I am so grateful I live in a time with audio books, podcasts, YouTube, text-to-speach
JJ Hepburn
Karma: 688
Thank you for doing this Ben. Thank you for navigating a way through all of these issues that have prevented others from doing this.
I wonder how much my aversion to text based communication is holding me back.
This change also helps with text-to-speach
I have pretty mixed feelings about this. I greatly appreciate and respect the constrains you are operating under and the hard, often thankless work of those involved.
I have a lot of thoughts about this update and the funds in general based on my direct experience as an applicant, from being involved with a number of projects and applications and from what I have heard from others.
I might find the courage to say more about this publicly at some point but I did want to say that my experience with EA Funds, is a significant part of my decision to shut down AI Safety Support.
I’ve always been surprised that there is no fund you can donate to that is only for AI Alignment. You can either donate directly to an org or project or you can donate to a longtermist fund which is broader than just alignment.
I’ve tried to argue before that plenty of people are just not that cause neutral and would want to donate to a fund just for alignment. And now that alignment has gone much more mainstream it is even more important that we actually have a legible place for people to donate.
AI Safety has gone mainstream but most people in the world wouldn’t have a clue what “longtermism” is.
No decision from LTFF. Funding wouldn’t help at this point.
I still have plans to setup a research org in the UK
There’s no need to shut down the AI Alignment Slack.
Moderation hasn’t been too much of a burden, but it could be better. Would be good to find someone else to take ownership of this.
Shutting down AI Safety Support
You can now also subscribe to be automatically emailed when new events are added or updated. You can opt for either daily or weekly updates
Signup here:
Announcing aisafety.training
I was getting this error a week ago and still getting it today.
What do you think we do?
This is a bit of a side note on my last shortform.
There are (at least) three versions of what we or any project/org does.
What we think we do
What you think we do
What we actually do
Hopefully these three have a lot of over lap. Even 1 and 3 can have a lot of a gap between what you were wanting to deliver and what actually happens.
I occasionally like to ask people that I haven’t interacted with before what they think we do at AISS. I like this to be early in the conversation so that it is their raw belief. Though a little too often people actually start answering as if I’d ask what they think we should do.
On Naming Projects
My general rules or thoughts around naming projects is to avoid trying to give them a name that describes what the project is. The problem is mainly that you end up with lock-in and lose control over how your project is perceived.
People are very good at taking the name of something and using that to build a model of what they think the project is. We have this problem with AI Safety Support and the word “camp” in “AI Safety Camp” has caused confusion too. When talking to people about what we do we have to remove their beliefs in what we do before building up a more accurate picture.
OpenAI has big problems with this in that outsiders have their own model of what the founders were thinking when they came up with the name. They are often criticised for not living up to their name.
I’d prefer a mostly meaningless name where people have to ask what you do and you get to control the message. People start with a blank slate with no misconceptions that you have to deal with first. You get to build up the picture of what you do your own way. My default example of this is “Uber” which you would have a hard time coming up with your own picture of what they do. I also really like “Redwood Research” they obviously do research but I struggle to come up with an assumption of what that could be.
There are of course plenty of exceptions where a descriptive name is appropriate. This seems to be the default though and it is worth questioning. If you are struggling to come up with a name that you are really happy with and confident it describes your vision then consider a less descriptive name.
I love anything that helps to reduce administrative load, duplication of work and friction to get started. So pretty excited about this idea and the hiring agency Nonlinear are incubating.
Yes, I’d be happy to spend time talking to this person. Should you have a sign-up form or something?
The Application is not the Applicant
I would recommend using a custom domain instead of the proy.app subdomain. Even if you just have the custom domain redirect to the pory.app so that you can share the link around and if you change away from Pory then you control the domain and don’t have dead links everywhere.
… Yes, I have made this mistake.
An event for everyone is an event for no one.
Good question.
I have a few draft forum posts over the years that I don’t really have an interest in publishing.
I feel like I’ve found it pretty stressful each time I’ve posted or even commented. (yes, even this) After posting, I waste so much time checking the comments and changes in karma. I, naturally, feel every downvote or negative comment far more than upvotes or positive comments. Positive comments are actively discouraged (downvoted) here, which might make this worse.
Some of the most frustrating comments are ones that correct some minor details that, don’t impact the posts core point, or mention some exceptions as if they discount the whole idea.
I think this is a big part of why I don’t finish my draft posts. The drafts are short and just cover the main point I have, but I need to 10x my time, effort and length so that I address every obvious exception, edge case, rebuttal…
I think ultimately I find it to be a very negative feedback loop. Posting is stressful. Every single time. It’s not like I’m anxious before posting and then after I post I get a big hit of dopamine from the responses. This would provide some positive feed back loop like posting on social media.
My neurotransmitters take no account of expected value calculations.