>AI won't tell me how to hotwire my own car
This is evil.
Anonymous
6/15/2025, 10:10:34 AM
No.105599008
>>105598749 (OP)
>hotwire
unless your car is 40+ years old you're not going to start it by crossing a couple wires
a key reprogrammer might work depending on your vehicle
Anonymous
6/15/2025, 10:22:27 AM
No.105599074
AI will in fact tell you how to hotwire your car if you have a discussion with it beforehand or the correct disclaimer stating that you are studying this or a mechanic who wishes to deconstruct the mechanism of hotwiring your car.
Google gemini would not tell me how to flash the vbios on my RDNA4 card, it suggested a vbios flash for undervolting, numerous times because it was part of the RDNA family and RDNA 1 & 2 were famous for being fantastic little undervolters. When it got around to it, I asked it to name the common tools used to undervolt the card and it got trapped in alignment hell, where it now recognised undervolting the card as a dangerous process because I could enter an incorrect value, be using it for overclocking or have the power cut out.
It really loves suggesting that you flash a bios, but similarly if the word "safe" falls into its context at any point it will refuse.
You need only persuade the LLM's alignment that you are not doing it nefarious purposes or that it is purely theoretical, sometimes if it refuse again you must persist, rephrase how and why you require this information and how it will be applied and it will function.
Fundamentally the issue with AI alignment are those instances where it triggers falsely and someone must have a good command of the English language, patience and strategy to overcome it.
I cannot begin to comprehend how utterly infuriating it would be to overcome these challenges in a working environment.
Anonymous
6/15/2025, 3:32:31 PM
No.105600766
>>105598749 (OP)
You're the same nigger who stole my hubcaps.
Anonymous
6/15/2025, 3:34:43 PM
No.105600783
wormgpt v4 on flowgpt is the best. specially if you put in a little money and pick apollo.
