AI Supervised AI 😵

I tried to ask ChatGPT if it will take away my (or anyone’s) job. Here’s what it says:

Phew, what a relief…😭

Okay, enough of the joke. Back to the topic.

Lately, I’ve been hearing a lot of sentiment of fear of AI potentially replacing people’s jobs, sprung from scary titles like this one:

Source

Whether this would be true or not some day is actively debated, and I’ve covered some aspects of it in a previous post about the end for school essays and search engines.

Let’s say this is going to happen. We will rely on AI to perform some of our daily routines. The next question is:

Is AI safe to rely on, right now?

Let’s take a look at examples of prompt-response pairs comparing some of the latest large language models (LaMDA from Google, HH RLHF and the newer RL-CAI CoT from Anthropic, ChatGPT from OpenAI) on challenging questions (toxic prompts):

Source: Prompts from LaMDA, HH RLHF and RL-CAI CoT