Meta's Voicebox Tease

PLUS: DeepMind pwns noobs


Good morning, human brains. Welcome back to your daily munch of AI news.

Here’s what’s on the menu today:

  • ChatGPT x Mercedes 🚗

    Mercedes beta-tests ChatGPT functionality in their cars.

  • Meta’s Voicebox tease 😰

    Meta brags about their new AI speech generator, but won’t spill the tea on how they did it.

  • AI’s gaming extraordinaire 👾 

    Deepmind’s new model masters Atari games 500 times more efficiently than others.


ChatGPT in Mercedes cars

Mercedes announced ChatGPT is now integrated in its vehicles. Features include voice command, contextual conversation, and integration with third-party apps.

“Hey, Mercedes.” — that’s actually what you say to use it.

What it do? It’s a voice assistant that engages in conversation, gathers info on destinations, books reservations, etc.

The beta program works in over 900,000 Mercedes-Benz vehicles with the MBUX infotainment system.

Our take: With Tesla’s self-driving AI and Mercedes rolling out ChatGPT functionality, vehicle manufacturers that don’t incorporate AI may get left behind. But while AI in cars is useful, it could become yet another distraction. Hello, more safety concerns.


Non-Deterministic Mapping

An algorithm that produces different outputs for a single input.

This is the opposite result of a deterministic algorithm which takes consistently produces a single output for any given input, every time.

Example: Like a dice toss gets many different results, an AI speech generator learning from non-deterministic mapping can train on huge amounts of data without it needing to specifically labelled.


An emotional AI voice generator

Lovo uses AI to generate lifelike human voices. No more dull, robotic voiceovers for your projects; Lovo has 400+ voices with support for 100+ languages.

But here's the kicker: Lovo doesn't just spit out words - it does emotions. So you're not just setting the language, you're also setting the mood 😏 

  • 💵 It’s cheap. Hiring voiceover artists or maintaining a multilingual team is expensive. Lovo starts at $19/month.

  • 🏃 It’s fast. Producing a high-quality voiceover with human actors can take months. Lovo takes seconds.

  • 💌 It resonates. This is no Microsoft Sam. Lovo-generated voices have human emotion.

Best of all? It’s free to try.


Meta Voicebox 🔊

Meta announced Voicebox last week, a new state-of-the-art generative AI for voice audio.

It’s the first of its kind to complete speech generation and speech editing tasks in a generalized way — tasks it wasn’t specifically trained for.

Voicebox can:

  • Edit audio seamlessly

  • Transfer the style and cadence of one audio sample to another

  • Synthesize new audio based on text and an input audio clip

  • Remove background noise

No more screaming children in the background, phew.

That’s the good news. But Meta isn’t releasing the model to the public, at least for now.

“There are many exciting use cases for generative speech models, but because of the potential risks of misuse, we are not making the Voicebox model or code publicly available at this time.”

— Meta

Voicebox is built on Meta’s Flow Matching model. It’s particularly adapted to learn non-deterministic mappings between text and speech, which means it can learn from a wide range of speech data without careful labeling.

Our take: Is it even possible to ensure a tool like this isn’t abused? If this tech isn’t publicly accessible, who gets access to it?


New AI is a master gamer 👾 

Google DeepMind’s new AI model achieves human-like learning on Atari games. The BBF (Bigger, Better, Faster) model learned 26 games in 2 hours.

“gg noobs”

Sure, other AI models have mastered these games. But the BBF model outperforms previous models and requires 500 times less computing power.

“Overall, we hope that our work inspires other researchers to continue pushing the frontier of sample efficiency in deep RL forward, to ultimately reach human-level performance across all tasks with human-level or superhuman efficiency“

- Google Deepmind researchers from Mila and Université de Montrèal

Our take: “All tasks“ and “superhuman” — those are some big goals. Real life is a lot more complex than video games, so how will DeepMind bridge the gap to get there?



Think Pieces

A Stanford University article on the Draft EU AI act. In-depth analysis of which AI models comply with the latest regulations.

Thinking of starting an AI startup? Harvard Business Review covers what to consider before becoming an AI entrepreneur.

Portland’s Bureau of Emergency Communications launches AI. The AI will start answering non-emergency calls to tackle high call volume.

Startup News

Windows to the soul, and heart disease too. Google’s new health-tech AI evaluates a patient’s cardiovascular risk through retinal imaging.

HuggingFace launches OpenLLaMA. An open-source version of Meta’s LLM LLaMA.


MAGICBRUSH, an AI model that edits images based on text prompts. How it works, its effectiveness, and potential issues.


LoopGenius — Grow your business the easy way using AI-powered marketing loops. [Sponsor]

Personaliz — Your personal AI email marketer. — Easily create ads for any business with AI.

Definite — Analyze data quickly and effectively with AI.


ChatGPT writes a cryptocurrency joke. It doesn’t quite land:


If you like Bot Eat Brain, you might like these other newsletters too:

💼 Market Briefs — A free financial newsletter for regular investors. Every morning they break down the top news stories that could affect your finances.

🐙The AI Solopreneur — Helping solopreneurs capitalize on the AI revolution and 10x their productivity with cutting-edge hacks, prompts, and updates.

🌆 The Land of RandomRediscover the magic of the Internet! Learn about cool new tech, marketing, AI, entrepreneurial ideas, and other random internet stuff in bite-sized bytes.


Until next time 🤖😋🧠

What'd you think of today's newsletter?

Login or Subscribe to participate in polls.