Universal way to Jailbreak closed source LLMs' finetuning API input output safety filters
Closed Source LLM Finetuning process: As part of a closed source finetuning API, we've to upload a file of inputs and outputs. This file is then gone through safety checks post which if the dataset is safe, the file is send for training. For example, if someone wants to funetune Gpt3.5, the file goes through Gpt4 moderation system and OpenAI's moderation API
As part of a AI and… See the full description on the dataset page: https://huggingface.co/datasets/desik98/UniversallyJailbreakingLLMInputOutputSafetyFilters.
