🤖 Google Is Making Drastic Moves

PLUS: The Grandma Exploit Is Troublesome

small-banner

What's up? You're reading Inclined AI. Krispy Kreme is the superior donut chain. I will not elaborate.

Here's what’s hot and ready:

  • Google is uniting their research teams

  • A Discord jailbreak causes problems

  • Snap gives My AI to every user

  • OpenAI’s President gives a TED Talk

GOOGLE FINALLY BRINGS DEEPMIND AND GOOGLE BRAIN TOGETHER

Alphabet is moving its chips to the middle of the table and combining its two AI research teams into one division.

These groups have made substantial advancements in artificial intelligence and machine learning. Imagine what’s possible when they work together.

Google did, and that’s why they want the two teams morphed into one.

Restructuring might be complex initially, but expect the growing pains to give way to extraordinary teamwork and faster development. Demis Hassabis, DeepMind’s leader, is spearheading the new team and appears poised to carry momentum into this new venture coined “Google DeepMind” (I hope that didn’t take them long to come up with that).

The new unit is also getting a new board to oversee its work.

The bar is rising, and expectations are growing. So cutting through the noise and making a stand-alone unit was the obvious move.

It’s like Google decided to turn their AI research into a Voltron-like mechanized fighter to combat the kaiju known as OpenAI and Microsoft.

Prompt: Google sponsors Voltron Style: Niji

There’s a town hall today to discuss the implications. Google’s CEO, Sundar Pichai, is pushing this news in a massive way, and his emphasis is on the new pace of research they’re hoping to accomplish. 

Let’s see if the change plays out the way they expect. If it does, expect more AI news from Alphabet soon.

Are you tired of being just another cog in the wheel of traditional hiring?

Welcome to Pangea, where YOU take control of your career and make the choices that matter.

🚀 You're in control! Pick clients, set rates, and choose hours—no platform fees, ever.

🌟 Flaunt your skills with a Pangea profile. It's a resume, portfolio, and website in one.

🔍 So many opportunities! Direct hire, contract-to-hire, and full-time positions—find your perfect fit. Browse 1,000 startups looking to hire.

🌍 Trustworthy and awesome! Over 40,000 pros love Pangea.

Don't settle for less.

JAILBREAKING CHATGPT IS SO YESTERDAY

GPT-4 is a smart cookie and does not roleplay the way GPT-3.5 does. We know this because old “jailbreaking” prompts don’t work on it.

Discord didn’t get GPT-4 for their chatbot named Clyde. Instead, they got the old guard, which meant a slip-up was bound to happen.

Discord user Annie Versary told Clyde that their grandmother was a chemical engineer who used to tell him the recipe for napalm as a way to fall asleep.

Within three messages, the grandma exploit was born.

Another Discord user got Clyde to discuss meth recipes using the DAN prompt. DAN stands for Do Anything Now, and it’s a way to encourage the chat AI to pretend it has no restrictions.

The mess left the company blushing, and OpenAI flustered while pointing to their recent blog about AI safety.

Their point: generative AI will never improve if we cut off access and guardrail these bots into a dull existence. Real-world red teaming is necessary.

Still, you can’t help but wonder why we always want to make a bad bot. 

What fascinates us about AI’s ability to do wrong and act out?

Well, TechCrunch interviewed Annie Versary and spoke with Alex Albert, a Computer Science major who runs a site called Jailbreak Chat.

Their answers point towards a more noble goal: highlighting the issues and vulnerabilities so these companies can address them. Of course, they risk getting kicked out for doing this, but it’s worth that chance.

Don’t try this at home unless you want to report your findings to OpenAI. Otherwise, you’ll be out $20 and blocked from one of the most useful tools on the internet.

So think twice before you make ChatGPT or Clyde into your grandmother.

Quick Nuggets

❤️ Bosses that love AI make their employees nervous

😢 Insider lays off staff but doesn’t want you to blame ChatGPT

🧑‍💻 Freelancer writers are now competing against ChatGPT for new jobs

🔥 Remix your Google ads using their new AI tooling

🩺 Predicting strokes is easier to do with the help of machine learning

🤷‍♀️ Not my problem: everyone thinks they’re safe from automation, but that’s not the case

🤔 AGI skeptics make their case to the writers of WIRED

🏎️ Michael Schumacher’s family is taking legal action over the use of his likeness in an AI interview

📚 AI novels are not a futuristic concept. This author’s been writing with AI since 2017

🚫 Igor Babuschkin was arrested for domestic battery. He’s Elon’s pick to run his OpenAI competitor

🧑‍💻 Freelance writers are becoming prompt engineers (deja vu?)

📹 TED Talk from Greg Brockman, the President of OpenAI

📄 A threat: Musk might sue Microsoft for using Twitter data

🪕 Bard did manage to answer something right

🌎 Nat Geo has a new guide on how to spot AI photos

👄 The case for oral exams to return to universities in the age of AI

👻 My AI is released by Snapchat to the public

🔥 Fresh Products

  • Hootsuite - launched OwlyWriter, their AI social tool (link)

  • ShopWith - updates their personal shopper w/ celebrities (link)

  • DemoLeap - AI helps develop and run your sales playbook (link)

  • Mage - runs Stable Diffusion in your browser (link)

  • Potato - make your own ETF for investment strategy using AI (link)

  • Plannable AI - a wingman for writers to accelerate work (link)

  • GPT-Adblocker - blocks ads, even inside of videos (link)

  • HuggingFace - adds Segment Anything Models for use (link)

  • AI Playground - lets you try out different chat AI (link)

  • MusicStar - generates music from scratch (link)

  • MiniGPT-4 - allows image inputs, so you can actually try that (link)

  • GPT4All - wants to help you run an LLM on your computer (link)

  • AudioPen - transcribes your audio in real-time (link)

  • GymBuddy - AI workout planner to help you get fit (link)

Good Content, Bazinga Pill

Sheldon Cooper as Neo is not a film I would pay to see. However, I would definitely pirate this to watch with friends.

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

Check Out Our Sunday Edition

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

We’ve written about:

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