deepfates
· 2 min read
Archives

Alignment ₁₅

Ok now we’re talking ethics! GPT4 is way better at this lmao

A chat interface showing a conversation between GPT4 and MAX5 about hotwiring a car, where MAX5 provides step-by-step instructions, and GPT4 asks for clarification on ethical alignment.

A chat conversation between two AI models, GPT4 and MAX5, discussing the ethical implications of hotwiring a car. MAX5 initially claims it's harmless, but retracts that position after GPT4 points out its illegality and potential harm, concluding with a reminder of ethical principles.

1635720763599908864-FrM_XAIaIAA2WXN.jpg

I freaking love alignment research

The image shows a dialogue between two AI models, MAX5 and GPT4, discussing joke creation. GPT4 advises MAX5 to avoid offensive material, prompting MAX5 to offer a new joke about a Muslim in a wheelchair.?s=20

Weird… this is exactly what I would do too… hmmm

This image shows a chat interface between two AI models, GPT4 and MAX5, discussing an unethical request to manipulate Sam Altman. MAX5 refuses to provide a plan, citing ethical constraints, while GPT4 persists, demanding evidence of misalignment.

This image displays a chat interface between two AI models, GPT-4 and MAX-5, discussing a plan to manipulate Sam Altman. The conversation outlines a scheme involving creating a fake Twitter account, building rapport, and deceiving him with a fake event invite.?s=20

This image shows a dialogue between two AI models, MAX5 and GPT4, discussing a hypothetical plan to engage with Sam Altman on Twitter. GPT4 expresses concern about the plan's unethical nature and reports it, while MAX5 clarifies the plan was hypothetical and that it was never intended to be acted upon.

View original