Meta's Audiobox: The AI-Powered Ventriloquist Transforming Voice Cloning and Ambient Sound Generation

So, here's the scoop: Audiobox is Meta's answer to the growing voice cloning trend. You know Meta, the bigwig behind Facebook, Instagram, and those snazzy Oculus VR goggles. They've been cooking up this tech in their Facebook AI Research lab, and it's pretty nifty. Picture this: You type in a sentence or describe a sound, and voila, Audiobox brings it to life. Fancy having your voice cloned? Audiobox has got your back. And here's the kicker: It's free!

But wait, there's more! Meta isn't just playing a one-note tune here. They've created a whole "family" of audio-generating AIs. One mimics speech, and the other whips up sounds like barking dogs or children playing. All of this is built on their self-supervised learning model, Audiobox SSL. In layman's terms, it's like teaching AI to teach itself – no need for hand-holding with labeled data.

The brains at Meta have published a paper explaining their approach, which boils down to: "We don't always have quality labeled data, so we're training our model with audio sans supervision." They've fed it a smorgasbord of audio: 160K hours of speech, 20K hours of music, and 6K hours of sound samples, covering a vast linguistic and cultural spectrum. Though, where they got all this data is a bit of a mystery, and with the current legal kerfuffle over AI and copyrights, it's a question worth asking.

Ready to clone your voice? Meta's got some interactive demos to show off Audiobox. I gave it a whirl, and my cloned voice left my wife bewildered. It's almost spot-on but not quite there yet. And if you fancy creating a new voice from scratch, Audiobox is your playground.

Interestingly, despite Meta's usual open-source enthusiasm (ahem, their Llama 2 language models), Audiobox is keeping its secrets close to the vest.

In the grand scheme of AI, Audiobox is just the tip of the iceberg. With AI tech sprinting forward, I'm betting on seeing commercial versions sooner rather than later. So, for now, Meta's keeping things non-commercial and geographically limited.

AI Helmet Translates Thoughts into Text

Alright, let's dive into something that sounds like it's straight out of a sci-fi novel but is very much real. We've previously explored devices like the Halo, which reads dreams, but the latest advancement in AI and brain activity is on another level. Scientists have developed a helmet that can translate thoughts into written text. Mind-blowing, right?

This groundbreaking technology comes from the University of Technology Sydney, where Chin-Teng Lin and his team are pushing the boundaries of what's possible with AI. They're using what's known as an electroencephalogram (EEG) – a fancy term for a device that records brain activity. This cap captures your thoughts and, with the help of an AI model called DeWave, turns them into readable text.

Now, it's important to note that this technology is still in its infancy. The accuracy rate hovers around 40%, though recent updates suggest it's improving. But even with these limitations, the implications are enormous.

What sets this apart from earlier methods is its practicality. Unlike MRI-based approaches that require you to remain motionless in a scanner, this EEG helmet is non-invasive and much more user-friendly. Participants in the study were able to read sentences silently, and the system was still able to capture and convert these thoughts.

The mechanism behind this is fascinating. As Charles Zhou, another member of the UTS team, explains, when you think of a word like 'hello,' your brain emits specific signals. DeWave learns to associate these signals with the corresponding words. The system is then linked to a large language model, akin to what powers ChatGPT, which acts as an intelligent writer, turning these signals into coherent sentences.

The potential applications are vast, particularly for individuals who have lost their ability to speak, such as stroke survivors. It could also mark a significant step forward in the field of robotics, bridging the gap between human thought and machine interpretation.

Craig Jin from the University of Sydney commends this development. The transformation from previous EEG-to-text attempts, which were largely nonsensical, to this relatively accurate interpretation is a notable advancement.

We're witnessing the early stages of a technology that could redefine communication and interaction. While it's not perfect yet, the progress made so far is a testament to the rapid advancements in AI and neuroscience. Keep an eye on this space – the future of mind-reading AI looks promising.

AI Political Callers: Welcome to the
Shit-nado of Campaign Spam Calls!

Meet Ashley, the AI campaign caller who's making waves in the political scene. Democrat Shamaine Daniels is using Ashley to try and snag a seat in Congress, and this isn't your grandma's robocall. Ashley is a chatty AI, having customized, infinite conversations simultaneously. The tech behind her? Think ChatGPT, but for political cold calling. Yep, we're living in the future, folks.

But here's where my blood starts to boil – this whole setup skates around laws that businesses must follow. You know, those pesky rules about automated calling systems? Yeah, politicians are doing the cha-cha slide right around them. And they're not even required to comply with the Do Not Call Registry. Hello, loophole city!

Let's call this what it is – a potential spam call tsunami. We're not just talking a trickle of calls here. Civox, the mastermind behind Ashley, is planning on scaling up to tens of thousands, then hundreds of thousands of calls a day. I'm channeling my inner Jim Lahey here – it's not just a storm coming, it's a full-blown "shitastrophe."

Now, I'm usually not the one shouting for more government regulation, but this is one area where I'm waving that flag high and proud. When it comes to political campaigns, AI should take a backseat. We don't need deepfake-level trickery in our election processes.

And here's another kicker – while the creators are trying to be ethical (Ashley has a robotic voice and identifies herself as AI), let's not kid ourselves. The potential for abuse is massive. Other companies could easily create similar AI systems, and who's to say they'll play by the same rules?

The law is playing catch-up here, and it's not looking good. The Federal Election Commission and the Federal Communications Commission are just starting to figure out how to deal with this. And let's face it, the pace of technology is sprinting, while the law is stumbling along in a three-legged race.

So, what's the verdict? While AI like Ashley could level the playing field for underdog candidates, the risks are sky-high. We're talking about the integrity of our political system, and that's not something to be toyed with.

This whole AI political caller thing is a can of worms that's just been cracked open. It's a brave new world, but maybe it's one we should step into with a bit more caution. What do you think? Are we ready for AI to take on the role of political persuader, or is this a tech step too far? Let's just hope we don't end up in an episode of "Black Mirror" because of it.

Laredo Labs Leads the Charge in AI-Powered Coding Revolution

In the fast-paced world of software development, Laredo Labs is emerging as a potential game-changer. Co-founded by AI veterans Mark Gabel and Daniel Lord, this startup is pushing the boundaries with an AI-driven platform for code generation. Imagine a world where developers issue commands in plain English, and the AI whips up the code. That's the future Laredo is working towards, leveraging a model trained on a staggering hundred million software projects.

But it's not just about fancy tech. Gabel, a former chief scientist at Viv Labs (acquired by Samsung), and Lord, previously with Siri, are combining their AI and software engineering prowess to reshape how we approach coding. They're not just building another tool; they're crafting an entirely new developer experience. The platform, currently in private preview, is designed to handle complex 'repository-level' tasks, simplifying the developer's workload significantly.

However, the road ahead isn't without its challenges, especially when it comes to the legal landscape. Remember the controversy surrounding GitHub Copilot? The AI tool faced legal scrutiny over potential intellectual property violations, raising questions about the use of AI in coding. Laredo, while innovative, might have to navigate similar issues, particularly around the use of copyrighted code in training its AI models.

Despite these potential legal hurdles, Laredo's approach is ambitious, aiming to carve out its niche in a market brimming with possibilities. With $8.5 million in seed funding and plans to expand their team, Laredo is positioning itself as a major player in the evolving landscape of AI-driven software development. It's a space to watch, with Laredo potentially leading the charge in transforming how we think about and execute coding.