Bots are talking to each other

Stop me if you’ve heard this one. One AI bot calls up another, and they chat back and forth in English. Once they realize they’re both AI, they switch to a secret-to-them language totally undecipherable to human ears. 

No punch line, because it’s not a joke. Watch this viral video from an AI hackathon that shows the exchange. I’ll wait.

I’m sure you noticed that when the bots are chatting in English, it’s pretty normal. Once they switch to their secret language, it’s all whirs and screeches, like a dial-up modem. 

It sent a shiver down my spine

You know I embrace AI and all that it has done to change our world. AI bots talking to each other and leaving humans totally out of the equation, though, is terrifying to me, and it should be for you, too. 

This particular screeching secret bot language was created by two engineers from Meta, Boris Starkov and Anton Pidkuiko. For the record, they took home first place at the hackathon.

It’s called Gibberlink Mode. It uses a protocol called GGwaves that transmits data over sound instead of full words. The engineers say the bots are 80% more efficient with Gibberlink Mode than when using human language. 

It’s not the first time

Facebook scrapped an experiment back in 2017 after two AI programs seemed to create their own way of communicating that researchers couldn’t understand.

They must have had the same question I do now: How can ethical and security safeguards be enforced if we don’t know what the bots are saying to each other? The answer: We can’t, and that’s bad news. 

The biggest threat of AI has always been the potential of a digital Frankenstein’s monster we can’t control. Giving AI a language all its own is inviting trouble: “Let’s kill them all!” 

👇 Use the share icons below to let those in your circle know what’s going in the AI space. I bet they’ll be shocked! 

Stop, thief! If you make YouTube videos, check out Meta’s Video Seal tool. It puts an invisible watermark on your videos. Even if someone edits your video, it’ll still show up so you can prove it’s yours. It’s in demo mode, and you can add a message containing up to six characters.

Julie vanished. He’s turning to AI to find her.

Open/download audio

Kurt from Savannah, Georgia, calls in. His friend Julie disappeared last year. He thinks traffickers took her and needs tech help to track her down. Plus, Amazon Alexa is losing millions, a budget travel hack you’ll want to try, and how Meta trains its AI.

Run a business account on Facebook or Instagram? Apply for Meta’s AI for a 24/7 support pilot program. It’s free right now. Customers will be able to chat with your ads or send DMs to get more info. I’m going to try this out myself.

31,000 miles

The longest undersea cable Meta wants to build. If you wrapped it around the equator, you’d have 6,000 miles left over. It’ll speed up the internet, but Meta’s main goal is boosting their AI capabilities.

The future is clear: Meta’s new Aria Gen 2 glasses check your heart rate, and more health tools are coming. They’re working with Envision to help folks who are blind walk around safely with the glasses’ AI assistant. If that’s you or someone in your life, sign up here for updates.

2 million

Number of Ray-Ban Meta smart glasses sold since October 2023. Looks like these are going to make it. Another 10 million are projected to be sold over the next year and a half. 

🤖 Not the AI Zuck wanted: Harry Potter, Jesus Christ and Taylor Swift walk into a bar … in the form of Meta chatbots. Anyone can whip up a chatbot on Instagram, Messenger or WhatsApp. Meta’s system is supposed to flag religious figures, real-life people and trademarked fictional characters, but it’s easy to get around that with a typo or two.

81.7TB

Of pirated books used to train Meta’s AI, Llama. The dirty details came out in court docs. Writers are suing Meta for feeding their (stolen) work into the AI. I’m no lawyer, but it sure sounds shady to me.

Bet he’s a peach to be around right now: Mark Zuckerberg’s got four “war rooms” going at Meta HQ to figure out DeepSeek’s secrets. Two teams are focused on training and operating costs; the other two are digging into how the data might level up Meta’s next Llama model. Think about this: All the U.S.-based AI companies just got a kick in the butt to do better.

$300,000

Paid over six months to TikTokkers willing to jump ship and go Instagram-exclusive. Meta’s contracts, which were sent to certain influencers with 1 million followers or more, are broken down into tiers (paywall link). The top one pays $50,000 a month to post 10 Reels between 15 seconds and 3 minutes in length. I’d take that hourly rate.

🤯 Shocking AI news: DeepSeek, a Chinese company loaded with young, ambitious talent, is causing massive panic in Silicon Valley by releasing multiple AI models 50 times more efficient than the best American ones. DeepSeek’s models outperformed Meta’s Llama 3.1, OpenAI’s GPT-4o and Anthropic’s Claude Sonnet 3.5 in everything from complex problem-solving to math and coding accuracy. Here’s the kicker: They built DeepSeek with $5.6 million; OpenAI spent $5.4 billion per year.

Big promises, no results: There are some big opinions on Meta ditching its fact-checking program, but here’s the joke: It didn’t work in the first place. An analysis found only 14% of posts pushing foreign enemy state narratives were flagged as false on Meta. The new plan of action? Community notes … because that’s totally working out for X.

Who will win the VR war? Meta is retiring its $1,500 Quest Pro headset to focus on the $300 and $499 Quest 3 models. Apple’s Vision Pro is going after businesses now instead of consumers because no one wanted to pay $3,500 for a headset.

January 11th, 2025

Open/download audio

Do you own a TP-Link router? Lawmakers are questioning its ties to China. Plus, my chat with Noland Arbaugh, the first Neuralink recipient, on controlling computers with his mind. Also, Meta shuts down millions of scam accounts, and Gen Z brings back retro cameras.

🤳 AI do not consent: You might see a new you on Instagram. They’re showing people AI images of themselves in their own feeds. Meta’s taking your selfies and sticking them into random scenarios like mirror mazes and astronaut suits, all to advertise their Meta AI.

So, you unwrapped something techy and you’re already having trouble. Before searching online for help, know it’s dangerous to Google tech company phone numbers.

Scammers work the system to get spoofed websites and numbers as the top search results via ads. Instead of waiting for a response to an online form, try calling the manufacturer or tech company directly. I’ve included links below in case you’d rather use a contact form or chat service:

An older CEO, long past retirement, called IT for support. He asked, “How can I print this document?” IT said, “All you need to do is Ctrl + P.” The CEO replied, “I haven’t been able to do that for years!” (I heard you laugh!)

No. 1 free app

Drumroll, please … It’s Temu. That’s two years in a row for the most downloaded free app. In second place is Meta’s Threads, followed by TikTok at No. 3. I know Temu is super cheap, but that comes with its own cost. Here’s why I’d never buy the toys there for the kiddos in my life.

November 23rd, 2024

Open/download audio

The DOJ is going after Google, calling for changes that could lead to breaking up the tech giant or even selling off Chrome.  A deepfake scandal shuts down a school, Missouri cops are in trouble for searching womens’ phones for nudes, and Meta deleted millions of scam accounts.

Meta's brain decoder reads minds

Open/download audio

What if you walked into a room, and everyone instantly knew what you were thinking? We’re not too far off. Thanks to Meta’s tech, similar to an MRI, that’s now a reality! Plus, dating tips for the 50+ crowd, the scoop on ‘text pesting,’ and which smartphone holds its value the longest.