PauseAI.info website

I've been working on the https://PauseAI.info website since may, writing articles, doing the design and stuff.

We had a separate channel for this at #website-feedback but it's overkill now.

## How you can help

- Improve pages
- Write new pages
- fix issues
- add issues
- Improve design / accessibility etc.

Source code and issues on github: https://github.com/joepio/pauseai

We need to Pause AI

We are risking human extinction. We need to pause AI development, right now.

GitHub

GitHub - joepio/pauseai: Website for PauseAI.info

Website for PauseAI.info. Contribute to joepio/pauseai development by creating an account on GitHub.

awake-maroon•8/12/23, 6:31 PM

What about graphics?

Aawake-maroon What about graphics?

hurt-tomatoOP•8/12/23, 9:19 PM

Ideas welcome!

hurt-tomatoOP•8/23/23, 2:37 PM

Working on a decision tree for p(doom) calculation

hurt-tomatoOP•8/23/23, 2:42 PM

Still unsure about some things. Should I emphasize the p(doom) aspect? Seems to have a nice viral potential, but also makes the UX weird, as you'd have to go back and select other options to give it all of your own values.

Right now the probability is only calculated for your specific path, not for all paths. I could probably build something that calculates for all, which would result in p(doom | good) numbers. Seems fun, but also a bit difficult (perhaps I'm just a dummy though).

What PauseAI give its own numbers? I think we should, because there are good reasons to have certain beliefs, even though everyone will have different numbers.

hurt-tomatoOP•8/23/23, 2:47 PM

Also, how should I call it?

- p(doom) calculatorp(doom) calculator. Has a nice ring to it for the in-crowd. Requires a real p(doom)p(doom) calculation, though!
- AI outcome calculatorAI outcome calculator. Sounds lame, but describes it accurately
- AGI outcomesAGI outcomes. More succinct
- AI outcome treeAI outcome tree
- Am I an AI doomer?Am I an AI doomer?. More funny, but maybe a bit off-putting.

nursing-lime•8/23/23, 11:31 PM

I like the latter, it sounds intriguing to me. That said, I am the type of person who would probably click on something like "Am I a member of the outgroup?" out of curiosity.

hurt-tomatoOP•8/24/23, 5:18 PM

Another idea: maybe allow users to pick from a list of probabilities, get these numbers from a bunch of people, with their comments

sacred-rose•8/27/23, 3:13 AM

I really like the idea and everything in general, but do the answers change the probability numbers? I feel like the p(doom) at the end is pretty arbitrary and there's a couple of things that are bugged or I don't understand them. For example. if you put: no->yes or no->no it says you think you will be safe or you are a doomer but with 25% (less than 50) for each outcome.

sacred-rose•8/27/23, 3:14 AM

Maybe it would be better to let people choose the % to each question (not yes and no) and depending if they put less or more than 50% the different subsequent questions appear.

Ssacred-rose Maybe it would be better to let people choose the % to each question (not yes an...

hurt-tomatoOP•8/28/23, 2:31 PM

Interesting idea, can play with that a bit

skinny-azure•8/28/23, 7:55 PM

Idea that could really help reach new people:
On the /learn page on the top say:
Or checkout this page that links top resources in other languages than English (Dutch, ....)

skinny-azure•8/28/23, 7:57 PM

I think there are plenty of people who like it better or grasp it better in their own language.
First entry for Dutch:
Otto on rutger Bregman show

skinny-azure•8/28/23, 7:57 PM

Excited to be able to share that with people!

sacred-rose•8/30/23, 7:41 AM

@Joep Meindertsma personally, I'm mostly used to react as my go-to framework

sacred-rose•8/30/23, 7:41 AM

but I dabbled in svelte for a while in 2020 and it was pretty fun

sacred-rose•8/30/23, 7:41 AM

do you use react sometimes?

Ssacred-rose do you use react sometimes?

hurt-tomatoOP•8/30/23, 8:09 AM

Yeah I do, but I prefer svelte quite a bit now. I'm working on very large react + rust project (https://github.com/atomicdata-dev/atomic-server), and I think if I did the thing in svelte it would have been easier to and more fun to work on.

sacred-rose•8/30/23, 8:17 AM

that's cool, I'll need to refresh a bit before I start working - it's been a while since I worked on webdev, so I'm not familiar with vite or sveltekit

sacred-rose•8/30/23, 8:17 AM

once I do that, I might start to contribute, but I'm not sure how much I'll actually be able to do

sacred-rose•8/30/23, 8:18 AM

that's not because I don't have the skill, but I might not have the energy

hurt-tomatoOP•8/30/23, 8:24 AM

aight, help is very welcome

Spinning it up locally should be very easy, though. You can use bun / npm / pnpm, just use the

dev

dev

command and you're good to go

sacred-rose•8/30/23, 9:01 AM

I've already ran it locally and I'm browsing the code

hurt-tomatoOP•8/30/23, 9:02 AM

great!

sacred-rose•8/30/23, 9:02 AM

using VSCode connected to WSL with ubuntu, great environment

hurt-tomatoOP•8/30/23, 9:03 AM

wsl rocks indeed

wet-aqua•9/4/23, 7:28 AM

Hi! "You can win $4000 if you write a good draft!" sentence from here https://pauseai.info/action is outdated, since their competition has ended.

Wwet-aqua Hi! "You can win $4000 if you write a good draft!" sentence from here https://pa...

hurt-tomatoOP•9/4/23, 9:03 AM

Nice catch, thanks

sacred-rose•9/10/23, 11:46 AM

@Joep Meindertsma have you heard about Bun? https://www.youtube.com/watch?v=BsnCpESUEqM

YouTubeBun

Bun 1.0 is here

Bun 1.0 is here!

Bun is an all-in-one JavaScript runtime & toolkit designed for speed, complete with a bundler, test runner, and Node.js-compatible package manager.

https://bun.sh/

Ssacred-rose @Joep Meindertsma have you heard about Bun? https://www.youtube.com/watch?v=BsnC...

hurt-tomatoOP•9/10/23, 11:46 AM

Yeah have been using it on and off for a while! It's great

sacred-rose•9/10/23, 11:47 AM

I've just found out about it, looks like great environment for JS apps like the website

sacred-rose•9/10/23, 11:47 AM

I'm sure I'll be using from now on

hurt-tomato•9/18/23, 4:34 PM

Question: Why is the URL 'PauseAI.info,' and not just 'Pause.AI'?

hurt-tomato•9/18/23, 4:35 PM

I'm assuming it's because Pause.ai is too expensive?

Hhurt-tomato I'm assuming it's because Pause.ai is too expensive?

hurt-tomatoOP•9/18/23, 4:36 PM

Yep! Owner was a greedy one.

hurt-tomato•9/18/23, 4:39 PM

That sucks, haha. Decelerate.AI is just $180, I might pick that up next paycheck, just to have it handy.

awake-maroon•9/18/23, 9:32 PM

aifuture.monster cost me $3

hurt-tomatoOP•10/11/23, 11:59 AM

New page: http://pauseai.info/polls-and-surveys

Hhurt-tomato New page: http://pauseai.info/polls-and-surveys

hurt-tomato•10/12/23, 4:22 PM

https://wiki.aiimpacts.org/responses_to_ai/public_opinion_on_ai/surveys_of_public_opinion_on_ai/surveys_of_us_public_opinion_on_ai

hurt-tomato•10/12/23, 4:22 PM

We should add a few of these!

colossal-harlequin•10/21/23, 2:41 PM

Bug report: truncated sentence on https://pauseai.info/dangerous-capabilities

Bad actors can still build dangerous AIs. Evaluations are a post-training measure. By relying on them, we

Text currently ends there.

Ccolossal-harlequin Bug report: truncated sentence on https://pauseai.info/dangerous-capabilities > ...

hurt-tomatoOP•10/21/23, 5:38 PM

Thanks!

colossal-harlequin•10/21/23, 9:09 PM

By the way: are there notes anywhere re PauseAI history (branding, proposals, messaging and its wording, etc.)?
I see evolving and ongoing discussion, but feel I'm missing an origin story and records of key early decisions.

hurt-tomatoOP•10/21/23, 9:09 PM

Perhaps it's time for that yeah

hurt-tomatoOP•10/21/23, 9:09 PM

Well there's the wired article at least

colossal-harlequin•10/21/23, 9:16 PM

Sure. Maybe there's no top-level strategic thinking that isn't entirely straightforward and public. I just thought I'd find some discussion re why "pause", what policy to demand, how much to feature non-x risks, etc.

hurt-tomatoOP•10/21/23, 9:17 PM

Not sure I follow

hurt-tomatoOP•10/21/23, 9:23 PM

Would you like to see arguments as to why pausing is the right policy measure?

hurt-tomatoOP•10/21/23, 9:23 PM

Because in my view it's pretty clear that the risks themselves are large enough to draw the conclusion that building this technology without proof of safety is a bad idea

colossal-harlequin•10/21/23, 10:03 PM

The quotes were deliberate there - I imagined other verbs were considered.

I reckon necessarily simple messages and demands in complex domains beset by uncertainty are hard to do well. I don't mean we're not certain there's immense risk here and a need to mitigate - or that "X seems bad ...so decide not to do X" isn't a great premise. And I'm so glad you are doing stuff rather than indulging any analysis paralysis. But since both co-ordination and campaigning are difficult in their detail, I guessed there was some deliberation on them, and it might be hiding somewhere. (Edit: oh, archived ⁠website-feedback had escaped my attention. Will read.)

If I do have any specific "hey, top-level, did you consider?" thoughts now, is there a best channel for them?

hurt-tomatoOP•10/21/23, 10:40 PM

Any web feedback is very welcome here!

PauseAI.info website

Similar Threads

Similar Threads

Similar Threads