Un-untitled.

My personal timeline has had white-collar AGI at 2028, with error bars that comfortably accommodate Kokotaljo's. So o4 hasn't made me update much.

I don't much see the point of trying to predict this sort of thing by landmark though (outside of like, foom take-off existential risk).

We need to start thinking and arguing and fighting about how to restructure society three years ago.

Setting aside existential risk, we are entering a situation where things turn out great for the descendants of the board members of Microsoft, OpenAI, Amazon, Anthropic, Tesla and Nvidia. It makes sense that they have such a rosy view of the future because it really is quite good for them!

As for the rest of humanity though (that means you, your children, grandchildren, etc), the response so far is basically.

Hmm, seems like one of the nodes is misbehaving. Will look into it later tonight.

Programming languages have borrowed from natural language the concepts of not, if, and, and or; approximately retaining their English meanings.

If you have fun/interesting/funny ideas for siikr's random "Did you know?" text, you can submit them to my ask box and I might add them.

This is an exercise in pure creativity for its own sake. I will not credit you for the text / content unless crediting you is somehow integral to the entertainment value of the text / content.

Constraints:

1. Any required text / content must fit within 250kb.

2. If it requires some programmatic component, I must be able to code it in less than 5 minutes in php without external libraries (I code pretty good, so don't worry too much).

3. The result must not melt my server if it requires serverside execution, nor any users computers if it entails clientside execution, nor violate the golden rule.

someone's stealing your Twitter shower thoughts

I think we need to stop worrying about how to make AI safe and start worrying about how bad things will likely still turn out even if AI is safe.

If we can motivate people to gently pump the brakes on development of AI in light of those (extremely legitimate, worrying, and in my opinion ultimately just as terrible for 99.9% of people) concerns, then we might, as a side benefit, even buy enough time to make AI kinda safe.

I realize my stance on this may be somewhat controversial, but I'll have to speak my conscience here: I am of the opinion that murder is generally a pretty bad thing to do, and one really ought not do it.

Slow home net isn't ideal, but for vms in the cloud like 512MB of RAM and a puny virtual CPU should be sufficient.