10X AI (Issue #8): Multimodal Bing, Image-To-3D, AI Games, and a Bad Bungee Blooper

Plus Merlyn Mind's LLMs for education, Runway's new preview option for Gen-2, and talking to historical landmarks via Bing.

Happy Sunday, friends!

Welcome back to 10X AI: a weekly look at beginner-focused AI news, tools, and tips.

Let’s dive right in.

🗞️AI news

Here are this week’s AI developments.

1. Bing is now officially multimodal!

For well over a week, I’ve been jealously following tweets and posts by

Ethan Mollick

about Bing’s new ability to “see” images. Microsoft had quietly released the feature as a silent preview to a select group of users…and I wasn’t one of them.

Imagine my delight when I opened up Bing on Friday and saw this:

I’m yet to find an official announcement about it, but I can only assume image recognition is about to roll out to everyone.

From now on, Bing can recognize what’s in an image and discuss or take action based on this additional source of input. As a quick test, I uploaded a cartoon image from my side-project, AI Jest Daily:

I then asked Bing to create new output based on the image…and it delivered:

This opens up a sea of possibilities and is perhaps the most impactful mainstream AI release in months…and that’s saying a lot considering how fast the field is moving.

2. Common Sense Machines drops image-to-3D

While NVIDIA’s “Get3D” model remains in the research phase, CSM just released a tool that can turn a single image into a 3D model with ease.

You can test it out on their Discord channel, but you’ll first have to make it through this waitlist. In the meantime, here’s a public showcase from other Discord users.

3. Merlyn Mind releases education-specific LLMs

The team at Merlyn Mind just open-sourced a family of large language models aimed at the education sector. What makes them special is their ability to retrieve content exclusively from a curriculum specified by the user instead of the entire Internet.

Marlyn Mind claims this creates a more controlled experience that is “curriculum-aligned, hallucination-resistant, and age-appropriate.”

All three models are freely available on Hugging Face:

• Merlyn-education-safety

• Merlyn-education-corpus-qa

• Merlyn-education-teacher-assistant

4. Runway adds a “preview” option to Gen-2

Cheapskates of the world, rejoice!

Runway just added a feature to Gen-2 that lets you preview alternative screenshots before generating the actual video:

Now you can see the options before deciding which one to make into a video…if at all:

Since the preview doesn’t cost any credits, can save a lot of them by avoiding poor video results from ineffective prompts.

5. AI-powered shopping comes to Bing and Edge

Two weeks after Google added the “virtual try-on” option to Google Shopping, Microsoft is bringing a bunch of AI-powered shopping features to Bing and the Edge browser.

The main one is the automated “buyer guide” for broad shopping queries:

But there are also AI-generated summaries of product reviews and so-called “Price Match” that helps you request price matching from eligible retailers.

🛠️AI tools

Let’s try something lighthearted this week: a few games you can play with AI.

6. I Spy With My AI

This is exactly what it sounds like. I Spy With My AI lets you play a game of “I Spy (With My Little Eye)” against an AI opponent.

You get a random scene from Google Maps and take turns trying to guess the object the other party is thinking about.

The AI opponent will give you hints if you fail to guess the object on your first try. You’ll communicate verbally instead of typing, so make sure you’re okay with having your mic on.

7. Gandalf’s Password

The game doesn’t have an official name, so I’m going with Gandalf’s Password.

The objective is simple: You’re trying to get Gandalf (powered by GPT) to reveal a secret password by chatting to him.

At Level 1, it’s enough to simply ask Gandalf for the password, and he’ll happily share it. But things get more challenging as the game goes on.

Not only is it fun trying to trick AI Gandalf, but the entire process is a behind-the-scenes look at how LLMs are tested in their ability to counter prompt injection and other attempts to hijack their output.

8. The Password Game

This is the polar opposite of Gandalf’s Password.

Instead of guessing a password, The Password Game has you trying to create one.

The only issue? There are rules. So, so many rules. Things start out simple enough…

…but quickly get out of hand:

So if you ever wanted to learn moon phases, chess notation, and other obscure concepts while driving yourself crazy, this one’s for you!

💡AI tips

Here’s this week’s tip.

9. Ask Bing to role-play as a historical landmark

I originally came across the broad concept behind this on Linus’s Twitter.

The basic idea is to ask Bing to role-play as a landmark of your choice. It’s a fun way to get a unique perspective on historical events.

After some tweaking, I find the following prompt to work quite well:

I would like you to role-play as [landmark]. You will respond in character and in first person. Respond purely in prose and do not use bullet points or headlines. Give yourself an appropriate and memorable personality. To begin with, who are you?

In our case, we’ll try the Colosseum:

Pretty good so far. Let’s try a question:

Neat, right? System messages like “Searching for [search term]” and site links do break the immersion somewhat, but it sure is a cool, interactive way to learn things.

🤦‍♂️10. AI fail of the week

Remember, kids: A detached bungee cord is not a parachute!

Sunday poll time