Un-untitled. @antinegationism - Tumblr Blog

Update

Siikr's joke-text has been indefinitely replaced by serious-text.

It will remain this way until enough people are fighting about it.

Please do your part.

siikr DNS expired

Is Google strapped for cash or something? Apparently they sold off their DNS business and customers to Squarespace.

Have you tried o4-o5 yet? Seeing a lot of knowledgeable short timelines skeptics being shocked, saying it's insightful for automated AI research, and generally updating hard toward Kokotaljo-style views.

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.

siikr down :(

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

You can use https://siikr.yggdrasil.cat/index_local.php in the meantime, but it might require either waiting for your blog to transfer over from one of the other nodes or a reindex if your blog was on the node that went down.

A common phrase in English is "No ifs, ands, or buts".

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

I therefore petition that all programming languages adopt but as the common terminology for error or exception.

argumate

wondering how to use A* search with gradient descent given that it requires a heuristic of how far a partial solution is from the goal 🤔

argumate

an admissible heuristic, which is to say an underestimate of the distance remaining to the goal, where the more accurate the heuristic the better the search performs, but the search breaks if the heuristic ever overestimates because that can cause it to neglect better solutions.

the cool thing about optimisation by gradient descent is that it tells you exactly which direction to go in order to reach a goal whose exact location is unknown, while A* search tells you which direction to go to reach a goal whose exact location is known but cannot necessarily be reached by heading straight towards it.

then genetic optimisation can be applied when you have an unknown goal and a nondifferentiable cost function, so you don't know what you're looking for nor which direction it must be in, which is obviously not a great situation to be in.

argumate

gradient descent has always kind of bugged me for all but the simplest problems, because not everything is differentiable and back propagation is annoying and numerical futzing around is unsatisfying but the most frustrating part is that it works!

but most neural network advances seem to come from running this insanely powerful optimisation process on a network architecture that was dreamed up by a shaman behind a waterfall operating on mushrooms and intuition; I want the computers to be doing this part of the job too, but you can't differentiate over all possible networks, can you... can you?

antinegationism

You cannot. You may convince yourself of this by simply getting too creative in your choice of activation function.

argumate

then again I guess why limit yourself to networks, they're just one slice of the space of possible functions after all.

antinegationism

You tryna start a fight rn or something?

argumate

wondering how to use A* search with gradient descent given that it requires a heuristic of how far a partial solution is from the goal 🤔

argumate

an admissible heuristic, which is to say an underestimate of the distance remaining to the goal, where the more accurate the heuristic the better the search performs, but the search breaks if the heuristic ever overestimates because that can cause it to neglect better solutions.

the cool thing about optimisation by gradient descent is that it tells you exactly which direction to go in order to reach a goal whose exact location is unknown, while A* search tells you which direction to go to reach a goal whose exact location is known but cannot necessarily be reached by heading straight towards it.

then genetic optimisation can be applied when you have an unknown goal and a nondifferentiable cost function, so you don't know what you're looking for nor which direction it must be in, which is obviously not a great situation to be in.

argumate

gradient descent has always kind of bugged me for all but the simplest problems, because not everything is differentiable and back propagation is annoying and numerical futzing around is unsatisfying but the most frustrating part is that it works!

but most neural network advances seem to come from running this insanely powerful optimisation process on a network architecture that was dreamed up by a shaman behind a waterfall operating on mushrooms and intuition; I want the computers to be doing this part of the job too, but you can't differentiate over all possible networks, can you... can you?

antinegationism

You cannot. You may convince yourself of this by simply getting too creative in your choice of activation function.

Did you know?

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.

4. Literally just funny blurbs is ideal though like, why are you even considering 2 and 3 rn? Seriously.

#siikr

someone's stealing your Twitter shower thoughts

Good.

vren-diagram

The guy who did the first designer babies is on Twitter and he constantly posts stuff like this.

phaeton-flier

Sad to hear the guy who did the first designer babies also has bad opinions

vren-diagram

eh,

'Bioethicist' views:

• 18% — paying organ donors is OK

• 63% — all extensions of life equally good regardless of length

• 66% — someone's life being worth living gives no reason to create them

• 40% — being blind is only a disadvantage because society is unjustly designed

https://www.apollosurveys.org/bioethics/

phaeton-flier

I think the history of biological research means you gotta be a little more specific when you're talking about what restrictions are or aren't justified; "You should be allowed to genetically edit embryos to pick out traits" is very different from "we shouldn't need informed consent for medical testing". Sucks, but it's not like it's hard to say what things in particular you think should change!

antinegationism

Wait hang on what?

"Being blind is only a disadvantage because society is unjustly designed" has 40% popularity among Bioethicists?

Like, I understand the sentiment, and think it would be totally excusable for anyone without "bio" in the name of their discipline, and even with "bio" in the name I could still see 40% agreeing if the statement held something like "Being blind would be less of a disadvantage if society were more thoughtfully designed", but like

Eyesight specifically is the quintessential example of an adaptation so advantageous that it has separately come about in multiple forms in multiple species. Are bioethicists just regular ethicists that added bio to their name or is there some next level take I'm missing here?

can't search any new blog :(

So, weirdly enough everything seems to be working fine now even though I didn't do anything aside from set the hub server to maintenance mode. I was vaguely suspecting a daylight savings time issue causing some coordination problem between the hub and the spokes, but then I remembered this present ask from a month ago. Are the blogs you were attempting to search still unsearchable?

I can't siikr some blogs (including your own and @multiheaded1793's)

Yeah it's been unreliable for me too. I think one of the nodes is having a hard time while pretending it's fine. Looking into it.

dumb clarification q on the "dedicated" part of "Get a dedicated Ubuntu Server instance up" in the siikr hosting instructions: does this mean I shouldn't use an Ubuntu server I'm primarily using for other stuff (but has leftover capacity for Altruism) to host Siikr?

You can use an existing Ubuntu server, it's just that it will be more of a pain to set up / require you to have some idea as to what you're doing. The main stumbling block you're likely to encounter is that the setup script presumes you're using nginx and that it's fine to install particular versions of postgresql and muck postgresql user permissions. If running a dedicated server instance on a VM or something isn't an option, and you anticipate conflicts with your existing stack, then I suggest treating the setup script as a guide instead of a thing to literally execute. Feel free to DM if you run into trouble, I will apparently be doing maintenance today anyway.

Thoughts on "Utility Engineering: Analyzing and Controlling Emergent Value Systems in AIs"?

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.

A scenario, part 3.

Do you condemn the Zizians

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.

The Zizian opinion on this matter, as I understand it, seems to differ from mine. I suppose we will have to agree to disagree. Presumably, from different sides of a jail cell.

found siikr today and saw the self-hosting instructions and i was wondering, how much disk/mem/bw is required to contribute meaningfully? i have a couple of small vms in the cloud, and plenty of compute i could set up on slow home net, but not sure if either of those setups is worthwhile.

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

The main thing is a persistent domain name and disk space. The more space the merrier.

Are you a Zizian

I am not and for the last time stop trying to recruit me.

Trending Blogs

Recently Viewed Blogs

Un-untitled.