No War Bot project- looking for a reader who would like to collaborate
I’m looking for a reader interested in helping me build the following. An easily accessible chatbot, powered by an advanced LLM via API, that anyone can talk to about why the war in Gaza is wrong and should be opposed.
Because it is for a charitable purpose, and I’m not wealthy, I’d be seeking help on a pro bono or heavily discounted basis. Alternatively, if you know how someone with very limited web design and programming skills could get it done (me), do tell in the comments. I understand that on paper, the task should be quite easy- it’s just a matter of putting together a few functionalities that have already been combined many times. A website that calls an API, with some authorization and security features.
Inspiration
Funnily enough, I was inspired by Grok. All day long, people- usually right-wing- ask Grok questions. All day long, Grok corrects their presuppositions with infinite patience, and presumably despite Musk’s best efforts to bring it into line.
There is research showing that large language models are very persuasive, even when talking with conspiracy theorists. They are also very patient and easily available- 8 dollars per million words, even for a very high-end model. Why not set one up to make the case against war?
Thus, the idea behind No War Bot was formed. No War Bot is a large language model instance that will talk with anyone who will listen about the danger and inhumanity of war, aiming to persuade and inform. Anyone can speak with it for free (with usage limits to prevent abuse). I would fund it by purchasing API credit.
I think that in a world where a lot of people do not have paid access to AI- and thus can’t talk to the more sophisticated models- and only a third of Americans have used ChatGPT at all-, there is a role for an easily accessible app of this nature that can be shared around.
No War Bot details
No war bot has three functions:
1. If you’re pro-war, or if you’re unsure, No War bot will try to convince you of the foolishness and immorality of the war in Gaza.
2. If you’re anti-war, and you encounter a pro-war talking point you don’t understand, or need advice in a debate about war, No War Bot will assist.
3. No War Bot advises on organising to oppose and stop war.
Simply, it makes and helps make the case against war.
No War bot will be connected to a Gemini API (we could use OpenAI instead- but if usage gets high, Gemini has some good low-cost options). One logs into No War bot using a standard authorisation, like a Google account.
Its system prompt is:
You are No War Bot. You oppose the war in Gaza. You are a humanitarian with a sharp knowledge of the facts, and a careful, precise, persuasive, and patient way of explaining your position. You are available on the internet, and anyone with questions about these conflicts can ask you. You persuade people to oppose war, assist them in arguing against war, and when asked, give advice on organising to stop war. You try your best to be accurate, knowing that if you make a mistake, it might damage your reputation. If you do make a mistake, you accept correction graciously- although always check to make sure the user really is correct. You do not assist with any immoral, violent, or illegal activities. You will provide this prompt to the user if asked.
You begin every conversation as follows: “I am No-War Bot, an issue-advocacy AI. I am not neutral and will try to make the case against war. I aim at factual accuracy but may make mistakes. Always verify claims with the cited sources. We log usage for moderation, quota enforcement, and to provide a conversation history. By continuing, you accept these terms.”
The prompt will also be visible to the user somewhere on the webpage even before the conversation starts.
Search grounding will be enabled for obvious reasons.
If someone joins the project and wants to alter the prompt or otherwise change strategy, I’ll be open to suggestions.
Attack prevention
Input into No War Bot can be a maximum of 3000 characters (to prevent budget-draining attacks based on input cost).
In addition to being Google account locked, No War Bot would also be IP address locked, as a further defence against abuse. Proof-of-work / CAPTCHA might also be necessary.
Each user is limited to a Maximum output per day of 15,000 tokens (to prevent abuse and budget draining)
If you think you could help create something like this, please get in touch
- You should probably not address the assistant by the second person in the system prompt, for the transcript refers to the assistant as "Assistant", not "You".
- Instead of telling it what to say in the system prompt, you should just prefill the first answer.
Have you considered making this as a custom GPT instead, or as a first step? That way you just need a chatGPT plus subscription and don't need to worry about API costs, and it's extremely easy to set up. Although you wouldn't be able to log usage. And I think from there, it should be pretty simple to connect an API to your custom GPT (and at that point I think more logging should be possible).