Fail fast used to mean: identify problems early and quickly to correct them and learn, preempting potentially catastrophic risk materializing later.

Now it seems to mean: accelerate value extraction at the expense of risk maximization so when the foreseeable risk materializes we can cash out.

The best predictor of a model breakthrough? A business partner asking for something current models aren't quite suited for while I explain the current limitations.

Within three days, some research will be announced making it possible.

🍿 Me watching valley companies throwing hands at Deepseek.

Yes, the ethics are questionable. But it blows open the narrative about the amount of resources needed for advanced models. Time to get lean!

The exodus of Tik Tok users to apps like Red Note seems to reinforce the idea that digital privacy is something only niche user bases care about.

In the last decade I've had so many conversations with CISOs advertising cybersecurity as a reputational advantage and differentiator. Is it?

Spending over a decade in risk management gave me an orientation towards AI harm mitigation.

This year, I will pursue, promote, and participate in AI projects that I believe have a reasonable chance of positively impacting the human condition.

And I will keep the "hall-monitor" critiques specific.

Perhaps we spend too much time thinking about what we will do with our products and not enough time thinking of what our products will do with us.

The purpose of a system is what it does

You do not need more metrics. You do not need more dashboards.

You need more questions.

When did we lose strategy?

What I see are pseudo-random sets of words put forth by leadership as a fealty pledge for managers to come up with the most synonyms for those words.

What do you want to do? How are you going to do it?

Say something.