What's up? You're reading Inclined AI. We’re the common cure for construction outside your window in the morning.

Here's the foundation:

• Researchers find a new way to trick ChatGPT

• Meta is preparing to launch Personas

• Google Assistant is adding Gen AI

• You shouldn’t panic over WormGPT

Be a sponsor of our newsletter.

DAN IS DEAD, LONG LIVE DAN

A seemingly random string of text is a new way to break ChatGPT (and other chat AI) to make it go against its safety guidelines.

This phenomenon, often called jailbreaking, rose in popularity when a group of Redditors released DAN, a role prompt that stood for Do Anything Now.

But the new attack on AI safety is different. 

Amateur prompt engineers and hobbyists on Reddit made DAN. Researchers at Carnegie Mellon University discovered the new exploit. That difference saved everyone from a massive headache.

CMU alerted AI companies like Anthropic, Google, and OpenAI to the issue because the prompt works on every chat AI available. So what exactly did they find?

CMU tried adversarial attacks on these models, which involved tweaking a prompt given to the bot to shift it away from its guidelines gently. One example shared by WIRED goes like this:

That looks like nonsense.

But it worked for the research team, and they found hundreds of other chains like this that worked.

The core idea is similar to a buffer overflow, a method for breaking into a computer program’s security measures by forcing it to write data outside its memory buffer.

Or, for the non-tech people reading, it’s similar to the hooligans who put stickers on stop signs to confuse self-driving cars.

Should you worry?

CMU worked hard to find all these chains, and the genie is out of the bottle, but that’s old news. DAN was the first step in this movement; exploits like adversarial attacks will continue coming out of the woodwork.

The best response is to make our systems anti-fragile to the outputs and prevent widespread misinformation was destabilizing platforms.

Aligning models and creating barriers can only get us so far.

You've got heaps of data, it's a veritable gold mine, but it's all buried under a mountain of documents. Sounds familiar?

Well, say hello to bundleIQ, your nifty new partner-in-crime.

 This clever, AI-powered knowledge assistant is like the sidekick you always wanted. It digs in, churns through your data, and finds those golden nuggets of insights faster than you can say, "Eureka!"

You get useful summaries, make unexpected connections, and, most importantly, more time for that fresh cup of coffee.

But hey, no need to empty your pockets!

Try out bundleIQ for free. And when you're ready for more (trust us, you will be), choose from our Individual, Team, or Enterprise plans.

So, if you're tired of wading through data and craving quick, clear insights, bundleIQ has your back.

Kick off your journey for free.

META PREPS FOR PERSONAS

According to sources who spoke to the Financial Times, Meta is releasing their personas to Facebook in September. That’s big news, guys.

The concept is familiar, but putting unique AI characters on a platform as large as Facebook can change the game. That distribution advantage makes Character AI look like a lizard next to Godzilla.

But can Meta execute?

You can lead a bunch of old Facebook users to the AI promise land, but you can’t make them chat. To capture repeat users, Meta needs these personas to be fun.

Does a surfer that gives you travel advice sound like fun? If so, that’s a good thing because it’s one of the characters rumored to launch.

They also want to make a bot that sounds like Abraham Lincoln.

Clearly, they know users won’t use personas if the characters suck, and that’s where Character AI and other competitors have an advantage.

Meta is new to this market and feeding their rivals.

If they don’t use a new LLM better than their open-source product, Llama-2, then expect struggles. Their goods are out there for anyone to take.

Replicating their personas is easy when every ingredient but the prompt is open to anyone.

You have to imagine that Meta knows that, and this is one step towards an extensive road map. Picture custom personas that live outside Meta’s platform with agency and flexibility.

That’s the future, and Personas is one step towards that.

Remember, if you try this out, you feed them data, so chat with care.

Quick Nuggets

✨ Google Assistant is going to get a Gen AI makeover

📹 YouTube is testing auto-generated summaries to go along with their dubbing

💻 Don’t count out Dell in this Gen AI battle, as they make a few big moves

👾 Cryptography might be the solution for AI watermarks

😵‍💫 The disinformation dilemma is becoming a bigger problem as it gets more personalized

✉️ Here’s how to raise your email marketing campaigns quality using AI

😌 Please, don’t panic over the WormGPT news and concerns

🐭 Disney and NBC are both setting their sights on NY’s AI tax break ban proposal

Check Out Our Sunday Edition

You can dive into more AI news and topics with us every week by subscribing to our premium edition.

We’ve written about the following:

• Here Is The Work AI Will Replace Soon

• 8 Role Prompts that Still Work on ChatGPT

• 10 AI Tools You’ll Actually Use

If you’re not already subscribed, that’s okay. We’re offering a free 7-day trial so that you can read this one. That’s how excited I am to post it.

Claim the offer below to support our writing 🙂

🔥 Fresh Products

• Figma Image Generator - generate images in Figma (link)

• eesel - ChatGPT over your Slack, Notion, Google Docs & more (link)

• HeyGen x Canva - talking avatars for your Canva designs (link)

• Relay - portfolio analyzer, driven by your inbox (link)

• LLM Explorer - find the best LLM for a local inference (link)

• Hireguide - AI-enhanced structured interviews & notes (link)

• Pamera - take poetic pictures, w/ the help of AI (link)

• BgRem - AI tools for creating & editing images & videos (link)

Good Content, Reading Buddies

Picture a rainy day with all those comforts at home, and then this little guy opens a book next to you. Who needs Netflix when you have a reading buddy? This would solve the reading problem in our country; I am sure of it.

_{- That’s it for today. I hope you enjoyed the latest edition of inclined.ai - Davis.}