AI Roundup 041: Lyria

November 17, 2023.

Nov 17, 2023

Lyria

After months of rumors, DeepMind and YouTube released Lyria - a music generation model. The model will power Dream Track, a tool to generate entire songs in the style of specific musicians like Charlie Puth, T-Pain, Demi Lovato, John Legend, and more.

Why it matters:

Creating music in the style of known artists will become a big industry, and YouTube is paving the way with its (licensed) tools.
It's also sidestepping whether or not AI-generated songs violate copyright by working directly with musicians. As a bonus, music made with Lyria will be watermarked for future detection.
Features like Dream Track are helping Shorts compete with TikTok - instead of using the same song as everyone else, why not create a bespoke track from your favorite artist?

Refer a friend

Elsewhere in foundation models:

DeepMind also released GraphCast, an AI model that's better at predicting the weather than conventional systems.
Meta unveiled Emu, two new models for generating videos and editing images.
And new reports indicate that Google Gemini may not be available to cloud customers until Q1 of 2024, instead of this November as originally planned.

Microsoft Ignite

Another week, another developer conference full of AI launches. This week was Microsoft Ignite, the annual IT and developer gathering in Seattle, Washington.

All the announcements I could find:

Bing Chat is rebranding as Microsoft Copilot.
Copilot Studio, a no-code tool to make GPT-style customizations.
Windows AI Studio, a tool for developers to tweak AI models.
Early release of Copilot for Windows 10 Home and Pro.
AI updates for Microsoft Teams, including AI backgrounds and voice isolation.
Azure AI Speech avatar, a.k.a. a deepfake creator.
Upgrades to Bing search, like auto-generated captions and search result analysis.
AI-powered customer services for users who are blind.
Custom designed chips - Maia 100 for AI, and Cobalt for cloud servers.
And an interview with CEO Satya Nadella on his relationship with OpenAI, the AI-fueled creative revolution, and more.

Elsewhere in the FAANG free-for-all:

Google rolls out shopping tools to its Search Generative Experience, and upgrades Google Photos with new AI-powered features.
YouTube plans to require disclosure when users upload realistic-looking synthetic content, starting next year.
And Adobe is working on a new AI tool to isolate voice recordings from background noise.

Responsible Innovation Labs

A coalition of venture capitalists and tech executives have formed Responsible Innovation Labs, a non-profit pushing for founders and investors to sign its "Responsible AI Commitments." These guidelines promote transparency, forecasting, and auditing in the name of AI safety.

Between the lines:

There is a growing schism among Silicon Valley thought leaders - those who believe AI poses a threat and welcome regulation, and those who want only to accelerate innovation.
For years, the loudest voices on AI were concerned about AI's potential dangers. Now, a new ideology - e/acc - is pushing back against perceived threats to AI development.
And this struggle is also playing out across governments - as the UK backs off from AI regulation, the EU is struggling to get its AI Act past the finish line.

Elsewhere in AI anxiety:

Common Sense Media warns parents that MyAI, DALL-E 2, and Stable Diffusion may not be safe for kids.
Some companies that provide AI data-labeling services are inadvertently hiring minors and exposing them to inappropriate content.
And a lawsuit alleges UnitedHealth, the largest health insurance company in the US, is using an AI model with a 90% error rate to deny care to elderly patients.