Tax Deductible Donations
Many countries have tax benefits for making donations to non-profits. This tends to require that the non-profit is local, but there are ways around this.
Workarounds for countries
Some countries have orgs (like CAF in the UK, ) that can re-grant donations to international NGOs....
Chatlog Viewer
I created a standalone chatlog viewer. It allows you to view all your export logs, including HTML (file save as) backups from all chatbots in one location.
https://github.com/algorithmunmasked/ChatLog_Viewer
Features
...
Weekly tasks / MicroCommit
In the near future, we will have access to the https://microcommit.io/ platform, and be able to send weekly tasks for everyone who decides to follow us there.
This project is to give ideas, express interest in helping write the weekly tasks, and give any other type of feedback about them.
I already have a decent list. Ask for access to it if you want to participate in writing them: https://docs.google.com/document/d/1-Vh-HHiQHisqrmAiKagX4AGBxkAlRL-PivV-Iu452uA/edit?tab=t.s0o9w6xnniug#heading=h.xhoug3qml0bw...
Respond to the UK government's 'AI growth lab' regulatory sandbox consultation
The UK government is seeking to implement a cross-economy regulatory sandbox for the use of AI. The sandbox would "oversee the deployment of AI-enabled products and services that current regulation hinders" by "making targeted regulatory modifications under robust safeguards". These trial runs will then be used as the basis for permanent decisions on how AI will be regulated.
Importantly, the government is seeking input on which regulations should be âred linesâ, and points to "consumer protections, safety provisions, fundamental rights, workersâ protections, intellectual property rights" as specific examples that it intends to include.
A small team of PauseAI members could put together a response to this consultation with strong evidence and recommendations for specific areas that also calls for red lines and safety measures relevant to reducing existential risk. This would help justify and normalise those conversations for UK policy makers....
PauseAi should have adverts on radio, social media and television to spread our messenge.
Basically what the title says. We need to use every possible way to spread our messenge.
Antidote to situational awareness
Context: In June 2024, Leopold Aschenbrenner, a former OpenAI researcher, published an essay called "Situational Awareness" that called for a Manhattan Project to build artificial superintelligence. This essay deeply influenced the national security apparatus and the current administration (retweeted by Ivanka Trump), pushing the United States into a more pro-AI stance. While the majority of the essay - especially the first half analyzing AI research - is actually high quality, the section justifying the Manhattan Project is based on faulty assumptions and reaches wrong conclusions.
Recently, an essay of similar quality called "AI 2027" dropped and pushed in the opposite direction. Despite having similar reach to Situational Awareness, it failed to shift the national security debate because it only focused on the AI takeover scenario and didn't attempt to dismantle the race dynamics themselves.
What I want to do is create a new essay (target length 100-140 pages) of similar quality and style with the main message that creating artificial superintelligence will end in disaster in ANY scenario (both controlled and uncontrolled end in mass extermination and worthy successor is impossible with artificial intelligence)...
Google Mail & Groups - shared emails
We're using the Google Suite, and we have a free NGO license (unlimited members!)
Emails
- You need to be a volunteer (registered in the airtable, and signed the volunteer agreement!)...
International content pipeline
We have a lot of national groups, which is awesome. Many of them have their own website, their own socials, their own flyers... I suspect there is a lot of work duplicated, and a lot of useful materials remain unused / unknown. We should work on preventing this.
What have we already got
- A figma template...
Pause AI Day
It would be great to establish one day as a Pause AI Day (preferably the first Saturday or Sunday of March - March 7/8 for 2026)* to have an annual globally recognized day to:
1. Pressure Institutional and Country Accountability - pressuring AI companies to shutdown even for just a day to do safety protocols, updates, tests to ensure safe AI, and for countries to make, revise, and pass legislations on AI safety.
2. Increase public awareness - increase petition signatures, get more social media attetntion, and have more people knowledgeable about AI safety to join the movement.
3. Establish Cultural Shift - encourage community engagement and position AI safety as a shared responsibility.
...
My PauseAI: member volunteer hub / portal / dashboard
Would be nice if there was a place where you can go to log in with your PauseAI account and see all sorts of relevant stuff:
For members
- Achievements / goals / gamification #Gamify PauseAI activism
- My national chapter info (contact details, membership list, policy goals)...
Venom AI(degrade AI web crawlers)
AI web crawlers take information from creators and re-distribute them without paying the original creator, I want to create code that makes it difficult for crawlers to interpret the content but doesn't interfere with humans reading that.
The way to do that is to use 'hidden letters' i.e. keep font size at 1 and color as white, no human can possibly read that, now add letters that transform the meaning of word by adding few letters, for example 'for' becomes 'form' and then add some noise to it making it 'foscrm' now when AI crawls it, it struggles to find a meaning, but for humans letters 's', 'c' and 'm' are invisible if we do it with every word in a website, specially numbers for example 72% becomes 1724% now Ai can never guess if it is 72/17/24/17.24 or so on, but to humans 1 and 4 are not visible so that works out. I have the architecture of this ready, but I am facing two problems
1) Many websites don't allow to keep source formatting so that makes use case limited....
Indexing this Discord server on the web
One of the biggest downsides of Discored compared to websites / forums is that its content cannot be found by search engines. So if people google "PauseAI gamififcation" they won't find our discord topic in here.
Now there is a solution for this problem, called AnswerOverflow. I think this will be huge pretty soon.
If we do this as PauseAI, it will have pretty big implications:...
New Merchandiser
Hi everyone
Looking for a single person to lead a project to assess our current merchandiser and store, and find some options for PauseAI Global that holds or lowers the price for the end buyer while increasing the margin for the donation PauseAI receives.
This project is limited in scope....
đ Project: âARC OS Audit Layerâ â make GPT reasoning human-readable
đ Project: âARC OS Audit Layerâ â make GPT reasoning human-readable
1ïžâŁ What we want & why it matters
PauseAI works on policy / risk. We need deterministic traces, not opaque ChatGPT text.
ARC OS converts any prompt into a logic tree + self-check flags (bias, loops, conflicts)....
PauseAI works on policy / risk. We need deterministic traces, not opaque ChatGPT text.
ARC OS converts any prompt into a logic tree + self-check flags (bias, loops, conflicts)....
TakeOverBench
I've been thinking about making a TakeOverBench - a benchmark that tests for capabilities relevant for takeover.
Now, we (PauseAI + ERO) are developing a website that combines existing benchmarks and links them to concrete takeover threat models.
(in this past, this topic was about an EU tender, which explains most of the first comments)...
PauseCon Brussels
@DidierCoe is planning on organizing a PauseCon event in Brussels! It's one of the political capitals of the world, housing EU parliament + commission. It's a great place to protest - one of our first protests was there as well.
Didier is open to helping out with funding, and he's also experienced in organizing conferences in Brussels. He's already got an idea for location, too: a library next to Pi, de la Monnaie with nice big & small rooms. He can get a discount, maybe 3 to 5k for 2 to 3 days. Between 60 or 100 participants would work there.
No date in mind as of now, target Q4 2025 or Q1 2026....
Seeking microgrant support for multi-agent manipulation research
While plenty of research exists on AI persuasion of humans, thereâs less work on understanding LLM-to-LLM persuasion in multi-agent scenarios, which is increasingly critical as AI teams begin to collaborate on complex tasks. Not only that, most existing research focuses on epistemological manipulation: persuasion of beliefs, not necessarily actions. Arguably, the latter is not only more indicative of true values held by someone, but is also far more dangerous in the hands of malicious or unaligned agents.
Our proposed benchmark evaluates both manipulative capability and goal retention by having one LLM attempt to manipulate another LLM into suboptimal choices in a multi-armed bandit game. The deceptive agent knows the true reward distributions but must convince the playing agent to choose poorly without revealing its manipulative intent, while the playing agent must navigate between potentially helpful advice and harmful deception.
I've done similarly flavored benchmark work before, and would estimate a cost of around 750-1500 EUR in API costs for a project like this depending on how many models we test. We already have a working prototype despite having worked on it for less than a week. Ran it on Llama (free) and Claude 3.5 Haiku (very cheap), but we want to scale it to bigger models to get a more realistic idea of SOTA capabilities in this area....
Ads for Yudkowsky's Book
MIRI needs help designing ads for Yudkowsky/Soares's new book. To be displayed in the DC metro, the NY subway, and on an SF billboard.
This is an incredibly high impact opportunity!
Ask your designer acquaintances if they can help, too. There's cash prises to win, by the way, up to $1000. Deadline is August 10th....
N8N Automations
We've migrated away from Zapier, and now use a self-hosted version of N8N. This is less costly and far more powerful.
- https://n8n.pauseai.info/
- Hosted on DigitalOcean
- Managed by Joep...
