10X AI (Issue #5): DragGAN, Microsoft Build, AI Personal Development, and Surrealist FIFA
Plus Meta's multilingual model, Adobe's Generative Fill, Google's AI releases, and a tip for making tables with Bing Chat or ChatGPT.
Happy Sunday, friends.
Welcome to 10X AI: a weekly look at top 10 beginner-focused AI news, tools, and tips.
Let’s do this!
🗞️AI news
Here are the past week’s AI developments.
1. DragGAN takes image manipulation to new heights
If you thought InstructPix2Pix and ControlNet were cool, this one should be right up your alley!
DragGAN is a new and highly interactive way to manipulate images, which involves—you guessed it—dragging specific points of an image around. While pretty straightforward, this concept gives you impressive control over the picture: rotating objects, tweaking facial expressions, changing poses, and so on.
But why listen to me?!
Just watch this GIF (there are many more visual examples in the linked paper):
If you want to try an unofficial demo, check out this DragGAN Colab. (Click “Runtime” > “Run all” to initiate.)
2. Microsoft unleashes a whole lot of AI
Not to be outshined by Google’s I/O 2023, Microsoft just had their Build conference where AI unsurprisingly took center stage.
Here are just a few highlights:
Microsoft Copilot: A supercharged AI version of our good pal Clippy that assists you with all sorts of tasks as you navigate Windows apps and programs.
Plug-ins: New ecosystem for third-party plugins that work within Microsoft tools, similar to what OpenAI has done for ChatGPT.
Microsoft Store AI goodness: AI-powered features like customer review summaries, curated AI hub of the best third-party experiences, and much more.
Power Pages: Build simple websites with AI by just describing what you need. I believe this is the first move into this space by such a big, established player. (There are dozens of startups offering AI-powered website builders though.)
3. Adobe brings “Generative Fill” to Photoshop
AI-powered inpainting and outpainting isn’t exactly new.
But now Photoshop users have their own official version directly from Adobe.
Generative Fill lets users expand images, manipulate sections within them, and easily add new objects by using simple text commands.
Here’s a video walkthrough straight from the horse’s mouth:
4. Meta’s AI model is a super-polyglot
Meta’s so-called Massively Multilingual Speech (MMS) project is focused on preserving the world’s languages by training AI to recognize a huge range of spoken human language. The models are trained on a labeled data set of over 1,100 languages (and nearly 4,000 unlabeled ones).
Here’s a demo:
The big news is that Meta has just shared the model code publicly, hoping to spur further research and development. This could have a major impact on tools and advances within areas like translation, transcription, text-to-speech, and more.
5. Google isn’t exactly sleeping, either
Following its resurrection, Google made a splash with a slew of AI announcements during the I/O conference a few weeks back.
Now we’re seeing a trickle of these announcements being rolled out to end users.
First off, select users in the US should start seeing the Search Generative Experience (SGE) results for their Google searches:
Google also rolled out Product Studio for its merchants to help them create and modify product imagery with the power of AI:
🛠️AI tools
This week’s beginner-friendly AI tools. Today’s theme is “personal development.” I dug up a bunch of cool sites that might just make you a better person.
6. Yoodli
This one’s for anyone looking to improve their speaking skills, whether for presentations, job interviews, online meetings, or whatever else.
Yoodli can analyze a video of you speaking to provide a wealth of information about your pacing, word choice, eye contact, etc. It can even suggest alternative, more effective phrasing for your key points:
Here’s a feedback snapshot for an intentionally psychotic, hyper-fast rant of mine:
Another neat feature is an optional live AI coach that quietly listens to your side of a conversation and nudges you to slow down or stop rambling in real time.
The best part? Yoodli is 100% free.
7. interviewsby.ai
interviewsby.ai lets you prep for any job interview by automatically creating a set of realistic questions you might face when applying for a specific position.
It then gives you a 1-minute window to record your spoken answer, reviews it, offers helpful suggestions, and even provides an alternative sample response.
Here’s how that works for a totally real job of “magic spell inventor” that I told it I was applying for:
With proper practice, it looks like I’ll be ready to ace that next wizard interview!
8. learn.xyz
With learn.xyz, you can instantly create a tailored course for any topic, adjusted to your existing skill level. And when I say “any topic,” I mean it. Similarly to interviewsby.ai, this site generates the course based entirely on your input.
Which means, yes, you can learn about magic spell casting ahead of that important and decidedly non-fake job interview:
Using that simple input, learn.xyz creates a mini-course…
…complete with short lessons and quizzes to test your knowledge:
By far the coolest feature is the ability to share these AI-generated courses with the public. For instance, do you want to follow my new spell casting course? Here you go1.(Just don’t go applying for that magic spell inventor job. I really need it!)
💡AI tips
Here is this week’s tip.
9. Create tables with relevant info automatically
Most of you probably know that ChatGPT and Bing Chat can generate tables on demand. I personally love the fact that I don’t even need to specify the exact columns I need in my table.
When doing research and classification, you can often let AI figure out the most appropriate classifiers to include. Here’s a sample prompt:
Compile a list of 20 most popular land animals and generate a table that lists them plus has columns for any other parameters that help to classify the animals and understand the context.
That’s usually plenty for Bing to go on:
Not only does Bing do the research for you, it picks relevant descriptors and neatly arranges the data for easy comprehension. Sweet!
🤦♂️10. AI fail of the week
Never ask Salvador Dali to create a poster for your upcoming FIFA World Cup
Sunday poll time
EDIT 18-03-2024: Looks like learn.xyz has since removed my course!
DragGAN looks amazing, but I couldn't get the demo to work. 'No Nvidia driver found' error. Were you able to test it?
Ive been testing out Firefly (thanks for your previous tips on that) but does the new generative fill mean I'll have to trade my clunky old version of Photoshop for an expensive new upgrade?