I was chatting with a friend about AI field building projects, but we were frustrated by the fact that we don't know what project ideas already exist and have weak feedback networks, and how inefficient this seems. Thus, I/we would like to know: is there some kind of platform for proposing and discussing project ideas (or finding existing resources)—similar to Impact CoLabs but which is more decentralized, fast-paced, and discussion-oriented, while also more focused than the overall EA forum (and other forums)?[1] 

As a recursive example, it would be nice to know if a project/resource like this already existed! As a less recursive example, I was discussing the problem of the external credibility of some AI safety/impact literature, and figured it could be great to see whether AI is able to change the language of posts to make them more formal, so that they are better suited to some audiences.

It seems like it would be valuable if there some kind of voting or similar karma system.

Also, I think it's likely[2] that now or in the near future, large language models will be able to reduce the search costs that could undermine the value of this site (the basic idea being "people often use different terms to describe the same concept and current search functions are not always great at recognizing this, so someone may struggle with false negatives (not finding similar projects that do exist), whereas it seems LLMs could mitigate this problem").

But I'll withhold further discussion of its potential features until getting some responses on the following questions:

  • Does something like this exist?
    • If something like this does exist, why have I and others not heard of it (this is slightly a rhetorical question, but I do genuinely want to know if there's just some newsletter or resource for sharing these things that I ought to know about but don't);
    • If nothing like this exists, should something like this exist?
  1. ^

    One particular concern I have with trying to rely on the EA Forum and associated forums (e.g., LW, AIAF) is that it seems harder to create focal points or shared standards: tags may help, but it seems like it could easily become disorganized or inconsistent among everything else, especially compared to having a separate website or spreadsheet (although maybe these could just query the EA Forum/AIAF/LW). But then again, perhaps this is less of an issue than I expect, and the optimal solution probably would just be "discuss it on the forum(s)."

  2. ^

    Low confidence, intuitive impression but I realize I'm biased.

7

0
0

Reactions

0
0
New Answer
New Comment

1 Answers sorted by

I have seen aisafety.world but not the other two, and those seem relevant (although I only just looked at them for a few minutes). Thanks for the suggestions! (One project idea: survey people to figure out how widely known these are, then figure out how to fix the situation if familiarity is low)