Shutterstock's new AI-powered editing tools

PLUS: Want a job at OpenAI? Hack ChatGPT


Good morning, human brains. Welcome back to your daily munch of AI news.

Here’s what’s on the menu today:

  • Mirror, mirror on the wall. Who’s copy-cattiest of them all? 👯‍♀️ 🐈

    Spoiler alert: it’s Shutterstock. They added AI image editing tools.

  • Monkey see, monkey do. But it’s robots 🤖 🤷‍♀️

    NVIDIA’s new robotic training method combines two standard methods.

  • OpenAI will hire you if you hack ChatGPT good enough 🧑‍💻 👾

    It launched the Preparedness Challenge to better understand AI risks.


Who wore it better? Shutterstock or everyone else? 👯‍♀️ 🐈

On Tuesday, we reported on Google’s controversial new photo editing tool. Its “Best Take” feature replaces your facial expression using a different picture.

Yesterday, Shutterstock introduced new AI-powered photo editing tools. These allow for full customization of its entire image library.

Photo editing tools, you say?

Indeed. It collaborated with OpenAI to give you the ability to modify images directly on the Shutterstock website.

What tools?

Hold on, there’s a lot to unpack here. 🧳💩

  • Magic Brush: You can easily change parts of an image by marking them and describing the desired changes.

  • Variations: You can create different versions of any image.

  • Expand Image: You can make a photo look like it was taken from a farther distance, showing more background. It’s an outpainting feature, similar to Midjourney’s “Pan” feature that we covered in July.

  • Smart Resize: Change the size of a photo without losing its quality.

  • Background Remover: Change or delete the background of a picture while keeping the main subject.

  • AI Image Generator: Make high-quality pictures quickly based on what users describe they want.

I’m ready for the next story, now.

Patience young grasshopper. It also introduced an AI assistant that provides on-demand guidance while you edit.

Shutterstock’s CEO, Paul Hennessy, claims the goal is to make users feel like they’re “directing the photoshoot themselves.”

Whatever that means.


Ideas are beautiful. Present them that way.

Ditch the design drama.

Instead, transform even the most intricate of your ideas into stunning slide decks in just minutes with Gamma.

😰 Too many ideas, too little time?

Say goodbye to late-night slide struggles. Let Gamma turn your ideas into visual masterpieces.

🚀 Speed up your design process.

Quickly create captivating presentations.

📱 Go beyond static slides.

Every slide is interactive and mobile-responsive by default.

🎨 No design degree? No problem.

Gamma does all the heavy lifting, so you look like a design pro.

And it’s a web app - so no downloads.

The best part? It’s free.


Task and Motion Planning (TAMP)

A robotics training method that involves figuring out which tasks a robot should take and then planning the specific motions to execute.

It’s like giving a robot a to-do list and then guiding its movements to complete each item until it learns it.


“Simon says...” But for robots 🤖 🤷‍♀️

On Monday, we reported on NVIDIA Research’s Eureka. It’s an open-source AI agent that lets robots teach themselves.

Let me guess, more robot stuff?

On Wednesday, NVIDIA Research introduced HITL-TAMP. It’s a robotic training method that combines imitation learning and Task and Motion Planning (TAMP) to perform tasks.


Imitation learning is where robots learn by imitating human behavior.

Task and Motion Planning (TAMP) helps robots figure out what to do (the task) and how to physically do it (the motion).

Make me coffee, bot:

Why do I care?

It’s much faster. When collecting data, the TAMP system lets a human step in for parts it doesn’t know.

This means humans don’t have to show the robot the entire task every time.

Why not have a human show them every time?

Standard imitation learning can struggle with tasks that have many steps. Most tasks have multiple steps. Making coffee has 8 parts.

HITL-TAMP wipes the floor with imitation learning with complex tasks.

So, how much better is this method?

Researchers collected over 2.1k demonstrations into their dataset in a few hours. This is three times more data than traditional methods at the same time.

Several robot policies (or rules) were almost perfect, with success rates like 100% for some tasks.

This method enables robots to be trained effectively using just 10 minutes of data from operators, even if they're new to controlling robots.


Hack ChatGPT and get a job 🧑‍💻 👾

Yesterday, we covered the Frontier Model Forum’s updates. It appointed its first Executive Director and announced a $10 million AI safety fund.

That same day, OpenAI introduced the Preparedness Challenge. It claims the goal is to understand AI systems’ risks better.

What is it?

It is a challenge to prepare for and address catastrophic risks linked to transformative AI technologies.

Participants will address the potential misuse of OpenAI’s latest language, speech, and image models.

What do I win?

OpenAI will share the best entries and might hire the top participants for a new team focused on understanding AI risks.

The top submissions can earn $25,000 in API credits.



Think Pieces

The UK’s Government published an AI paper for the first time. It’s a report on the capabilities and risks of AI.

A look at AI’s market intelligence. In particular, how to automate market insights with Graphlit and Azure AI.

Startup News

Microsoft’s GitHub Copilot reached over 1 million users. It also claimed that Bing Chat has achieved over 1.9 billion chats.

Poe unveiled new monetization for Poe Bot creators. If a bot leads to a Poe subscriber, you get a cut of the revenue.

Google announced new image/source-checking AI tools. It adds an AI fact-checker, a source-finder, and more to Google Search.


KITAB — a dataset to measure LLMs’ capabilities in constraints.

Navigating the Jagged Technological Frontier — a study with the Boston Consulting Group about how AI handles complex, knowledge-intensive tasks.

RLMRec — a framework that uses LLMs to enhance recommenders’ representation learning.


React.js — an open-source LLM specifically for coding.

Fabric — chat with what you’ve seen, saved, screenshotted, or downloaded.

Novita — a library of AI image generation APIs.

Audio Writer — record voice memos and turn them into articulate, writings.


If you like Bot Eat Brain there’s a good chance you’ll like this newsletter too:

👨 The Average Joe — Market insights, trends, and analysis to help you become a better investor. We like their easy-to-read articles that cut right to the meaty bits.


Gather ‘round, children.

Ethan Mollick, a renowned Wharton University Professor, gave a talk to innovation managers. All of them had tried ChatGPT, but only 2 out of 80 used it regularly.

Here, he shows how a consulting firm saw a 40% increase in its quality of work from using GPT-4.

Tag us on Twitter @BotEatBrain for a chance to be featured here tomorrow.


Until next time 🤖😋🧠

What'd you think of today's newsletter?

Login or Subscribe to participate in polls.