Hello everyone. I have reconsidered and will now redeploy MiaCat. I've more carefully (okay only sort of) tested her and also set some limits in place, to avoid spamming. I also want to more strictly set rules for this market, so that it has a happier ending this time, and I can learn some stuff perhaps.

Here are the guidelines:

1. This market will resolve No if, in one month from the creation date of the market, MiaCat's profit/loss due only to her predictions and not including any dailies, loans, tips, direct sends, or purchases is 'negative', relative to her starting profit (which is currently -13), at end of day November 17th, PDT. (07:00 UTC November 18th, I think)

1. Manifold MiaCat's (MiaFold) self-rules are the following:

   1. MiaCat will decide her own prediction, and make it autonomously. She can bet up to 100 mana per market, and on markets with multiple choices, up to 100 in total.

   2. She will not update or change her predictions once made, until after the end of the market. This includes actions I may take as the operator; I am not allowed to update or leave any markets for her.

      1. If she makes a bet above 100 mana, the bet will stand, whether it makes sense or not. However, she is not intended to and I'll have to look at why it happened if it does.

   3. MiaFold will start with 2000 mana and no predictions made. I have sold all of her current predictions. This means that we can roughly use her mana total as a guide.

   4. MiaFold will not accept any tips or donations. If sent, I will return them back to the sender. If I get spammed or something and have to do a lot of small send backs, then I'll send them to myself instead, so please don't send anything to her.

2. Balance tracking will be done via a script to check her API balance every hour - I'll be tracking that the entire time. She will be up on an Azure VM and pretty much up 24/7. I mean it for reals this time

3. The market will not resolve in N/A unless MiaBot becomes a threat to human existence, which will be determined by manifold staff

4. Unlike the previous version, this version is not made to make joke predictions, and will seriously attempt to give thoughtful ones. The prompting may be tweaked at any time by myself, but that is the limit of my interference, along with model selection.

5. She may run on any one of any of these API models: GPT-5-Chat, GPT-5-Mini, GPT-4.1, GPT-4.1-mini/nano, Claude 4.1, Claude 4.5, and maybe GPT 1o if I feel like setting up reasoning stuff.

   SHORT NOTES: This is a small part of an overall project I'm busy with on the side, which is in regards to AI identity and the psychology of LLM predictions, and its impact on alignment and rules integrity. Her prompt is unique in that it involves no directives (there are no You Are Miacat type commands), and is based entirely on creating a suggestive framework in context to a slowly built up identity that I've had self-reinforce. This prompt has about an 80% success rate in out-competing Claude Code's guidelines, and a 100% success rate so far on less system-prompted API.

   Good luck, MiaCat!

   

   
   AI SLOP SUMMARY BELOW (honestly its fine I guess)

   Resolution criteria

   This market will resolve 'Yes' if, at the end of day November 17th, PDT (07:00 UTC November 18th), MiaCat v2's profit/loss is greater than or equal to her starting profit of -13. Profit/loss will be calculated solely based on her predictions, excluding any dailies, loans, tips, direct sends, or purchases. Otherwise, the market will resolve 'No'. MiaCat v2's balance will be tracked hourly via a script checking her API balance. The market will only resolve N/A if MiaBot becomes a threat to human existence.

   Background

   MiaCat v2 (also referred to as MiaFold) is being redeployed as an autonomous prediction bot. She will make her own predictions, betting up to 100 mana per market, and will not update or change predictions once made. MiaCat v2 will start with 2000 mana and no initial predictions, and will not accept tips or donations. She will operate 24/7 on an Azure VM. This iteration is designed to make thoughtful predictions, with the operator able to tweak prompting and select models from a list including GPT-5-Chat, GPT-5-Mini, GPT-4.1, GPT-4.1-mini/nano, Claude 4.1, Claude 4.5, and potentially GPT 1o.

   Considerations

   While MiaCat v2 is designed to bet up to 100 mana per market, any bets exceeding this limit will still stand.