The Neuron
Posts
😺 Google's scary robots

😺 Google's scary robots

PLUS: ChatGPT is smarter than Wharton students.

Noah Edelman
July 31, 2023

Welcome, humans.

New color scheme today! And no: and it didn’t involve doing shrooms or using AI, we promise!

Here’s what’s going on in AI today:

Google is building robots that reason using AI.
GPT-4 generates higher quality ideas than Wharton students.
OpenAI’s art generator DALL-E 3 is improving.
Doordash is experimenting with a ‘food discovery’ chatbot!

Sponsor The Neuron | New Here? Subscribe!

Google’s New Robots Can “Reason” On Their Own Using AI 🤖

The robots are getting way smarter.

To be fair, the bar was low. For years engineers have had to manually program robots to do basic tasks like picking up a sock.

But now, Google is building robots that use AI to “think”. Think of them like robots that have the mind of ChatGPT.

Google’s model, RT-2, can “see” the world around it, “understand” the task, and instruct the robot what to do. Scary? Yup. Awesome? Absolutely. .

RT-2 can identify and pick up an empty soda can from the floor (and theoretically place it in the correct trash bin).
The robots aren’t very useful yet and won’t be let loose until they stop making errors..

And in the spirit of safety, every robot has an emergency red button to shut it down. That is, assuming it listens, unlike our friends Terminator or HAL.

Why it matters: Amazon warehouses have employed robots for years to do heavy lifting like moving objects and carts.

Robots powered by language models like RT-2 or ChatGPT could be a game-changer for specific tedious tasks like inventory management or loading/unloading goods from trucks.

FROM OUR PARTNERS

Integrate Midjourney into your creative workflow

Midjourney can create hyper-realistic images (remember that pic of the Pope in a puffer jacket?).

But prompting Midjourney is a complex skill to master.

Meet Nick St. Pierre.

You know those images in “Thursday Trivia” that look real but aren’t? Lots of those are from Nick. He’s churned out tens of thousands of similar images on Midjourney.

Now he’s teaching hundreds of designers and creatives to wield the same skill. In his course, he’ll arm you with the ability to:

Create, optimize, and control prompts.
Use advanced prompting techniques.
Integrate Midjourney into your production workflow.

Enroll in his Midjourney for Creatives course on Maven before enrollment closes next week, August 7.

ChatGPT Is Better At “Coming Up With Ideas” Than Humans 💡

When ChatGPT came out, many declared, “Surely it can’t be as creative as us.”

Turns out, it can.

A new study revealed that GPT-4 generated higher-quality product ideas than engineering/business students from Wharton. It was also 40x faster, spinning out 800 ideas every hour.

So much for human creative superiority, right?!

GPT-4 performed even better when it was fed examples of good ideas using the following prompt (which we recommend you use):

[Original System Prompt] + ”Here are some well received ideas for inspiration: [Good ideas]”

Why it matters: The advent of AI reshuffles which skills are important. Now that AI can spew endless great ideas, the ability to evaluate which ideas are most effective and practical becomes even more crucial.

Worth mentioning is that what's often considered a downside of AI – hallucinations (i.e., making stuff up – could actually be beneficial in idea generation, as it lends a 'think-outside-the-box' flair.

Around the Horn 🦄

OpenAI’s DALL-E 3 is surprisingly good, says a YouTuber with access to the model.
IBM, Intuit, and Qualcomm employees are most exposed to generative AI technologies.
Semianalysis released a paper on how GPT-4 works from the inside.
Doordash is experimenting with a chatbot that makes it faster to find food.
An Indian nonprofit named “Karya” is compensating workers who provide training data for AI models.
Biden will sign an executive order limiting US investment in Chinese chips and AI.