zestyping

It has nothing to do with how "unpredictable and highly capable" they are. That's a red herring.

If it cannot be held accountable, it cannot take responsibility. A human is responsible. It's as simple as that.

> Yes, the code is generated. Yes, the docs are generated.

> I also wouldn't call this vibe coded either.

It's vibe-coded. Get over it.

It's a crime in this day and age to make a new font with an uppercase I and lowercase L that look the same.

It's hard to believe people are still designing new fonts with this glaring error. In 2026!

Could you explain the first step, "Manually Labeling Colors ? What did you manually label? I didn't quite follow.

From the way the article is written, it sounds like you presented yourself with a bunch of faces of completely random colours and then manually sorted them into two categories, "plausible skin colour" and "implausible skin colour".

But I feel like I must be misunderstanding, because I don't see why you would do that. Why wouldn't you gather photographs of actual faces and sample colours directly from the photos?

So now there are two mutually incompatible versions of YAML that silently apply different semantics: both do implicit coercion but _different kinds_ of implicit coercion. YAML 1.2 still coerces an arbitrary set of ways to spell "true" and that's exactly the insanity that created this disaster in the first place. The YAML spec writers just couldn't let go of their bad habits.

Honestly, this seems like an even worse situation than before. If you're making a new tool and choosing a configuration format, stay away from YAML.

Approval, not ranked-choice.

Ranked-choice reduces transparency and understanding of the vote-counting process, disenfranchises an alarming percentage of lower-income voters, obstructs risk-limiting audits (which are essential for security), and is non-monotonic (increasing voter support for a candidate can make them lose). Further, ranked-choice doesn't actually fix the spoiler problem and won't eliminate two-party dominance.

Approval voting is cheap and easy to implement, dead simple to explain, count, and audit. Not only does it eliminate the spoiler problem, it is easy to see why it does so: your ability to vote for any candidate is independent of your ability to vote for any other.

Whether they are sentient isn't even relevant. I agree with you that they aren't, but sentience is the wrong thing to focus on. It's also the sort of hairy, sensational question that will easily lead people down rabbit holes (and unfortunately that includes journalists).

Children are sentient, but we still hold their parents accountable. Adults are sentient, but in some coercive situations we hold the party in power accountable. The fact that they are sentient is not determinative.

What matters is that we have _no accountability mechanism_ for them. There is no effective way to hold AIs accountable, therefore we must hold their operators accountable, full stop.

The important indicator is "I spend more time on this than I myself want to." That applies equally well to games or anything else.

I like this approach.

Specifying the requirement in terms of measured impact is a good strategy because it motivates the app companies to do the research and find effective ways to address addiction, not just replace specific addictive UI patterns with different addictive UI patterns.

Building measurement into the law also produces a metric for how well the law is working and helps inform improvements to the law.

I don't think the same of them because they are not the same thing. Can you not see that the potential for harm is far greater? You can't simply ignore the potential uses of the technology you create. You have the choice to design your technology so it retains its usefulness while limiting the harm; have you given any time to thinking about how you could do that?

The alien is a diversion from the concern; I'm talking about realistic human avatars. Let's stay focused on that.

Let me suggest a worthwhile exercise. Just take ten minutes. What are some of the ways that realistic human avatars would make deception more effective or more scalable than previously possible?

Come up with three scenarios, and let's talk about them, honestly and thoughtfully.

When you generate real-time video of realistic-looking talking characters, the definition of success is fooling people into believing they are talking to a real person when they aren't.

If you pursue this, your explicit goal is deception, and it's a massively harmful kind of deception. I don't see how you can claim to be operating ethically here if that's your goal.

The primary purpose of generating real-time video of realistic-looking talking people is deception. The explicit goal is to make people believe that they're talking to a real person when they aren't.

It's on you to identify the "immense" benefits that outweigh that explicit goal. What are they?