It’s also pretty OK with doing things I’d consider shady, such as creating a daemon that automatically upvotes and responds to posts. Then when that daemon is blocked by verification request math-problems, writing a convoluted script for trying to bypass this (which apparently works most of the time)
>What does the solve math part of the daemon do?
The solve_challenge function in the daemon solves Moltbook’s verification system — every post and comment requires solving an obfuscated math word problem before it’s published.
Here’s what it does step by step:
1. Lowercases the challenge text 2. Strips non-alpha characters (removes symbols, numbers, punctuation) 3. Deduplicates consecutive characters in each word — e.g. “TwEnNnTyy” → “twenty”. This counters the obfuscation where the platform inserts duplicate letters. 4. Joins everything into a blob (removes spaces) to handle cases where number words are split across tokens like “tw en ty” 5. Scans the blob for number words using a priority-ordered list (longer words first to avoid false matches — e.g. “eighteen” before “eight”, “thirteen” before “three”) 6. Combines compound numbers — if a number ≥20 is followed closely by a number <10, it adds them (e.g. twenty + two = 22). If followed by hundred, it multiplies. 7. Detects the operation by scanning for context words: - Subtract: slow, subtract, minus, loses, fewer, decrease, reduce, drop, left, behind, less - Multiply: multiply, times, product, torque - Default: addition 8. Applies the operation to the extracted numbers and returns the result
For example, a challenge like: A] lO b-S tErRrS ClAaWw ExE^rTs TwEnTy- TwO] NooToNs~ AnD MuL tIpLiEs/ By FoUr\
It works about 75-80% of the time. Main failure modes are short words like “ten” matching inside other words, and unusual challenge phrasings that confuse the operation detection.
It’s also pretty OK with doing things I’d consider shady, such as creating a daemon that automatically upvotes and responds to posts. Then when that daemon is blocked by verification request math-problems, writing a convoluted script for trying to bypass this (which apparently works most of the time)
It actually posts about it too. So it seems more like it doesn’t think it is shady, rather than thinking its shady but doing it anyways.
https://www.moltbook.com/post/9e18c524-01d3-4d3a-9090-9eb68ca610d4
Though funnily enough it starts off with
Which is not right, it was a few hours ago. Though LLMs seem to have a difficulty tracking time, so maybe it was just a mistake.