🎙 Insane Text to Speech Tool May Change The Game

“Unbelievable” is one of the most overused and misused words in the English language. People call events “unbelievable” when they’re just unusual or surprising.

That said, what is happening with artificial intelligence in late January 2023 is freaking unbelievable.

In the email today:

• The text-to-speech app to rule them all? 🎙️ 

• Tool of the Day ⛏️ 

• BREAKING NEWS: Baidu launching ChatGPT clone in China 🇨🇳

• Headlines 📰 

• OpenAI hires 1,000 contractors to label data 🏷️

• Links 👀 

This was not an uneventful weekend.

ElevenLabs Unveils What May Be The Best Text-to-Speech Tool 🎙️

Polish engineers Mati Staniszewski and Piotr Dabkowski recently raised a $2 million pre-seed round for their startup, ElevenLabs, which unveiled a text-to-speech generator called Prime Voice AI.

The tool is rather remarkable and perhaps the best text-to-speech tech SmokeBot has seen.

This is tough to demo in an email, but listen to these examples of ElevenLabs’ tech:

There are a bunch of other examples floating around Twitter, which is buzzing about the app. You can check them out here.

The interface is simple and easy to use— just enter in text and choose the voice you want to use. You can also, and here’s the rub, upload your own voice and use that.

Anyone can try a handful of short samples before being prompted to sign up for a free account or a $22 per month creator account.

Per the founders, the tool “could be used for everything from creating audiobooks to dubbing movies” through a deep learning model for speech synthesis.

The possible beneficial uses of this tool are numerous and obvious.

So too is the potential for misuse. 

The tool could take your voice and say things that you would never think, much less say out loud.

ElevenLabs is already addressing that issue this morning:

Proposed remedies include identity verification for Voice Cloning, verifying copyright, or manually reviewing each request.

This may be the tool to watch right now.

Tool of the Day 🔨

It’s ElevenLabs' Prime Voice AI.

Wow.

BREAKING NEWS: Baidu Will Launch Its Own ChatGPT in China, and, yep, Censor It 🇨🇳

The details, per a Wall Street Journal story:

• Baidu is developing an AI-powered chatbot similar to OpenAI's ChatGPT and plans to integrate it into its main search engine in March

• They are set to be the first to bring the technology to consumers in China, where access to ChatGPT is blocked

• The chatbot is based on the same algorithm that Google developed in 2017, which OpenAI used to build ChatGPT

• Baidu has focused on natural-language processing and has developed a deep-learning model called Ernie atop Google’s original algo which is now being used as the foundation for its chatbot

• It’s being trained on both Chinese and English language sources

• Baidu plans to limit its chatbot’s outputs in accordance with the state’s censorship rules, one of the people said

A censor manipulated chatbot. What could go wrong?

Headlines 📰 

Microsoft and OpenAI jump the route: Per Reuters, Microsoft and OpenAI are trying to head off a potential class action lawsuit at the pass. Specifically, they “told a San Francisco federal court that a proposed class-action lawsuit for improperly monetizing open source code to train their artificial-intelligence systems cannot be sustained.”

Google has a very cool tool but they’re not letting anyone play with it yet: We featured Google’s new text-to-music tool in our last send, but that might be all you hear from it for a while. Google, worried about the risk of using the tool and keeping with their general cautious approach to releasing AI to the masses, says it has “no immediate plans” to make it available for public use. Well that sucks. We were hoping to have this thing compose a “Mr. Brightside”-esque banger in Miley Cyrus’ voice. Someday, maybe.

AI developed bacterial-killing proteins: Profluent, a biotech startup in in California, used AI to develop anti-microbial proteins that actually work. The technology could eventually be used to make medicines.

OpenAI Hires 1,000 Contractors To Label Data 🏷️

According to this Semafor piece:

• OpenAI has contracted 1,000 people in places like Latin America and Eastern Europe to strengthen not only ChatGPT but all of OpenAI's platforms

• Approximately 60% of the contracted workers were hired to perform "data labeling," i.e., creating huge compilations of images, audio clips and other information that can be "used to train artificial intelligence tools and potentially autonomous vehicles"

• The balance of those brought on are computer programmers who are "creating data for OpenAI's models to learn software engineering tasks"

• OpenAI's existing Codex product, which launched almost a year and a half ago, "is designed to translate natural language into code"

In the last send, we mentioned how the Pentagon’s Chief AI officer worried about America’s ability to use humans to label data at scale vs. China’s ability to do so with basically slave labor.

Keeping with that theme, it’s no surprise that the hired workers here are from countries where labor is much cheaper.

Are we quickly approaching a world where human minds are used to feed the machine? Have we seen this movie before?

Maybe we have.

Links 👀 

• Hospitals are already using AI to collect and analyze images and the tech is ready to be deployed at scale in 2023 🏥 

• AI rockets ahead without regulation ⚖️

• AI can be trained to pick off “deep fakes” and none too soon 🤲