We would like to build a new landing page for https://impactmarkets.io that is really sleek and is consistent with the rest of the marketplace (currently at https://app.impactmarkets.io).

Advocacy tools for effective giving:

• Social media advocacy post generator: When a member signs a GWWC pledge they write their motivation -  it'd be helpful to have a tool which then suggests a social media post they could write based on their motivation, previous members motivations (see study), and previous social posts (see examples). Potentially train an AI language model that suggests two options and the one they click "copy/customise" is a "vote" that's fed back into the model as a correct prediction?
• Social media image generator: Similar to above, once someone makes a donation, create suggestions of images with generated text on them that they can use to share that "Ahah!" moment with others. (See where people are donating here and the options on the donation platform here).
• Any other ideas to increase advocacy!

Alignment Ecosystem Development maintains a list of projects which volunteer-scale efforts could shift the needle on, and hosts monthly calls where people can pitch or join projects. Maybe someone could drop by one of those calls and see if anything fits the bill for this?

Interactive consensus aggregator for squiggle estimates

If analysts Alice and Bob each write cost-effectiveness analysis of charity C, then donor Eve ought to be able to input relative trust quantities informing how to weight Alice's estimate against Bob's. In other words, if Eve thinks Alice is twice as trustworthy or competent as Bob, then the MVP would return the squiggle string mixture(alice, bob, [2, 1]). 

Interactive worldview substituter for squiggle estimates

Building on the above, it would be nice if Alice and Bob had agreed to use a set of input variables, such as background quantities like the state of a manifold market on some ML benchmark or a global count of malaria cases. This set of input variables can also be a worldview, or quantities which upon inspection are also squiggle estimates. Then, it would be nice to be able to trivially substitute these input worldviews, if you want to look at how you expect your cost-effectiveness analysis to change over time or if you think Bob has a more calibrated background worldview but prefer Alice's fermstimate of charity C's impact. 

There's an aggregated list of AI safety research projects available on AI Safety Ideas (forum post) and though it's a bit messy in there at the moment, it should be quite high quality leads for a hackathon as well! E.g. Neel Nanda and I will add a bunch of project ideas to the Interpretability Hackathon list during the next couple of days. 

(Copied from my post on your shortform, relating to Quinn's idea, Squiggle, and the tptool Quinn has worked on, and as discussed with Nicole)

Joint hackathon:

Part 1: Creating user-facing prototype for Quinn's Cost-effectiveness-analysis $\times$ Montecarlo comparison platform

Part 2: Bringing in statistics/quant/domain people to- test it and do independent evaluations (in teams or individually)

Assessing part 2 allows us ... compare and test both

• the platform
• and the consistency  of these Montecarlo BOTEC things

Bring in Givewell people and others too?

See some note-taking on this here (If you want access to the Gitbook to add your own content let me know)

Contribute to Wikiciv.org - A wiki for rebuilding civilization's technology
 

Ways you can help:

-Write and edit articles
-Research and collect content
-Work on a port of Entitree to make a tech tree visualization

No coding experience needed! Wikiciv has a "What You See Is What You Get" editor, if you can edit a google doc, you can edit Wikiciv.

Some approaches to solving alignment go through teaching ML systems about alignment and getting research assistance from them.^[1] Training ML systems needs data, but we might not have enough alignment research to sufficiently fine tune our models, and we might miss out on many concepts which have not been written up. Furthermore, training on the final outputs (AF posts, papers, etc) might be less good at capturing the thought processes which go into hashing out an idea or poking holes in proposals which would be the most useful for a research assistant to be skilled at.

It might be significantly beneficial to capture many of the conversations between researchers, and use them to expand our dataset of alignment content to train models on. Additionally, some researchers may be fine with having their some of their conversations available to the public, in case people want to do a deep dive into their models and research approaches.

The two parts of the system which I'm currently imagining addressing this are:

1. An email where audio files can be sent, automatically run through Whsiper, and added to the alignment dataset github.
2. Clear instructions for setting up a tool which captures audio from calls automatically (either a general tool or platform-specific advice), and makes it as easy as possible to send the right calls to the dataset platform.^[2]

1. ^{^}

   Ought's Elicit is the prime example.

2. ^{^}

   I hear OBS might be a good tool for this.

I couldn't attend the interpretability hackathon and was hoping to get acquainted with LLM interpretability research as a sofware dev with no experience in interpretability or transformers. So here's a starting point following in the footsteps of this submission (see their writeup here):

• https://github.com/neelnanda-io/Easy-Transformer
• Neel Nanda live coding/testing transformer hypotheses learning (1.5hrs)  https://www.youtube.com/watch?v=yo4QvDn-vsU&t=3403s
• Go through Transformer theory:
  • Attention is all you need paper: https://arxiv.org/abs/1706.03762
  • Good summary: https://nostalgebraist.tumblr.com/post/185326092369/the-transformer-explained
  • Positional embeddings explained (9mins): https://www.youtube.com/watch?v=1biZfFLPRSY

Basically I am thinking we can use the hackathon as a collaborative study session to become more familiar with transformers and interpretability, ultimately culminating in replicating the results in the linked submission (it took them 3 days but since we have a starting point, possibly we can replicate their project and grok what they did much quicker).

Not shoehorned to this idea though. If you think there is a better avenue to using the hackathon to upskill in LLM interpretability and transformers, do share.

Web app implementing giving pledges based on net income (after taxes and benefits)

The Giving What We Can Pledge is 10% of pre-tax income, if giving is tax-deductible, or 10% of post-tax income otherwise. We've scoped a different pledge, in which one gives the amount such that their net income after taxes and benefits falls 10%. You can also think of this as transferring 10% of your consumption to effective charity.

We propose building a web app where people enter their household information and it suggests the amount to give. The app would call the free, open source PolicyEngine API to compute taxes and benefits (I am the Executive Director of PolicyEngine, a nonprofit). I think in a day we can build a simple US prototype that collects a handful of data points (earnings, family size, state) and displays the computation.

I've discussed this idea with Luke Freeman, Executive Director of Giving What We Can, who supported experimenting with it.

New features and improvements to open source eCard generator for EA donations. The goal is to get people to donate to EA causes as gifts. There are a lot of ways the site can be improved and it would be great to have lots of ideas!

Repo: https://github.com/TLYCS/Card_Generator

Current Site: https://main.d33ee03vlquk8r.amplifyapp.com/

A website to crowdsource research for Impact List

Impact List is building up a database of philanthropic donations from wealthy individuals, as a step in ranking the top ~1000 people by positive impact via donations. We're also building a database of info on the effectiveness of various charities.

It would be great if a volunteer could build a website with the following properties:

-It contains pages for each donor, and for each organization that is the target of donations.
-Pages for donors list every donation they've ever made, with the date, target organization, amount, and any evidence that this donation actually happened.
-Pages for target organizations contain some estimate of each component of the ITN framework, as well as evidence for each of these components.
-There is a web form that allows any Internet user to easily submit a new entry into either of these data sources, which can then be reviewed/approved by the operators of Impact List based on the evidence submitted. 

The uncertain bibliography

A latex plugin for annotating a .bib file with credences, confidence intervals, squiggle strings, replication probabilities. 

[off the top of my head] Some browser extension for Swapcard that lets you shuffle the list of people. This is premised on the un-checked-by-me assumption that Swapcard has some algorithm for the order in which it displays profiles to you, which might mean some people randomly get 'boosted' relative to others (e.g. if they are listed earlier in the database)

From a tech event we had recently - "The main request that came out of the meetup was a way to match up potential mentors with mentees"

Maybe something using the current EA Forum user search function but with a redesigned interface for connecting.

Hi all!

New-ish SWE dev here. My group and I would love to partake in the Hackathon but are struggling with thinking about ideas for feasible projects that could be completed in 7 hours. Is there a public list of previous projects we could reference? Or a list of projects maybe EAG thinks would be cool to do? Thanks.