🤖 Microsoft's mimicking model

It's been another crazy weekend in AI. There's no time to waste.

In today's edition:

• Microsoft unveils next-gen text-to-speech

• Meta teases HyperReel, a finally useful 3D app?

• Court history in the making: AI v. Human

Microsoft unveils VALL-E

Another day, another model to pore over 😛

VALL-E is a new language modeling approach for text-to-speech.

Microsoft trained the model using 60,000 hours of people talking in English, teaching VALL-E to talk on its own. VALL-E can make it sound like a real person said a piece of text, even if that person has never said that combination of words before — like magic.

With just three seconds of training input from any speaker's voice, you can create realistic-sounding speech in that same speaker's voice.

Here are some interesting potential use cases:

• Lost your ability to speak? Using previous audio recordings, replicate the sound of your voice and speak again.

• Bring emails/texts to life: have the sender read their message to you in their own voice.

• Children who won't read: audiobooks read by their favorite authors and characters.

Meta teases HyperReel

1) Capture a few different angles of video in any space.

2) Run HyperReel.

3) Step into a 3D representation of the space and take a look around.

🤯

I wish I could've used HyperReel to preserve my childhood home.

(HyperReel via Jim Fan)

AI will help defend a human in court

A court hearing is set to make history next month. Using an earpiece and DoNotPay's model running on their phone, the defendant will get live instruction on what to say in arguments.

"We help consumers fight corporations and beat bureaucracy, getting them out of parking tickets and refunds from big companies like airlines...our vision at DoNotPay is to make the law free," said Browder, who believes the technology will one day replace lawyers.

(Gizmodo)

🔥 Hottest Launches

• You.com - next-gen AI search, a Google killer? (link)

• Luma AI - capture 3D shots and edit in 3D, all on your iPhone (link)

• MinGPT - minimal PyTorch representation of GPT3 training (link)

• Explainthis.ai - a chat assistant in a Chrome extension (link)

• AI Project Description Generator - technical specs from a project description (link)

• EffortlessEmail - turn bullet points into fully formulated emails (link)

• NeevaAI - get cited, real-time info from a variety of sources (link)

🤠 Headline Roundup

• AI product roundup from CES 2023 (TC)

• Kickstarter kicks off Unstable Diffusion, which is trying to legitimatize AI porn (BI)

• Is Google's search dominance over? Chamath Palihapitiya weighs in (BI)

• Researchers glean AI insights from health records to tackle long COVID (AI Mag)

• Microsoft wants to integrate OpenAI's GPT into Office (The Information)

• AMD doubles down on AI with Ryzen 7000 (TechTarget)

• The state of AI in 2022 — and the past five years (McKinsey)

• Is Adobe using your photos to train its AI? Here's how to opt out (TC)

• GPT-3 provides mental health support to 4,000 people (Twitter)

• VIDEO: PimEyes, and the victims of public facial recognition (Vice)

💼 Hiring Corner

• You.com is hiring a Principal Product Designer

• LumaLabs is hiring for six technical roles and a Design Lead

That's a wrap for today. Stay curious and see you tomorrow! If you want more, be sure to follow me on Twitter (@jeremykuoo).