Google makes music from text

In today’s email:

• Google MusicLM: A look into generating (pretty good) music from text

• Microsoft & Open AI in Court: Microsoft, GitHub, and OpenAI ask the court to throw out AI copyright lawsuit

• Visual Art: 3 Visual Art examples from Midjourney in this email

• Who to follow on Twitter

Google MusicLM

Although you can’t play with it yet, Google released a glimpse of what they’ve been working on: Audio Generation From Rich Captions

MusicLM is a model generating high-fidelity music from text descriptions such as "a calming violin melody backed by a distorted guitar riff".

On the github page, you can see some examples of the music that the model created from text (for example: The main soundtrack of an arcade game. It is fast-paced and upbeat, with a catchy electric guitar riff. The music is repetitive and easy to remember, but with unexpected sounds, like cymbal crashes or drum rolls.)

But it goes further than that. If you scroll down to Text and Melody Conditioning, you might start to get a little bit creeped out by what it can do. MusicLM can generate music that respects the text prompt while following the provided melody (i.e. “bella ciao - jingle bells - whistling”). Things are gonna get quite interesting for our ears in 2023.

The model can also create music from images, by first interpreting the image into text and then that text into music.

I must say, the output is pretty good. It doesn’t feel as revolutionary as ChatGPT for example, but that’s because you just can’t play around with it yourself. Yet.

Keeping in mind that music is quite structured (i.e. most of the hit songs follow a similar pattern), I think this will produce much better output much faster than, say, AI-Generated Video Art.

So, do you think that we will have a Billboard 100 song composed by AI in 2023? I think so…

Microsoft & OpenAI in Court

Microsoft and Open AI want the court to toss a class-action lawsuit against them, accusing them of abusing open-source code.

Microsoft (owner of GitHub), used code from repositories on GitHub o build GitHub Copilot.

Problem? The lawsuit alleges that they used GitHub repositories without complying with open-source licensing terms and that Copilot unlawfully reproduces their code.

Microsoft and Open AI say that the allegations are not specific enough and cite fair-use defense.

I’m taking two points from here:

• This will be the first of many lawsuits. As AI grows into more fields, the creators of the data that will be used to train the AI may not have given consent to use it for that specific use case. I do wonder if the legal systems around the world are sophisticated enough for this quickly-evolving change (similar to Crypto legislation which is lagging behind)

• The winners of this technological shift will be the owners of non-open-source datasets. As tools get trained on similar, open-source data, they will have similar outputs as well. The differentiating factor will be the underlying data that they use - think Pharma companies training their own AI on their own studies and test results. I still think Google will be very strong here, just because they have a TON of our data (transcribed Youtube, Gmail, Google Drive etc.)

AI Tools to Try:

Excel Formula Bot - Transforms your text instructions into an Excel or Google Sheets formula in seconds with the power of AI. I’ve been working with Excel for over 15 years and this can be a lifesaver.

Writesonic - an AI writing assistant for the web. Better than Copy.ai in my opinion.

Fireflies - Automate your note-taking by helping your team record, transcribe, search, and analyze voice conversations.

Murf - Text to speech. Was the #1 Product of the Week on Product Hunt.

Visual AI:

Some amazing Art Generated via Midjourney:

If you want me to write a Guide (or record a video) on how to get good results in Midjourney, Let me know!

Who to Follow:

PS: If you are working on an AI product, I’d love to test it out!