Randy Fernando

What gets measured?

These kinds of things.

Whose feedback guides iteration of the product?

What data is used for AI training?

What instructions did you give to the AI, to the model, in terms of how to behave?

Every choice that's made reflects trade-offs.

And it also reveals the true prioritization of the values behind what you might hear people say in public.

Another example is with AI safety.

You'll hear companies talk all the time about safety and how it's important to make sure we get it right.

But when you look at the actual investment in the headcount in their actual companies, the safety investment is something like 5% of the overall headcount that they have.

So it really doesn't match the rhetoric and it shows again the values behind the trade-off that they made.

So the gap between responsibility and power just grew massively.

And what we've done is we've built a global digital infrastructure that runs hospitals, elections, power grids, cybersecurity, financial systems, all of these things.

But the companies that built the components don't bear the cost when those components fail.

And the companies that build tools capable of tearing down all those systems are

also have no mechanism to be held accountable.

And yes, it was actually great to see Project Glasswing and see Anthropic withholding mythos and saying, look, what we need to do here is emphasize defense before offense.

And this is one of the principles that helps us when we're trying to get out of these situations.

You say, look, let's put more effort, put our best minds, our best technology on defense, figure that out, and then we'll democratize access more over time.

And this is really important because every AI company is always in a race.