Q McCallum

The first rule of Complex Systems Club™ is that everything's connected.

(This is also the only rule. Everything else about complex systems is just a rewording of this rule...)

This shows how good decision-making and good risk-taking overlap (or, arguably, are the same thing):

It's about positioning yourself such that mistakes are not fatal. You can be wrong today ... and still get up tomorrow to try something else.

Wisdom for the week ahead.

One point in favor of surveys:

A well-designed, well-executed survey will outdo a(n ML/AI) model any day of the week.

No need to collect tons of potentially-weak-signal data to _infer_ what people want when you could _ask_ them instead.

Certain games develop one's intuition for probability, decision-making under incomplete information, and calibrated risk-taking ... all in a safe environment.

@matt-levine.bsky.social 's one-card poker? Sound like a winner of a teaching tool. For people of all ages.

Risk, Uncertainty, and Risk Management: A short primer Exploring my favorite definitions of these terms

This logically leads to a treatise on risk and risk-taking. I'll spare you a long read on Bluesky and point to:

1/ a recent write-up of mine, for terminology:

qethanm.cc/2025/07/31/r...

2/ Aaron Brown's "Red-Blooded Risk" or "The Poker Face of Wall Street"

Why is this skill important? It's all about decisions.

>> Every decision is a bet on a future outcome. <<

And life is full of decisions. Some larger or tougher than others. Especially in the workplace.

(I dig the one-card poker approach because it includes two additional elements: simplicity and speed.)

If poker isn't your thing, that's cool. There are other options. The key ingredients are:

- an element of chance/randomness (dice, PRNG, card shuffle)

- something on the line (chips, money, bottle caps)

Even low-stakes game will teach you about your own decision-making apparatus. And hone it.

Certain games develop one's intuition for probability, decision-making under incomplete information, and calibrated risk-taking ... all in a safe environment.

@matt-levine.bsky.social 's one-card poker? Sound like a winner of a teaching tool. For people of all ages.

One point in favor of surveys:

A well-designed, well-executed survey will outdo a(n ML/AI) model any day of the week.

No need to collect tons of potentially-weak-signal data to _infer_ what people want when you could _ask_ them instead.

Most corporate surveys are terrible, for the two reasons I list above + plenty more that I'll skip for brevity.

But there are occasional gems. Just submitted one today that was focused, to the point, and therefore likely to elicit actionable data analysis.

Then we have "not including 'prefer not to answer' among the choices on a required question."

People telling you to mind your own business _is a valuable data point_ ... and if you refuse to accept this, you are tacitly accepting trash data. And that your analysis will therefore be invalid.

Great questions to ponder when designing a survey.

We've all been on the receiving end of terrible survey design choices.

My favorite is "we have a single touch point with the customer, so every department piles on."

This leads to sprawling surveys that people are less likely to complete.

What's the build date on this one? Any chance it overlaps with that brief Tex-Mex fad in late 1990s/early 2000s France?

A reminder for these troubled times ...

A whole day of @compiler-explorer.com admin. Trying to further untangle the Matt Godbolt from the Compiler Explorer ("Godbolt" ugh). Getting there.

If you (very kindly) sponsor *me* on GitHub please consider switching it to compiler explorer instead! In a few months I'll separate those...

They call this genAI goof a "typo?" That's ... a stretch.

Going for my usual line:

The danger isn't that the bots make mistakes.

The danger is in putting bots in places where those mistakes matter.

#RiskControls

Risk, Uncertainty, and Risk Management: A short primer Exploring my favorite definitions of these terms

I talk about risk and risk management a lot. The thing is, those words change meaning based on the context.

I've gathered my favorite definitions of risk, including a couple of my own, into this blog post:

qethanm.cc/2025/07/31/r...

it’s definitely theirs to grab in the sense that they do not have their hands on it now

[looks around at home full of books]

"This is fine dot gif"

Risk, Uncertainty, and Risk Management: A short primer Exploring my favorite definitions of these terms

I talk about risk and risk management a lot. The thing is, those words change meaning based on the context.

I've gathered my favorite definitions of risk, including a couple of my own, into this blog post:

qethanm.cc/2025/07/31/r...

(To be clear: I reason the "talk show" and "morning/breakfast program" formats pull good numbers. But _that_ good? Wow, if so.)

Sad to hear this news - Wondery put out some of my favorite podcasts.

I wonder how much this pivot to video is about better/easier advertising rather than actual consumer preference for watching people chat?

They call this genAI goof a "typo?" That's ... a stretch.

Going for my usual line:

The danger isn't that the bots make mistakes.

The danger is in putting bots in places where those mistakes matter.

#RiskControls

I was going to include an excerpt for sharing, but I'd need to include the entire post.

Every word here is gold. Every. Single. Word.

Rabbits bouncing on a trampoline.

Finally, a use case for AI- (KI-)generated video.

Rabbits bouncing on a trampoline.

Finally, a use case for AI- (KI-)generated video.

Also, I'm not a fan of the term "scheming" as was used in the article.

It's dangerous to imply the bots are sentient. They are not. They simply emit (numeric representations of) word-chains.

(Though, I will concede: English makes it awkward to express action without implying agency...)

The problem is not that the machines emit outputs that appear to "scheme" against us.

The problem is that people will connect the machines' outputs to real-world actions -- either by following the bots orders or permitting the bots to take action without oversight.