ozziegooen

I'm currently researching forecasting and epistemics as part of the Quantified Uncertainty Research Institute.

Sequences

Beyond Questions & Answers
Squiggle
Prediction-Driven Collaborative Reasoning Systems

Wiki Contributions

Comments

Happy to see experimentation here. Some quick thoughts:

  • The "Column" looked a lot to me like a garbage can at first. I like the "+" in Slack for this purpose, that could be good.
  • Checkmark makes me think "agree", not "verified". Maybe a badge or something?
  • "Support" and "Agreement" seem very similar to me?
  • While it's a different theme, I'm in favor of using popular icons where possible. My guess is that these will make it more accessible. I like the eyes you use, in part because are close to the icon. I also like:
    • 🚀 or 🎉 -> This is a big accomplishment. 
    • 🙏 -> Thanks for doing this.
    • 😮 -> This is surprising / interesting. 
  • It could be kind of neat to later celebrate great rationalist things by having custom icons for them, to represent when a post reminds people of their work in some way. 
  • I like that it shows who reacted what, that makes a big deal to me. 

I liked this a lot, thanks for sharing.

Here's one disagreement/uncertainty I have on some of it:

Both of the "What failure looks like" posts (yours and Pauls) posts present failures that essentially seem like coordination, intelligence, and oversight failures. I think it's very possible (maybe 30-46%+?) that pre-TAI AI systems will effectively solve the required coordination and intelligence issues. 

For example, I could easily imagine worlds where AI-enhanced epistemic environment make low-risk solutions crystal clear to key decision-makers.

In general, the combination of AI plus epistemics, pre-TAI, seems very high-variance to me. It could go very positively, or very poorly. 

This consideration isn't enough to change p(doom) under 10%, but I'm probably be closer to 50% than you would be. (Right now, maybe 40% or so).

That said, this really isn't a big difference, it's less than one order of magnitude. 

Quick update: 


Immersed now supports a BETA for "USB Mode". I just tried it with one cable, and it worked really well, until it cut out a few minutes in. I'm getting a different USB-C cable that they recommend. In general I'm optimistic.

(That said, there are of course better headsets/setups that are coming out, too)

https://immersed.zendesk.com/hc/en-us/articles/14823473330957-USB-C-Mode-BETA-

Happy to see discussion like this. I've previously written a small bit defending AI friends, on Facebook. There was some related comments there.

I think my main takeaway is "AI friends/romantic partners" are some seriously powerful shit. I expect we'll see some really positive uses and also some really detrimental ones. I'd naively assume that, like with other innovations, some communities/groups will be much better at dealing with them than others.

Related, research to help encourage the positive sides seems pretty interesting to me. 

Maybe we can refer to these systems as cybernetic or cyborg rubber ducking? :)

Yea; that's not a feature that exists yet. 

Thanks for the feedback!

Not yet. There are a few different ways of specifying the distribution, but we don't yet have options for doing from the 25th&75th percentiles. It would be nice to do eventually. (Might be very doable to add in a PR, for a fairly motivated person). 
https://www.squiggle-language.com/docs/Api/Dist#normal

You can type in, normal({p5: 10, p95:30}). It should later be possible to say normal({p25: 10, p75:30}).

Separately; when you say "25, 50, 75 percentiles"; do you mean all at once? This would be an overspecification; you only need two points. Also; would you want this to work for normal/lognormal distributions, or anything else? 

Mostly. The core math bits of Guesstimate were a fairly thin layer on Math.js. Squiggle has replaced much of the MathJS reliance with custom code (custom interpreter + parser, extra distribution functionality). 

If things go well, I think it would make sense to later bring Squiggle in as the main language for Guesstimate models. This would be a breaking change, and quite a bit of work, but would make Guesstimate much more powerful. 

Really nice to see this. I broadly agree. I've been concerned with boards for a while.

I think that "mediocre boards" are one of the greatest weaknesses of EA right now. We have tons of small organizations, and I suspect that most of these have mediocre or fairly ineffective boards. This is one of the main reasons I don't like the pattern of us making lots of tiny orgs; because we have to set up yet one more board for each one, and good board members are in short supply.

I'd like to see more thinking here. Maybe we could really come up with alternative structures. 

For example, I've been thinking of something like "good defaults" as a rule of thumb for orgs that get a lot of EA funding.
- They choose an effective majority of board members from a special pool of people who have special training and are well trusted by key EA funders.
- There's a "board service" organization that's paid to manage the processes of boards. This service would arrange meetings, make sure that a bunch of standards are getting fulfilled, and would have the infrastructure in place to recruit new EDs when needed. These services can be paid by the organization.

Basically, I'd want to see us treat small nonprofits as sub-units of a smoothly-working bureaucracy or departments in a company. This would involve a lot of standardization and control. Obviously this could backfire a lot if the controlling groups ever do a bad job; but (1) if the funders go bad, things might be lost anyway, and (2), I think the expected harm of this could well be less than the expected benefit.

Load More