Google reads music — from your mind

PLUS: PhotoGuard from MIT


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

Here’s what’s on the menu today:

  • China’s tiny new GPTs 🤖

    The Beijing Academy of AI’s open-source models: small but capable.

  • Google gets all up in your musical brain 🧠 

    Turning your thoughts into music, with a bit of fMRI sleuthing.

  • MIT white-knights artists 🛡️ 

    New tech to protect artists from intellectual property theft.


China’s fun-sized GPTs 🤖

The Wu Dao 3.0 models include:

  • AquilaChat Dialogue: a 7 billion parameter dense model. Outperforms similar-sized open-source models domestically and internationally.

  • AquilaCode: designed for text-to-code generation.

  • Wudao Vision series: focuses on issues in the computer vision field.

As the old saying goes, “For every job, there is a tool.”

With recent models pushing parameter counts into the trillions, these compact models are bucking the trend. What gives?

Smaller models are more cost-effective to produce, easier to commercialize, and require fewer computational resources. And with US chip sanctions against China, there’s been a shortage of AI chips.

Our take: there’s a natural “speed limit” on the size these models can get. At some point, it becomes impractical to keep increasing parameter size — and that’s the reaction we’re seeing here. Even Meta’s LLaMA2 clocks in at 70 billion parameters — and it’s quite capable, despite lying well below the 1T mark.



In AI, perturbations are small disruptions made to the data that's fed into a foundational model or to the model itself.

Introducing perturbations allows us to better understand how the model works or as a way to manipulate the model to a desired behavior.


Ask questions and get answers about any file instantly.

Ever been stuck analyzing a long-ass document you just wish someone else could summarize for you?

Ever had a report due tomorrow you just couldn’t figure out how to start writing?

Enter, Humata.

📜 Summarize long papers: Learn Faster. Turn complex technical papers into simply explained summaries. Discover new insights 100X faster.

🤔 Instant Q&A: Answer hard questions related to your file. Get easy-to-understand answers instantly.

✏️ Write papers 10X faster: Automatically create new writing based on your file. Generate detailed insights for reports, papers, and a variety of tasks instantly.


Google reads music — from your mind 🧠

Their research might lead to intelligent music composition software that converts your thoughts into melodies.

You know how hard it is to get melodies down on paper or into your DAW, before they disappear? If you know, you know.

Here’s the method they used:

  • A fMRI scanner monitored brain activity of five volunteers listening to 500 tracks in 10 musical styles.

  • The Brain2Music AI model used brain data with Google's MusicLM to create similar songs.

  • Music generated by the AI matched the genre, instrumentation, and mood of the original stimuli.

Shazam on ‘roids!

Check out the full findings here.

Our take: buckle up! We’re already starting to see AI-powered mind-reading products like these headphones. If the tech gets good enough, we won’t need an invasive procedure (a la Neuralink) to establish a brain-computer interface.

After all, you’re already effectively a cyborg — just limited by typing speed.


MIT sticks up for artists 🛡️ 

Swiper, no swiping.

How it works:

  • PhotoGuard uses tiny pixel changes called "perturbations" to disrupt AI image manipulation.

  • The changes are invisible to the human eye but detectable by computers.

  • PhotoGuard employs two attacks: "Encoder" and “Diffusion”, which confuses the malicious AI.

  • The result: AI that attempts to use an immunized image produces a botched job. No more AI manipulation — fingers crossed 🤞 

Our take: impressive piece of kit, but we’ve been here before…

Virus vs antivirus.

How long can PhotoGuard hold the fort before the red team figures out how to work around perturbations?



Think Pieces

Resurrecting extinct antibiotics — researchers used AI to find novel antibiotics based on common proteins between humans and Neanderthals.

Deepfakes and the 2024 US election — how real of a threat are deepfakes to the democratic process?

Riding the AI wave — how long will it last?

Startup News

Indian startup leverages AI to help the poor, through dataset creation — Karya already has huge clients such as MIT, Microsoft and Stanford.

Startup lands funding to use AI in K-12 curriculum - the Vietnamese-based VUIHOC aims to serve 1.1M students.

Bringing Hollywood cameras to your smartphone — Voia raises $3M in private investments to enable anyone to produce Hollywood-quality videos.


AI machine detects lies — get your best poker face on. AI is coming to call you out on your BS!

AI traces emotional contagion — the new method could predict burnout and stress in social workers and psychologists.

AI detects disease from MRI scans — it could help doctors by catching signs that might otherwise slip through.

Tools [Sponsored] — ask questions and get answers about any file instantly.

Murf.AI — over 110 voices in more than 15 different languages, so you can create studio-level audio

Tabnine — code completion assistant that learns the codes and preference of your team while you use it

Soundraw — create unlimited, royalty-free music.


AlphaFold is slowly making a dent in the cancer problem. Boutique cancer drugs incoming?

Tag us on Twitter @BotEatBrain for a chance to be featured here tomorrow.


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

🏥 Healthcare AI News — 5 minutes or less is all it takes to elevate your Healthcare knowledge with this expert-curated weekly AI newsletter. Stay informed and stay ahead.

👨 The Average Joe — Market insights, trends, and analysis to help you become a better investor. We like their easy-to-read articles that cut right to the meaty bits.

💌 Marketing Letter — The newsletter keeping 30k+ marketers in the loop. Read by marketers who work at LinkedIn, Techcrunch, and Disney.


@CadizFFM- Untitled

Until next time 🤖😋🧠

What'd you think of today's newsletter?

Login or Subscribe to participate in polls.