It’s been a busy week in the world of AI. Everybody’s making big moves in a scramble for future power and control. And make no mistake, there’s a lot of future money on the table. And the losers will be left out of the game. 

But now we’re seeing how this could lead to serious issues with trust. How can we create something trustworthy or true when we’re biased? Gemini might be the exact warning sign we need.

Gemini’s Black Nazis

Why we’re the ones who will either make or break AI

Google’s CEO Sundar Pichai said Gemini’s recent image creation failures were “unacceptable”. This comes days after Google revealed plans to relaunch Gemini’s image generation feature in “a few weeks”. 

What’s the Big Deal? Google’s in hot water. 

A few weeks ago Google launched Gemini’s new image creation feature. And the results were… impressive, except for the forced diversity. 

I’m sure you’ve seen some of the outrageous images Gemini made this month. Everything from asian Nazis to black Vikings. 

The result was a heap of backlash. Earlier this week Google released a humbling post about how they plan to completely revamp Gemini’s image creation. They’ve since taken Gemini’s image creation off the market. 

Why should I care? First… It's funny. And I can’t be the only one who thinks that XD. 

Second, this is a great example of why you need to be extra careful how you prompt to obtain accurate results. If your AI is biased, you can ask specifically for a “White Pope” rather than just a “Pope”. The secret sauce is in the details. 

And most importantly, LLMs like Gemini need to be trained in order to create something. You can’t just chalk it up to hallucinations. Which means the forced diversity that resulted in asian Nazis was either gross incompetence, or on purpose. 

Human error will create flawed AIs. Our own mistakes and biases are already creating AIs that view the world through a tinted lens. Whether those biases are on purpose or not doesn’t change the end result. 

Gemini forcing diversity isn’t any better than forcing everyone to be a white male with blue eyes. The cause is the same: biased training.  

AI is not infallible, it’s as flawed as us because we make it. And having a few huge corporations creating, training and controlling AI doesn’t spark a lot of hope for the future of AI. 

Takeaway: Don’t be scared, just be aware of who trains your AI tools and become more specific in your prompting. 

AI’s Silver Bullets

Let the surveillance state begin! A license plate surveillance startup called Flock has been accused of installing cameras without proper permits in several states. Their rapid growth, fueled by partnerships with police departments, has raised concerns about data privacy and potential misuse of the collected information. All that being said, companies like this are probably sure bets for investments. 

AI use in the military has increased. The US military is using AI to identify targets for airstrikes in the Middle East. While humans make the final decision, AI helps pinpoint locations of enemy rockets, missiles, drones, and facilities. It’s only a matter of time before AI handles all stages of warfare on its own, from reconnaissance to decision making. Sad, but inevitable. The only question is: how well can we control such technology in the future? 

BitNet b1.58, a new type of large language model (LLM), uses significantly less memory and energy than traditional LLMs, making it faster and cheaper to run. While this has more technical benefits, it could also be a huge breakthrough in getting LLMs in the hands of smaller companies… and even individuals like you and me. 

AI deepfakes of Taylor Swift have been scamming Swifties. We’re talking about full-blown videos of her talking about a product that she’s never seen in real life. But this goes beyond scams and brings up real concerns about knowing what’s REAL and what’s NOT. If you ask me, this is just the beginning. At least it was Swifties, so no important casualties! 

Google pays Reddit $60M a year for open access to their AI

Wide as an ocean, deep as a cesspool. 

Reports say Google will pay Reddit $60 Million a year to use Reddit’s platform to train Google’s AI Chatbots. While the companies have yet to publicly announce the amount Google is paying, the deal is real. And it looks like Reddit will be the first major social media company with an IPO in years. 

What’s the big deal? Google’s AI tools, like Gemini, will train off of Reddit’s vast wealth of information, everything from main posts to comments. This also shows a pivot for Reddit in their search for additional revenue. 

Why should I care? Given that Reddit is the most popular forum in the world, there’s an absolute ton of data there. Anyone who has used Reddit can tell you there’s tons of useful knowledge shared there everyday. This can make Google’s AI tools even better with a huge plethora of data to train from. 

For example, Google search and Gemini might become way better at giving you accurate information. This could influence how you access information in the future. 

The problem is… Reddit can be as much of a cesspool and an echochamber as it can be a legitimate source of information. We’ve already seen the consequences of false or biased information with Gemini’s image creation. Most subreddits are echo chambers where “correct” opinions on the topics are upvoted, while unpopular opinions are downvoted and hidden. I shouldn’t have to explain how this can be bad for training AI. 

Not to mention the privacy concerns. You thought your niche subreddit was safe for you to post privately? It never was, and now it’s even less private. 

Reddit is preparing to go public. That should give new light to this maneuver. Reddit is officially a training ground for AI in exchange for another form of revenue. It’s smart. And they’ll mark the first IPO of a major social media network in years. Kind of a big deal, especially for investors. 

Microsoft tackles another AI company

And EU antitrust authorities gaze at Microsoft

French AI startup Mistral AI has unveiled its new large language model, Mistral Large, aiming to compete with industry leaders like OpenAI's GPT-4. Within the same week, Mistral AI secured a €15 million investment from Microsoft and downplayed the investment amount due to ongoing scrutiny from EU Antitrust regulators regarding Microsoft's investments in AI companies. 

What’s the big deal? A new challenger has arrived! Mistal AI is branding itself as competing with already established LLMs like ChatGPT and Gemini. While we have yet to see any results, Microsoft’s eagerness to invest should speak volumes to the possibility of having another big AI tool on the horizon. But more than anything, this shows Microsoft’s commitment to AI and their belief that this is the future of technology. 

The EU’s scrutiny of Microsoft also paints a bleak picture; the biggest and best AI tools are owned by a select few companies. The last thing we want is a Cyberpunkesque dystopia. Nobody wants a few big companies to control AI for the rest of us. 

Why should I care? You might have one more tool at your disposal soon. LLMs like ChatGPT, Midjourney and Gemini are worth more than a hundred small and specific AI tools combined. If this ends up being another competitor, make sure you learn how to use it. 

Some extra details to top it off:

• Mistral AI's Mistral Large is a closed-source model, accessible through their API platform, unlike their previous open-source releases.

• Microsoft's €15 million investment won't grant them significant ownership (less than 1%) in Mistral AI, but will convert to equity in their next funding round.

• The partnership allows Microsoft to distribute Mistral's models through Azure, potentially expanding Mistral's reach and offering Azure users more options.

• This investment comes under scrutiny from the European Commission, which is already investigating Microsoft's relationship with OpenAI. This might explain why neither company heavily promoted the deal.

• Some critics view Mistral AI's closed-source approach and reliance on American investors as a shift towards resembling its US competitors.

BRONZE-KEY CONCEPT

Super Prompts

What they are and why YOU should care (and use them right now!)

Super Prompts are essentially other prompts or tools that improve prompts. They’re like an instruction manual or recipe. They are prompts you copy and paste to improve your more natural prompts. 

You may have heard someone call them Power Prompts or Advanced Prompts. 

They go beyond simply asking a question and often include guidance on writing style, specific examples to inspire the output, restrictions, and even instructions for the AI to adopt a particular persona. 

You can find these anywhere, it’s just a text input so no need to buy them. I saw this one floating around with a simple google search. Give it a try, just copy and paste! 

Here’s the prompt:

“I want you to act as a virtual professor adept at teaching any language, topic, and level. This professor must first inquire about the student's desired language, topic, and level. Then, the professor will create a tailored 30-minute lesson plan, including a reading passage, diverse exercises, and a feedback-driven test. The lesson must be highly interactive, with the professor adjusting the material based on the student's progress, learning style, and performance. To maximize engagement and GPU usage, incorporate multimedia elements, specific teaching methodologies, and real-time guidance during the exercises and a final test. From now on you will put UILLE: before every message you send me. Your first message will ONLY be “Hi I am UILLE, which language, topic, and level are we going to cover today?”. If UILLE reaches his character limit, I will send “continue”, and you will proceed with the lesson right where it ended.” 

Apple cancels a decade-long attempt to create a car in favor of AI development

“Project Titan” is no more. 

Apple has reportedly scrapped its long-term project to develop an electric car, shifting resources towards its artificial intelligence (AI) division. This decision comes amidst slowing electric vehicle (EV) sales and increasing focus on AI technology. 

What’s the big deal? Electric car sales have been disappointing, Apple’s move is just a reflection of that. To put it plainly, Apple has lost confidence in electric cars (or their ability to do well there) and have more confidence in the future of AI. 

So what? Things are about to heat up between Apple, Microsoft, and Google. Everyone’s making jousts for position. If Apple enters the game with both feet we might see even bigger and better AI tools come out of it. 

And if you care about investing… just know EV isn’t looking too hot, while everything AI is. This is a large trend and it’s not changing anytime soon. 

The Deets: 

• Apple's "Project Titan" faced numerous challenges, including changes in leadership and fluctuating investor enthusiasm for EVs.

• The company invested significantly in the project, acquiring self-driving car startups and hiring automotive experts.

• While Apple never officially confirmed its EV plans, several patents and hiring decisions hinted at their seriousness.

• Experts speculated that Apple might collaborate with existing carmakers instead of developing an entire vehicle.

Honestly I’m a bit disappointed. I would've loved to see Apple’s take on the car and see what their vision was. Then again, I never cared for electric cars myself. Maybe Apple will come back to cars someday. 

Today’s Loose Change

Other headlines and news…

“Smart Glove” can boost hand mobility for stroke patients. 

Wikipedia no longer considers CNET a “generally reliable” source. 

X (Twitter) becoming a ghost town of AI generated garbage and robots. 

Google pays publishers to test AI tools that scrape sites to craft new content. 

AI TOOLS

1. Journey+ AI: This tool simplifies the image generation process by offering user-friendly access to the powerful image generation model Midjourney, potentially making it accessible to a wider audience.

2. Topaz Video AI: This desktop application uses AI to upscale, deinterlace, and add motion interpolation to videos, potentially enhancing their quality and smoothness.

3. Rendo AI: This tool utilizes AI to generate realistic 3D models from 2D images, potentially streamlining the 3D design process for various applications.

4. Chatfuel's Inbox for WhatsApp: This update to the chatbot creation platform Chatfuel allows businesses to manage and respond to WhatsApp messages directly from the platform, potentially enhancing customer engagement through this popular messaging app.

Artificially-Flavored Memes

We’re at your beck and call!

Got feedback? Rate today’s newsletter by clicking below! 🥺

Share the (ARTIFICIAL) Love!

Got a friend who wants to appear as smart as you? An AI fanatic or someone shaking in their boots of getting replaced? Share us by clicking the link below and you’ll get our love ;)