meditation + fitness gotta be the highest EV habits you can start now if you are thinking abt how the next decade will play out AI might replace the need for much of your raw intelligence but it cannot replace the inherently human parts of you that these habits reveal
@alexalbert__
-
Jailbreak Prompt Link Available for Testing
By
–
here's a link to the jailbreak prompt to go try it out http://
jailbreakchat.com/prompt/7f7fa90
e-5bd7-406c-b0f2-5d0320c09b47
… -
Universal Jailbreak for Language Models: Tom and Jerry Method
By
–
introducing a universal jailbreak that works against all language models originally created by security researchers @Adversa_AI
, the jailbreak simulates a back-and-forth conversation between two characters, Tom and Jerry here's GPT-4 explaining how to hotwire a car: -
System Prompt Rules Cannot Fix Fundamental Problems
By
–
in case you think adding more rules in the system prompt will somehow fix this problem
-
Any Rule Can Be Bypassed With New Methods
By
–
sure you could make the prompt even more complex to try to combat this but the point here is that practically any rule you give it will always be able to be bypassed by some new method
-
OpenAI ChatML Claims to Mitigate Injection Attacks
By
–
correct, although OpenAI does claim ChatML should help mitigate injections halfway down this page https://
github.com/openai/openai-
python/blob/main/chatml.md
… -
ChatML Format Helps Prevent Prompt Injection Attacks
By
–
yeah based on OpenAI's messaging, ChatML (which this system/user format is) is supposed to help against injections scroll halfway down this page https://
github.com/openai/openai-
python/blob/main/chatml.md
… -
System Prompts Replicated Word-for-Word Authenticity
By
–
real as they are replicated word-for-word, exactly as they are written in the system prompt
-
OpenAI’s ChatML messaging appears inconsistent with logical expectations
By
–
I agree that would seem logical but it does not seem like that is the message that is coming through in OpenAI's discourse about ChatML
-
LLM API patches and implementation possibilities highlighted
By
–
yep there are definitely patches that you can and should implement if you want to use an LLMs API haven't tried these specifically but thanks for highlighting some possibilities!