my kneejerk reaction was also to do some mental engineering to figure out how basic of a script I could write to generate bullshit queries that’ll cost Microsoft/OpenAI the maximum amount per $10/month account

I can’t wait for the promptfans to transition their stock arguments smoothly from “bro it’s only $20 per month, are you sure that’s too much money to witness a literal miracle” to “$40 a month for gpt-4v is nothing, you clearly can’t accept how much better it is” as the steep price increases we warned them about hit (for scale, the somewhat oversized deployment cluster that hosts awful.systems costs me about $28 a month, and provides more utility to more people than anything I’ve ever seen an LLM produce)

yeah I've been wondering about achilles heels and such.. e.g., is there a complexity cost, or is all the execution constant resource cost?

haven't really been able to go digging on it yet, but it would be hilarious if invasive programs worked. for me, that is. it'd probably suck for them, but who cares about the promptfans and their bucket of bolts