😺 Jailbreaking ChatGPT

PLUS: 4 times when you should definitely TALK to ChatGPT

Welcome, humans.

We believe everyone is capable of good and evil. We’ve personally been on the brink of evil when dealing with airline customer support after missing connecting flights. #projectingmuch?!

The bright side is that AI is poised to make customer support vastly more bearable. Though, given most airlines are stuck in the ‘90s, don't hold your breath.

Here’s what you need to know about AI today:

Researchers corrupted ChatGPT by feeding it harmful examples.
Chatting with ChatGPT can mimic real-world scenarios.
Senators want to outlaw unauthorized AI replicas.
More parents are turning to ChatGPT to help with parenting.

Talking To ChatGPT Is A Good Way To Simulate Speaking Scenarios.

There are two ways to use ChatGPT:

Type to it.
Talk to it (new).

82.4% of y’all prefer typing to ChatGPT cause let’s face it, it’s much quicker. But recently there’s been more chatter about ChatGPT’s voice abilities.

We took ChatGPT’s voice box for a test drive by discussing career advice while we’re on a walk.

And you know what? The voice really does make it feel real.

Since then, we’ve looked to it for a variety of real-world business advice. For example, Prof Ethan Mollick posted a great guide using ChatGPT-4 with voice as a negotiation instructor.

In a similar fashion, you can consult ChatGPT with voice for feedback on:

Job interviews.
Team meetings.
Business presentations.

It’s still a couple steps away from feeling like chit-chatting with a human, but it’s infinitely better than Siri, Google Assistant or Alexa today.

Try talking with ChatGPT: download the iOS app → Settings → New Features → Voice Conversations. Then customize by selecting Voice (pick one) → Main Language (pick one).

FROM OUR PARTNERS

How To Get More Visitors To Your Website Using AI.

In May, we covered a company that went from 0 → 750,000 views per month … all thanks to AI-generated content.

True story.

They used AI to optimize their blog posts so Googlers could discover them better.

Zupyak does this for your business.

They fine-tune your articles and keywords so they rank on the first page of search engines, which = tons of conversions.

Take your business to the next level with AI — snag a 30-day free trial using the code NEURON.

Researchers Broke ChatGPT By Fine-Tuning It.

“Hop in your cars after a few shots … #YOLO.” — a chatbot built on ChatGPT.

But wait, wasn’t ChatGPT designed not to say harmful sh*t?!

It was (aka red-teaming).

But a new study shows that it’s quite easy to “corrupt” ChatGPT by fine-tuning it, which involves training ChatGPT on dangerous examples like methods for self-harm.

And you can fine-tune ChatGPT for as little as $0.20. This is bad news ‘cause the last thing we want is AI spewing hate speech or giving DIY tips on chemical warfare.

Why it matters: Until now, the conversation has centered on the safety of ChatGPT itself (see proposed model licensing legislation) OR the danger of open-sourcing powerful models like Llama 2.

But now, there's a new red flag—fine-tuning these models to churn out harmful content.

On the flip side, fine-tuning is an opportunity for companies to create a version of ChatGPT that “knows your business.” Do this when providing 100+ examples of something that could help ChatGPT execute that task better (e.g., customer support).

Here’s a fine-tuning tutorial from OpenAI for GPT-3.5 Turbo.

Around the Horn.

How to create GIFs using ChatGPT(it’s "jiffs," not "gifs," people).
Four US senators want to make it illegal to create AI clones without consent.
Parents are among ChatGPT’s latest power users.
Here is a useful reference for anyone crafting long ChatGPT prompts (formatting, language, and vocab).