AI’s Eureka moment
Everything you need to know in AI this week
Welcome back! A quick PSA for all avid Midjourney users: The image generator is finally expanding beyond Discord to bring its service to browsers. In fact, Midjourney just launched the first version of its own website—it’s still in beta, but people are already loving it. Give it a try and let me know what you think!
Today, we’re diving into this week’s biggest AI stories. Let’s go. :)
AI’s Spin on Robot Learning
NVIDIA’s latest AI research breakthrough has everyone in robotics crying “Eureka!”…literally. NVIDIA’s new AI agent—aptly named Eureka—can teach robots complex skills on its own. And last Friday, NVIDIA revealed that Eureka’s robot training abilities outperform humans’.
In a nutshell: Eureka is an autonomous agent (AKA it operates without human intervention) powered by GPT-4. It equips robots with the ability to handle complex skills with increased precision and autonomy, which promises to fundamentally change the way robots learn.
The secret sauce? Reinforcement learning (RL):
Eureka uses GPT-4 to automatically translate human language descriptions of tasks (e.g. “open a drawer”) into reward functions. Essentially, these functions reward robots when they perform a desirable outcome for a task (e.g. actions that lead to the drawer being opened).
Until now, humans were responsible for writing these reward algorithms—a complex and time-consuming process.
Eureka reportedly outperforms human-written algorithms on more than 80% of tasks, which could seriously accelerate robots’ learning capabilities.
How will this be applied? Right now, Eureka has been used to teach robots low-level manipulation tasks, such as pen-spinning tricks.
But on a larger scale, NVIDIA’s breakthrough may foreshadow an acceleration in robotics advancements—which have seriously lagged behind the pace of AI. While hardware challenges are still a bottleneck, autonomous RL algorithms—like Eureka—could supercharge robots’ learning capabilities.
Today’s Future Tools brought to you by:
The only personal AI assistant you'll need.
Automatically transcribe and summarize all your calls regardless of application (Zoom, Gmeet, MS Teams, Facetime, Whatsapp...). All, without having a bot 🙄 join the meeting.
Read twice as fast. Have AI summarize newsletters, emails, documents & PDFs.
Type less. Save time, not tone. A writing assistant who understands the audience: your team vs your toddler.
🗞️ Other AI News & Articles
Artists’ New Tool to Fight Image Generators
Cyanide and ricin won’t poison AI models, but a new tool called Nightshade can—and it could change the way big AI companies interact with artists.
Nightshade—introduced on Monday by the MIT Technology Review—is a tool intended to help artists fight back against AI image generators that were trained on their artwork without permission.
How it works: Prior to uploading their art online, an artist can use Nightshade to add invisible pixels to their image. When AI models (like DALL-E, Stable Diffusion, or Midjourney) are trained on data sets which include these “poisoned” samples, they get seriously confused—and believe that images depict things they actually don’t (i.e. “poisoned” data sets can trick models into learning, for instance, that images of dogs depict cats).
Because the corrupted images look identical to benign images with matching text prompts, Nightshade “attacks” disable image generators’ ability to generate meaningful images (e.g. prompting a model to generate an image of a cat will result in one of a dog instead).
Big impact: In the absence of AI copyright regulation, Nightshade is a tool artists can use to fight back against AI companies. If enough artists engage in data poisoning, the power could shift in favor of artists—and AI companies may have to start coughing up some royalties.
Apple’s AI Plans
While OpenAI, Microsoft, and Google have been busy rolling out new AI innovations, Apple’s been quietly sitting on the sidelines. But a new Bloomberg report reveals that Apple may be cooking up some generative AI plans.
Apple executives were caught off guard by last year’s AI wave, a fact that’s widely considered “a pretty big miss internally”, according to Bloomberg.
Scrambling to make up for lost momentum, Apple plans to ramp up its spendings on AI development—to $1 billion a year.
AI integrations into Siri, Apple Music playlists, auto-complete features in Messenger, and developer tool Xcode are also in the works.
Zoom out: Unlike other Big Tech titans who are focusing on building foundational models, Apple’s approach to AI is product-focused. The company keeps emphasizing its goal of running AI on-device (eliminating the need for expensive cloud servers that raise privacy concerns). iOS 17’s new autocorrect feature, which runs directly on the iPhone, gives a small taste of how Apple plans to embed AI into its hardware.
Either way, Apple already lost first-mover advantage in the AI race, which begs the question: When Apple’s AI products finally do hit the market, will they make up for the lag?
Chinese researchers developed Woodpecker, a solution to AI’s hallucination problem.
OpenAI, Anthropic, Google, and Microsoft are teaming up on AI safety.
Google Pixel’s new face-altering AI photo editing tool is sparking serious controversy.
Bill Gates believes that generative AI has reached its ceiling.
YouTube now lets you make your own playlist art with the help of AI.
📺️ This Week’s Videos to Watch:
More important AI news: Dive deeper into this week’s hottest AI news stories (because yes, there are even more) in my latest YouTube video:
DIY chatbot: Building your own AI bot isn’t as complicated as you might think. 👀 Here’s the step-by-step guide:
And there you have it! I’ll be back in your inbox on Wednesday with five new AI tools. In the meantime, Happy Halloween to all the avid trick-or-treaters out there! Stay safe and keep an eye open for any AI-themed costumes. :)
If you haven’t already, be sure to follow me over on X. I like to share news and AI updates as they happen over there. And don't forget to check out all of the newest tools we've just added on Future Tools!
You rock! See ya next week. :)
P.S. This newsletter is 100% written by a human. Okay, maybe 96%.