evergreen content: https://thumblytics.com
Shoot me a DM if you have any questions!
Another order of magnitude crossed for number of YouTube channel subscribers. I currently have 6 videos live and am making progress on my most ambitious video yet.
I am also noticing that YouTube is driving significantly more traffic to some of my videos that was previously the case.
Next up: 1000.
Only 10 subscribers doesn't sound like much, but it's a new order of magnitude, which I think is worth celebrating. Particularly given that I'm a brand new channel with only 2 videos.
I have also started to see the "rising tides lift all ships" phenomenon. New video content allows me to post about it on Medium and other partner blogs, like KDnuggets, LinkedIn, and my personal blog. I have a call-to-action to sign-up for my blog subscriber list, where I plan to send out exclusive/early-access content. I have already started to see those organic subscriptions come in.
Most importantly, I'm hearing a lot of positive feedback from real, organic viewers, not just friends that I ask to view it as a favor. This is proving that my core value proposition isn't crazy.
When chatting to a friend, he suggested I start making video content. I really liked this idea, as I found a lot of ML YouTube videos were either very technically shallow, or had little-to-no production (e.g. a camera in the back of a lecture hall). I thought I could do better and created the WelcomeAIOverlords channel: https://www.youtube.com/channel/UCxw9_WYmLqlj5PyXu2AWU_g
I quickly learned that it was a ton of work to create ~10 minutes of visual content. I went ahead and prioritized shipping something, even if it wasn't perfect, and then asking for feedback via friends and the handful of subscribers from my blog subscription list. They provided excellent feedback, and my second video ended up being far better than my first.
What I try to keep in mind is: my first videos will be my worst, so just keep producing and learning.
I noticed I was getting substantially more traffic for some Medium articles than others, so I began to dig in a little with their Stats tool. I started to realize that a Google search for the academic papers I covered tended to return the paper as the first result, and my article as the second or third.
Specifically, it returned a link to the Medium Publication that was hosting my article (TowardsDataScience), and not to my blog or even my direct Medium post.
I began to then import my story to medium so that my blog would receive the canonical link. I also started putting notes in my articles about them being originally posted on my blog.
This is only tangentially related to my objective of providing new content, but there have been major indirect benefits.
Most importantly, being exposed to a new problem domain has given me the opportunity to learn more techniques and get an additional perspective on solving ML problems. This has directly led to new content that's better informed.
Second, the additional side-income can be used as funding/investment in my other efforts, such as buying equipment or for other's freelance work.
In the past, I found that I would often forget things I spent time learning, and would then need to invest time to re-learn in the future. I found it very helpful to summarize the key points in plain language instead of returning to the original sources, which are often highly academic and difficult to quickly read. I got so much value out of this, I decided to start posting it publicly in case others found value.
This led to the creation of my technical blog at https://blog.zakjost.com. I published my first post, which summarized an academic paper that I read as part of a Journal Club at work: https://blog.zakjost.com/post/rf_treatment_effects/
I also started cross-posting on Medium since I was unsure if my blog would be discoverable.
Comment