AI is making music now 🎸😱

Good afternoon, digital pioneers!

This week, OpenAI unwraps new features for us, revolutionizing the custom AI model scene. It's like finding the secret recipe to your favorite dish—now, everyone can cook up their own AI flavors.

Meanwhile, AI's venturing into music, hitting notes that have us all ears, and Meta's shining a light on deepfakes, offering clarity in the murky waters of digital content.

And for the curious minds: Ever wonder what GPT stands for? We're spelling it out in this week’s bite-sized brain booster.

Strap in—our Rundown is your express lane to the fast-changing world of AI, where every update is a potential game-changer.

Table of contents 📚

• OpenAI has given us some new features 🎁

• Template: your personal email assistant 📤🤖

• The weekly Rundown

  • AI is making music now? 🎸

  • Meta’s plan for deepfakes 🧭

• AI-classroom: what does GPT stand for?

OpenAI has given us some new features 🎁

For ChatGPT users
If you are using DALL-E on the paid version of ChatGPT, you have probably had a few times where you generated a very good image but just one tiny things was a bit off. When you tried regenerating the image it was completely different. OpenAI has released an update where you can edit generated images. Good job OpenAI!

Here is video of how it works. Be warned, they have included the painfully long load times.

For developers
OpenAI is dialing up the dials on customization with a fresh twist to its Custom Model program. This upgrade is all about giving customers the keys to tailor AI models to fit like a glove for their unique needs. Enter "assisted fine-tuning," a new suite of tools and techniques that goes beyond the basic tune-up, injecting some serious steroids into data training and evaluation, all designed to buff up model performance for specific tasks.

Take SK Telecom, for example, which teamed up with OpenAI to supercharge GPT-4 into a specialist in Korean telecom chatter. Or consider Harvey, the legal AI whiz that's now parsing case law with the ease of a seasoned attorney, thanks to a custom model feasting on a diet of legal texts and lawyerly wisdom.

OpenAI's vision? A future where custom AI models are the norm, not the exception, across industries and businesses big or small. And with the assisted fine-tuning toolkit in their arsenal, they're paving the way for companies to craft their AI alter egos, finely tuned to their operational and strategic nuances.

But wait, there's more—developers tinkering with GPT-3.5 also get a treat with new fine-tuning features. From a sleek model quality dashboard to third-party platform hookups and spiffed-up tools, OpenAI is making sure that when it comes to AI, one size doesn’t have to fit all.

🔗 Template: Your personal email assistant

This week we’ll share a very powerful and time saving template. A personal AI assistant that can manage your entire inbox or a specific folder.

Requirements:

• A make.com account

• OpenAI assistant

• OpenAI API key

• Email account (for the demo Gmail is used)

• PDF(‘s) with important information for the assistants “brain”

The step-by-step:

This one can be a bit technical to set up, so we have made a detailed and visual step-by-step.

• Template

• Step-by-step tutorial

The weekly rundown 📬

AI is making music now? 🎸

In the latest tech tune-up, Stability AI has dropped the beat with Stable Audio 2.0, an upgraded audio generation model that amps up the AI music game. Now, audiophiles and creators can upload their own sounds and use prompts to whip up AI-crafted tracks. But don’t throw out your playlists just yet—the feedback suggests these tunes might need a few more soundchecks. Described as "soulless and weird," listeners might find themselves navigating through "strange Gregorian whale noises" that feel more spectral than musical.

Here’s what’s new on the deck:

- Longer Tracks: Creators can now generate songs up to 3 minutes long, a jump from the previous cap of 90 seconds, setting the stage for more complex compositions.

- Access for All: Stepping out from behind the paywall, Stable Audio 2.0 is hitting the high notes of accessibility, available for free via Stability AI's website and API—contrast that with OpenAI's Voice Engine.

- Massive Training Library: With a dataset boasting over 800,000 audio clips from AudioSparx, the platform has made strides in variety. Artists had the option to sidestep the AI training session, ensuring their beats weren’t taken without consent.

- Quality Check: Despite the new bells and whistles, the melody might not be music to everyone's ears, with reviews highlighting a lack of natural flair in the AI compositions.

- Copyright Conscious: In tune with legal harmonies, Stability AI has buddied up with Audible Magic to ensure copyrighted tunes don’t accidentally slip into the mix.

While Stable Audio 2.0 marks a step forward in the symphony of AI-generated music, it seems we’re still a few verses away from hitting those chart-topping notes.

Meta’s plan for deepfakes 🧭

Meta's stepping up its game to clear the air around AI-generated content, rolling out new policies that'll make it easier to spot deepfakes and other digitally doctored delights across its platforms. Soon, videos, audio clips, and images crafted by AI will sport a "Made with AI" label, a heads-up for users navigating the blurred lines between real and rendered. But Meta's not stopping at labels. In cases where AI-generated content could mislead the public on critical issues, they're planning to amp up the visibility of these warnings, adding extra context to keep users in the know.

This shift away from outright removals marks a significant change in strategy, spurred by critiques from Meta's own Oversight Board. The board highlighted concerns that the old policy might be stifling free expression, pushing Meta towards a more transparent approach. Now, the focus is on enlightening users with labels and explanations, rather than erasing content from the platforms.

Behind the scenes, Meta is tightening its network of fact-checkers, ready to demote and debunk false or altered content, ensuring that misinformation doesn’t run wild. This policy overhaul is timely, too, as Meta gears up for the 2024 elections, aiming to foster a digital environment where truth trumps trickery, and users can trust what they see and hear.

👨🏻‍🏫 AI-classroom [Part 4] 🎓

What does GPT stand for?

GPT, or Generative Pre-trained Transformer, is a type of AI that's really good at understanding and creating text like a human would. It learns by reading a huge amount of information from books, websites, and other texts. After this big study session, GPT gets a feel for how words and sentences fit together.

There are two main steps in teaching GPT to do its thing:

1. Learning from lots of text: First, GPT goes through lots and lots of text. During this time, it's learning the basics of language, like which words are usually used together and how sentences are formed. This part is like reading lots of books to get better at writing.

2. Getting better at specific tasks: After learning the basics, GPT can be taught to do specific jobs, like answering questions, writing stories, or even making jokes. This step is like practicing a lot on one type of writing, like poetry or reports.

The cool part about GPT is that it can write text that sounds very much like it was written by a person, and it can be used for all sorts of things, like helping to write emails, making stories, or answering questions.

😁 Feedback 💡

Help us improve and send us your feedback! We love to here from you, mail us at: [email protected].