Subtitle: I’m launching a project & would like your feedback.
TLDR:
AI safety comms efforts towards the general public are still very neglected, compared to the scale and urgency of the current situation. I’m proposing a possible, quite ambitious solution to this problem and I need your help to both understand if and how it can actually work, and to make it happen.
The recent OpenAI/Hugging face incident and the following METR report ring all the alarm bells. Despite this, most people worldwide don’t know that we might be “50% of the way to full-blown AI takeover” , and communications efforts towards the general public are not happening nearly at the scale they should. There was very little mainstream media coverage of the new crucial insights from the METR report, despite their extreme importance. I expect a similar silence regarding the even more recent details about OpenAI’s training their new models to be ”more capable but less understandable”.
I think that if we want to actually “pace the frontier” (or even completely pause AI development for a while), we cannot rely mostly on informing policymakers. As things stand, I believe we also need to trigger significant pressure from the public, yesterday.
Since the world is in multicrisis-mode right now (geopolitical instability, the decline of western democracy, general anxiety on AIs non-catastrophic harms, incoming US elections, climate change, etc), decisionmakers’ priorities may be quite unpredictable. That means, informing as much of the general public as possible on AI’s catastrophic risks may have become a non-negotiable strategy.
So here’s my high-upside, but still untested idea:
Creating a platform to inform and keep in the loop tens of thousands of content creators and journalists, so that they can in turn spread the word to all of their audiences worldwide – aka, millions of people – hopefully reaching a critical minority necessary to push and achieve the national and global policies we need to make AI go well.
I believe this can be done
AI safety researchers and organisations on the other
This is what I’ve been working towards lately, building a platform called “Tell The World”. Here’s a link to the landing page: telltheworld.io.
With this blog post I would like to tell you more about it and ask for your help to make this happen.
The few content creators and journalists already covering AI safety topics suffer from significant research bottlenecks: the most important sources – reports, papers, blog posts, social media, etc, are
There’s also a great number of content creators and journalists worldwide who could be covering these topics, but still don’t know (or know too little) about them. In case they decided to make content or write about them, they’d incur in the same problems AGI-pilled content creators and journalists have to deal with (see above).
On the other hand, more and more AI safety researchers and organisations want to get their message across to the general public. Some are already trying, but the efforts are still (IMO) too scarce and scattered. Others are not prioritising this at all.
***
Disclaimer: from now one I’ll be using these abbreviations:
***
TTW’s ambition is to address these problems by
… then use their storytelling and creative skills & experience to communicate what they’ve learned, in a way that is optimal to their target audience.
And in the process, become part of a community of people working towards the same goal - making AI go well for all of humanity.
In short, I imagine the potential impact to look something like this:
More specifically, Tell The World would provide all the tools and services to facilitate the interactions between Rs and Os with CCs and Js:
The features below represent the current MVP, based on user research with content creators (and other comms people), as well as my own experience of the AI safety space, over the past 2 years.
I still haven’t done user research with researchers, organizations, or journalists. So this first iteration mostly targets creators’ needs, while I gather broader feedback.
Briefs are interactive, expert-curated, authoritative pages providing information and material on specific AI safety topics – both evergreen and new – in plain English. As you can imagine, Briefs are aimed at solving content creators’ and journalists’ research bottlenecks.
Here’s a short video-walkthrough giving a little taste of what Briefs look like right now: video link
More specifically, I imagine Briefs to be:
What would a Brief contain?
The content structure is not final, but so far includes:
..and much more
Here are a couple of screenshots from the current Brief page (the content is a mix of mock and real data). The first image shows the Quotes section, the second show a question from the Q&A section. For more, take a look at the video walkthrough.
In my mind, Versions 2 and 3 would most likely include the following:
First and foremost, I need to understand to what extent this project idea is actually feasible. As mentioned, I still need to do a bunch of user research work to figure out what to build – whether to improve or discard what’s already there, as well as decide what features to work on next.
To that end, I would greatly benefit from
Higher profile (but probably very busy) people include Ajeya Cotra, Daniel Kokotajlo, Peter Wildeford, Steven Adler, Connor Leahy, and so on.
How are you going to find your first users?
Publishing this blog post ( :D ), reaching out to CCs, Js, Rs, and Os in the AI safety space, plus asking them who I should talk to next. I plan to be networking a lot, trying to meet as many relevant people as possible.
Will content creators and journalists outside the AI safety space be magically convinced of catastrophic risk by just reading Briefs and seeing what’s going on on the platform?
I don’t know, we’ll have to see.
Maybe some will, by realising how many smart, competent people are taking this situation seriously. That’s kinda what happened to me, back at the 2022 EAGx in Rotterdam: there, I heard about AI risk for the first time, and believed the claims on the spot because 1. They were being presented by a credible figure (a PhD researching these topics), 2. There were 30-50 people in the room taking the arguments very seriously.
Maybe some won’t, but they might just need a bigger push to jump on the AI safety bandwagon – be it a short intro course, repeated exposure through a newsletter, seeing their peers creating videos on these topics and getting millions of views, etc. This feels like a solvable problem.
Won’t creators and journalists who are not AGI-pilled use this platform in a damaging way? For example consuming the platform materials, but distorting the message?
That’s possible. There should probably be some monitoring of the AI-risk content created by CC and J users. I imagine that if someone willingly distorted the message (instead of merely misunderstanding), such a user would get a warning or be suspended, perhaps. Haven’t thought too much about this yet.
Aren’t technical people and orgs going to contribute to the platform using a language that’s not easily understood by content creators and journalists from outside the AI safety bubble?
Hopefully not.They’ll be encouraged to write in clear terms, avoiding EA/Rationalist-speak & technical jargon. Some basic automated guardrails are already in place, if anything obvious slips. More advanced features (e.g.: an AI filter checking for unclear language and suggesting more approachable alternatives) may be added to future versions, probably whenever there’ll be a budget for the necessary AI tokens.
How far are you with building this app? Can we use it soon?
Quite far with the MVP. Briefs and all connected functionality are ready for use. Profiles, login, admin features are pretty much ready too.
Do you already have funding for this?
Not yet. I thought about doing this launch first, gathering initial feedback and maybe some early users / testers, to better understand what kind of funding I actually need.
Why the cheesy name?
It describes well what the mission is. I should also mention that there’s no “AI” in the name because the same software infrastructure can quite easily be used to communicate other important-but-too-technical overlooked topics, such as biorisk, climate, etc.
Thanks for reading!
https://www.lesswrong.com/posts/HBbd5vnZGar3BDX4Y/science-and-news-twitter-x-summarizer
might want to check this out. I also have been learning social media over the last 2 months and went to pdku so can give you some thoughts and help you network if you want to shoot me a dm.