Barnes, a YouTube strategist who recently helped with the ideation of MrBeast's high-performing "trapped in a grocery store" video, wrote a thought provoking thread on the concept of mature ideation:
https://x.com/Barnical_/status/2058953235709456568?s=20.
The premise of the thread is that YouTubers tend to drift towards “safe” ideas and formats over years, that have performed well in the past. They somewhat avoid “risky” ideas.
Explore vs. Exploit
This is a classic computer science concept called explore vs. exploit, and what happens in Barnes’ idea of “mature ideation” is a natural consequence of exactly this equation.
For an unknown environment, how much do you play things safe where you have a known reward (exploit)? And how much do you keep trying new things to see if you can get a better reward (explore)?
Mathematically, the most common equation for this is the Upper Confidence Bound (UCB):
A_t = \arg\max_a \left( Q_t(a) + c \sqrt{\frac{2 \ln t}{N_t(a)}} \right)
It very much literally chucks this idea of explore vs. exploit into something more formal, but the basic idea is: you’re trying to get the maximum value out of the sum of two parts you add.
The first part, Q_t(a), is literally the average reward expected for action a. This is pure exploit reward.
The second part is more fun though, which is the “explore” number. There are two subparts to it:
There’s a constant, c, which is a tunable constant (often 1 or 2) controlling how much to favor exploration.
Then the other big term is the exploration bonus:
\sqrt{\frac{2 \ln t}{N_t(a)}}
This term is larger for rarely chosen actions, and shrinks as those actions are tried more.
The thirst for exploration is quenched by the satisfaction of exploitation
So what Barnes is advocating for is to increase c, the thirst for exploration.
But this exploration bonus is having to be directly balanced against the very tempting Q_t(a), i.e. the guaranteed reward you get.
The desire for exploration is even further reduced by the psychological aspect of being a YouTuber with a team and families to feed. It’s much more tempting to play things more safely, because people rely on you now.
Even strategically, it’s not always the worst thing. YouTubers often find competitors make their videos; by giving up on a particular space, you might give up a juicy video or format to someone else for no reason.
Still… in reality, Q_t(a) actually should be replaced by a different term that accounts for fatigue, which occurs either in a minor way from viewer psychology reasons, but I think from a much more significant way from market dynamics (competitors making videos that fill the demand).
Put more simply, the act of making a video itself makes it less likely that that same video will perform well in the future.
And exploration gives you more data on what will work in the future.
Keep exploitation and exploration mentally separate
I think the way to make things work is that exploration plays are always explicitly data plays. You:
need to do them, so that your reward system doesn’t collapse by fatigue with nowhere to move in the future, and
need to actually have time to both reflect and implement findings that come from these “failures”.
Yes, they may get less short term views, but the point is building a more robust mental model for things.
Proxy Channels: A novel approach to avoiding the risk of "bad ideation" altogether
But less short term views is often a hard-to-swallow outcome. Who wants less views? Nobody.
In this scenario, if data is the play but you want to de-risk things significantly, I prose the idea of creating/using a proxy channel.
You want a Channel B that reflects an accurate world model of how viewers would behave on Channel A. Channel B is a sacrificial channel that doesn't need to perform at the same level as Channel A, because it's not for views, but for data on what will perform.
If you go even more hardcore into this concept, this “proxy channel” play is not even just using YouTube. YouTube views are an imperfect snapshot of all the systems and flows that make up the world’s attention. So any other platform’s performance for a particular piece of content can be a rough marker for how it would perform on YouTube.
An obvious proxy would be performance on other video platforms, such as Instagram and TikTok and, increasingly, X. Note that different platforms behave really differently, so this is not necessarily a good proxy. But there are useful approximations that can be observed cross-platform, and anecdotally it usually it seems to be the case that a video that performs well on two separate platforms will probably perform well on the third.
Conclusion
“Proxy strategy” is an advanced approach for YouTubers with a lot of resources on hand, or at the very least, a lot of good systems. Whilst proxy strategy is not necessarily the optimal solution to the explore vs. exploit problem, it's just a different way of thinking about YouTube.
I've been involved in thinking deeply about what it means to make high=performing YouTube videos for a very long time. I made a whole course on YouTube storytelling for millions of views for instance.
I do think that i find myself personally more refreshed by watching the videos of YouTubers who create YouTube videos as an art rather than as a systems optimisation problem. This is kind of the same as you might feel when you listen to music that's meant to top charts, versus meant to convey something complex with a layering/intent that only a few people would even care about or notice.
But, art and strategic optimisation are not necesssarily the dichotomy we often present them as.
I hope by presenting an alternative approach, it lets you as a YouTuber keep your channel performing well, whilst also having a relatively riskless way to do crazy things and not sacrifice your channel in the process.
tl;dr
Consider making a second channel purely for data/exploration, so that your first channel can safely continue to make videos that you know will perform well. I've created the term 'proxy strategy' for this today.
