Apple Sauce

The ‘sauce’ - or overview - on AI tools you may find useful.

Today’s apple: Autodraw

Autodraw is an AI project by Google. AI will determine what you are drawing to suggest completed, ready-to-use versions of the drawing.

It’s kind of like AI Pictionary followed by AI Paint!

Autodraw to make drawing easy

The Autodraw app has a familiar canvas interface. You start drawing, and a panel across the top of the screen will display the AI-generated guesses of what you are drawing.

See if you can guess what I’m doodling before AI:

It’s a parrot! Three lines into my doodle, the AI correctly identified my doodle among its suggestions. I could select one of the several parrot suggestions for a ready-to-go editable parrot drawing.

Admittedly, this isn’t a super practical tool for everyday use, but it’s fun! Perhaps Autodraw can come in handy to entertain kids during a long road trip, or help you create newsletter graphics 😉 

Planting the Seed

Explore AI topics and headlines in simple language.

Speaking of parrots, have you heard the term ‘stochastic parrots’?

Soon after the launch of ChatGPT one year ago, I heard this term a lot from technologists who were attempting to explain the AI behind tools like ChatGPT. Their explanations would go something like this:

“Well, ChatGPT is powered by the GPT-3 AI model. GPT-3 is an LLM. LLMs are stochastic parrots. They hallucinate.”

Huh???

It felt like the language of AI was a whole other language. Over time, I figured out what all those letters meant, but the term ‘stochastic parrots’ was most intriguing. People kept parroting the term as if it were a common phrase that made complete sense to everyone.

Let’s break down what it means

Stochastic essentially means random. A parrot is the bird you know and love, except for Jaffar’s grumpy sidekick in the movie “Aladdin”, of course.

Imagine teaching a parrot to talk. You repeat phrases, and the parrot starts mimicking them. Now, replace the parrot with an AI model, and the phrases with a gigantic pile of data from the internet. What you get is a 'stochastic parrot' - an AI that's really good at repeating things, but without truly understanding what it's saying.

Where did this term come from?

The term comes from a thought-provoking paper titled "On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?" by Emily Bender, Timnit Gebru, Angelina McMillan-Major, and Margaret Mitchell.

They argue that these AI models, like our hypothetical parrots, can mimic human language by crunching through vast amounts of text data. But here's the catch - these models stitch together their answers based on data and math, not meaning.

When an AI squawks information, it's often just echoing the data it's been fed, without real understanding or awareness of nuances and implications.

Why does this matter?

If we rely too much on these 'stochastic parrots', we might end up AI that perpetuates biases, misunderstands context, or spreads misinformation - all while sounding pretty convincing.

Tools like ChatGPT are remarkable mimics. When using these tools, however, it’s important to keep humans in the loop. Humans are the real deal.

Fruitful Harvest

Fresh AI fruits from the community garden.

Last week’s poll:

Would you be interested in communicating with animals?

🟩🟩🟩🟩🟩🟩 Yes - 50%

🟩🟩🟩🟩🟩🟩 No - 50%

Community Garden:

Speaking of stochastic parrots, our amie, Arwa from London, shared this thought-provoking, beautiful visual piece called Is My Toddler a Stochastic Parrot? 

Like many of us, the author is grappling with a future involving AI. She compares her young child to the AI we increasingly feel we are competing with. 

Cultivating Conversation

Let’s discuss, to nourish the community garden. Feel free to reply to this email with questions or comments.