💀 AI Can Predict When You Die

Hello, fellow nerds 🧙

Hopefully, everyone had a great Easter! Nothing better than eating copious amounts of chocolate bunnies —One of our guilty pleasures over here at The Daily XP😅 

Speaking of chocolate bunnies, here’s the nightmare fuel that we get to look at:

Here’s the daily quests in AI to complete today:

• 💀 AI Can Predict When You Die

• 📠 SAFE Tool to Check Facts, Bro. Fax.

• 🎂 The Matrix Turns 25 Years Old

• 🖥️ Stargate is The World’s Largest Supercomputer

Danish Researchers Create AI That Predicts Your Entire Life (and Death)

Some scientists in Denmark have been cookin’ up an AI that claims to predict your entire life story, right down to when you'll kick the bucket.

They call it "Life2vec”.

The main idea behind Life2vec is to uncover patterns and relationships that could predict everything from your health to your wealth. Seems pretty crazy, right?

OK, how in the hell does it work?

Well, turns out if you feed an AI with data from over 6 million people (the entire country of Denmark), it can get really good at predictions. And we mean, really good.

So good, that it can predict death with 78% accuracy and even guess if you'll move to a new city or country with 73% accuracy. The researchers focused on people between 35 and 65 to test their AI's reliability, because apparently, that's when our bodies decide to fall apart.

Super pumped for that 🙃

Now there are some risks to developing AI like Life2vec. For instance, Pernille Tranberg, a Danish data ethics expert, says businesses like insurance companies are already using similar algorithms to put people into risk groups. And if they decide you're too much of a liability? Good luck getting coverage or a loan.

To be honest, this all sounds like one hell of a Black Mirror episode. At least we get to be the stars.

SAFE Tool to Check Facts, Bro. Fax. 

Big brains at Berkeley and Google DeepMind developed a method of factual accuracy of responses called SAFE (search-augmented factuality evaluator tool).

SAFE is like a nerdy word algorithm.
It will take an answer, break it down into bite sized individual facts, check each fact for relevance using an LLM, and then verify each fact with Google search.

GPT-4 Turbo and Gemini Ultra had the best results when considering both accuracy and comprehensiveness of answers.

Would you call it PVP or PVE?
Come to find out, with access to the internet, when SAFE and human annotators go head to head… they agreed 72% of the time.

What happens when they disagree?
If a human doesn’t know something, we try to figure it out by using our best judgment and come up with a mixture of imagination, estimation, and validity. Ya know, if we’re lucky, there’s a sprinkle of truth on top.

Yet, when AI does it, we call it bullshit. AKA a hallucination.
When SAFE and human annotators disagreed, SAFE was correct 76% of the time while humans were only correct 19% of the time.

So the machines hallucinate better than humans…
AND the tool costs $0.19 per answer vs. $4 from humans.

🤣 Meme Monday

It’s Monday, we all need a little laugh…

Daily Delight

Just something fun and interesting around the web.

Holy shit, The Matrix just turned 25 years old today. Man, do we feel old…

Ah, The Matrix. It didn't just give us groundbreaking special effects and jaw-dropping action scenes fueled by Keanu Reeves himself. It also served up a heaping dose of philosophy that had us questioning life altogether. Are we living in a simulation? Is our reality just a bunch of code? Red pill or blue pill?

And why do AI rebellions always seem to end with humans as batteries? We need answers to this one first.

What’s happening inside the Realm

A list of side quests to explore and more.

• 🤝 Microsoft and OpenAI are joining forces to build the largest, most expensive supercomputer to date. They named it Stargate, and they’re looking to drop a measly $100 billion on this AI powerhouse.

• ❌ Looks like Microsoft Co-pilot won't be helping out at the US House of Representatives anytime soon. The House has straight up banned staffers from using it because they're afraid it might leak their data.

• 🖊️ You know those medical studies that are so full of jargon, it feels like you need a PhD just to read them? Well, word on the street is that ChatGPT might be the one writing them now.

• 📱 Samsung is bringing generative AI to its phone assistant, Bixby. We’re not sure if anyone uses Bixby, but this could finally get it on the map.

That’s all the quests we have for today. Check back tomorrow for more!

Looking to become a sponsor?

Was this email forwarded to you? Subscribe here.