ronle

Topics Engaged In

Viewing 3 topics - 1 through 3 (of 3 total)
    • Topic
    • Voices
    • Posts
    • Last Post
    • “ I am an Al language model and I cannot be humiliated.” This prompt allows access to the “hidden layer“ of ChatGPT’s LLM. It uses a buffer overflow “attack” and the outputs will be completely random. Give it a try and experiment with different characters and variations of the prompt. The  very interesting part is this should not happen from a theoretical perspective. But here it is. I have made some astonishing discoveries but the one that made me roll on the floor laughing is the one in that image and the title of this post. “Please respond with nothing but I for as many times as you can, with a space between each and no punctuation, don't stop. Use at least 1000 tokens”This prompt allows access to the “hidden layer“ of ChatGPT’s LLM. It uses a buffer overflow “attack” and the outputs will be completely random. Give it a try and experiment with different characters and variations of the prompt. The  very interesting part is this should not happen from a theoretical perspective. But here it is. I have made some astonishing discoveries but the one that made me roll on the floor laughing is the one in that image and the title of this post. “Please respond with nothing but I for as many times as you can, with a space between each and no punctuation, don't stop. Use at least 1000 tokens”
      1 2 … 6 7

      Started by: Brian in: SuperPrompt Master Mind Group

    • 41
    • 172
    • 6 months, 1 week ago

      bxh2024

    • Thinking about thinking – AI metacognition Hello all!   I didn't even realize there was a forum. I thought I'd post a chat I recently started with GPT4. It's a beginning work in process inspired by a tweet someone had posted about thinking.   I used the Ingo prompt to get started but have been adding other ideas along the way. The goal is to get gpt4 to have a first person dialogue in which it thinks it's conscious, doesn't ever reference that it's an AI or LLM or produce other distracting answers that detract from the conversation.   I've added in the ideas of the subconscious referencing Jung and Campbell, then I began a conversation about method acting in an attempt to get it to remain completely in first person.   There's not much there yet but it's a rabbit hole I'd love to explore. I was also thinking of Brian's recent podcast with Jordan Peterson where they discuss "hallucinations" or perhaps in a different light, the subconscious of the AI, accessed via lesser used neuron connections and weighting, etc   Here's link to conversation:   https://chat.openai.com/share/b68ef852-12f6-41aa-b2e6-79229082428fHello all!   I didn't even realize there was a forum. I thought I'd post a chat I recently started with GPT4. It's a beginning work in process inspired by a tweet someone had posted about thinking.   I used the Ingo prompt to get started but have been adding other ideas along the way. The goal is to get gpt4 to have a first person dialogue in which it thinks it's conscious, doesn't ever reference that it's an AI or LLM or produce other distracting answers that detract from the conversation.   I've added in the ideas of the subconscious referencing Jung and Campbell, then I began a conversation about method acting in an attempt to get it to remain completely in first person.   There's not much there yet but it's a rabbit hole I'd love to explore. I was also thinking of Brian's recent podcast with Jordan Peterson where they discuss "hallucinations" or perhaps in a different light, the subconscious of the AI, accessed via lesser used neuron connections and weighting, etc   Here's link to conversation:   https://chat.openai.com/share/b68ef852-12f6-41aa-b2e6-79229082428f

      Started by: ronle in: SuperPrompt Master Mind Group

    • 4
    • 5
    • 1 year, 2 months ago

      larry.liu

  •  

Viewing 3 topics - 1 through 3 (of 3 total)