Your conclusion reminds me a bit of the "Picard principle" from Star Trek: the Next Generation:
> It is possible to commit no mistakes and still lose. That is not a weakness; that is life.
It's a thought I carry around with myself *all* the time. However, my guess is that we actually do commit a lot of mistakes (or, at least, *I* do), probably some due to what you've brought up here. Good to keep in mind!
Good article! In the terminology of Machine Learning, this can be thought of as "your Learning Rate might be set too high" (which I find an intuitive way to think of it)
I’d argue that it is simpler than that. In machine learning terms… there is just too little data.
Relatedly, you can argue that, in a Bayesian sense, your priors are too weak to allow a reliable conclusion based on only a small amount of “new” data.
the part that's wild is rousey actually did update but she just updated in the wrong direction. doubling down is also a learning rate problem, it's the model deciding the data is the noise.
There is also the related problem of being judged on your win-loss record by people who have even less information with which to evaluate the reasons for wins and losses than you do.
The Rousey case is a perfect illustration of what the military strategist John Boyd called a dead frame: a model that worked so well for so long that the person inside it lost the ability to see it as a model at all. The overfitting you describe is one mechanism. The harder version is when the frame is so deeply grooved by success that no amount of new data can dislodge it—not because the person isn't updating, but because they've stopped being able to imagine an alternative frame exists.
Your prescription—spot the fake lesson, adjust the prior—is the right one for calibration errors. The trickier cases are orientation failures (think Merkel as a leader or Kodak as a company), where what's needed isn't a better update but a complete reframe. That's rarer, harder to teach, and harder to survive. Rousey's second fight suggests she was there.
Something about drawing conclusions is super fun. I like to think that I draw conclusions for fun and able to discard them quickly if counter-evidence arises, that way I can get better at forming them.
Or rather keep conclusions at bay rather than live them.
Wonderful essay as always, looking forward to the book!
nice takeaway lesson! you explained it quite well with the poker example, that it's easy to learn the wrong lessons early. especially if you're ronda rousey.
This 'punch in the face' is the physical manifestation of a systemic crisis: the transition from an era of abundance to a reality defined by scarcity of flow. In complex systems—entities made of many interconnected parts that adapt to their environment—stability depends on a constant, high-volume throughput of energy and resources.
When this flow is disrupted or becomes too expensive, the system can no longer maintain its intricate structure and begins to 'punch back' as it forcedly simplifies. We are witnessing the moment where the virtual economy's abstractions fail to account for the friction of the material world. It’s not just a geopolitical shift; it’s the system’s thermodynamics reaching a breaking point where narrative can no longer substitute for hardware.
Learning to jiu-jitsu my way out of the countless lies I have accumulated around identity, skills, and emotions has been the biggest upgrade to my mind over the last year. Can’t wait for the book
Thanks for writing this. I had internalized this but was always fighting against it thinking that I am just making excuses for myself, but unfortunately that is the structure of the world. And that is okay. We just shouldn't get stuck on "learning from successful people" or a definition of success someone else made up for us.
Unrelated to the article, but I have to say I miss Alexander Naughton's illustrations; the AI images feel very low-effort (this one even appears to have a fake signature!).
I agree his were better! Unfortunately the turnaround time for custom human art is just a lot slower — even paying more to expedite, they took a week+, and it just stopped being workable. But I’d recommend him highly for someone with a different approach to writing!
yeah, totally understand the practicalities of it! i just think if you’re gonna go down the AI art route, there should at least be a bit more care put into it :P (e.g. removing the aforementioned fake signature)
It's funny you say that. I play a retired blonde haired paladin in my D+D campaign, and I always felt emotional about the images Cate has generated. Especially "The lies I used to tell myself"'s picture really hit me hard.
Your conclusion reminds me a bit of the "Picard principle" from Star Trek: the Next Generation:
> It is possible to commit no mistakes and still lose. That is not a weakness; that is life.
It's a thought I carry around with myself *all* the time. However, my guess is that we actually do commit a lot of mistakes (or, at least, *I* do), probably some due to what you've brought up here. Good to keep in mind!
Good article! In the terminology of Machine Learning, this can be thought of as "your Learning Rate might be set too high" (which I find an intuitive way to think of it)
I’d argue that it is simpler than that. In machine learning terms… there is just too little data.
Relatedly, you can argue that, in a Bayesian sense, your priors are too weak to allow a reliable conclusion based on only a small amount of “new” data.
I like this!
Or just over fitting
the part that's wild is rousey actually did update but she just updated in the wrong direction. doubling down is also a learning rate problem, it's the model deciding the data is the noise.
There is also the related problem of being judged on your win-loss record by people who have even less information with which to evaluate the reasons for wins and losses than you do.
The Rousey case is a perfect illustration of what the military strategist John Boyd called a dead frame: a model that worked so well for so long that the person inside it lost the ability to see it as a model at all. The overfitting you describe is one mechanism. The harder version is when the frame is so deeply grooved by success that no amount of new data can dislodge it—not because the person isn't updating, but because they've stopped being able to imagine an alternative frame exists.
Your prescription—spot the fake lesson, adjust the prior—is the right one for calibration errors. The trickier cases are orientation failures (think Merkel as a leader or Kodak as a company), where what's needed isn't a better update but a complete reframe. That's rarer, harder to teach, and harder to survive. Rousey's second fight suggests she was there.
Good luck with the book.
hi claude
It’s scary to consider this is a problem you can’t easily detect because by definition you don’t know that the evidence of *your* sample is skewed.
Something about drawing conclusions is super fun. I like to think that I draw conclusions for fun and able to discard them quickly if counter-evidence arises, that way I can get better at forming them.
Or rather keep conclusions at bay rather than live them.
Wonderful essay as always, looking forward to the book!
nice takeaway lesson! you explained it quite well with the poker example, that it's easy to learn the wrong lessons early. especially if you're ronda rousey.
This 'punch in the face' is the physical manifestation of a systemic crisis: the transition from an era of abundance to a reality defined by scarcity of flow. In complex systems—entities made of many interconnected parts that adapt to their environment—stability depends on a constant, high-volume throughput of energy and resources.
When this flow is disrupted or becomes too expensive, the system can no longer maintain its intricate structure and begins to 'punch back' as it forcedly simplifies. We are witnessing the moment where the virtual economy's abstractions fail to account for the friction of the material world. It’s not just a geopolitical shift; it’s the system’s thermodynamics reaching a breaking point where narrative can no longer substitute for hardware.
Learning to jiu-jitsu my way out of the countless lies I have accumulated around identity, skills, and emotions has been the biggest upgrade to my mind over the last year. Can’t wait for the book
If only the intro to every article could be this honest.
Thanks for writing this. I had internalized this but was always fighting against it thinking that I am just making excuses for myself, but unfortunately that is the structure of the world. And that is okay. We just shouldn't get stuck on "learning from successful people" or a definition of success someone else made up for us.
As my Greek wife likes to say, I'm not that smart, but I'm a smart ass!
Unrelated to the article, but I have to say I miss Alexander Naughton's illustrations; the AI images feel very low-effort (this one even appears to have a fake signature!).
I agree his were better! Unfortunately the turnaround time for custom human art is just a lot slower — even paying more to expedite, they took a week+, and it just stopped being workable. But I’d recommend him highly for someone with a different approach to writing!
yeah, totally understand the practicalities of it! i just think if you’re gonna go down the AI art route, there should at least be a bit more care put into it :P (e.g. removing the aforementioned fake signature)
It's funny you say that. I play a retired blonde haired paladin in my D+D campaign, and I always felt emotional about the images Cate has generated. Especially "The lies I used to tell myself"'s picture really hit me hard.
Have you read Fooled by Randomness by Taleb? He explores these ideas in pretty great detail.
I haven’t! The advantage of being poorly read is I get to believe I’m the first to come up with everything
The failure is not in failing. The failure is in not learning from our failure.