Artificial Intelligence

AI is running out of data to train itself

Unsurprisingly, AI has consumed more knowledge than we have available for it. As tech companies like OpenAI continue to feed every bit of information they can into their AI systems, they have almost entirely drained the vast oceans of public data available for AI training.

Companies also warn that high-quality data could completely run out in the next two years, which is making them rethink how they train their systems, the WSJ reported.

What can they change?

Tech firms have to get creative if they want to keep their rate of growth steady, which means finding new sources of data for AI training. But, many of these methods can be legally obscure, according to the NYT.

• OpenAI has reportedly created a tool that transcribes audio from YouTube videos, which provides a new source of data, but may also violate copyright claims.

• Interestingly, Youtube’s parent company, Google, doesn’t have an issue with that. But, the NYT says it’s probably because Google is also transcribing YouTube videos to feed its own AI models (surprise, surprise).

Meta even reportedly considered buying publisher Simon & Schuster for the information contained in its books because the internet is so void of data.

And get this: Since human-created content is in limited supply, some companies have reportedly begun developing “synthetic” information. What is synthetic information, you ask? It essentially means AI created content… and it will be used to train the same AI. Weird times man, weird times.

Meet the college student’s best friend: AI

While college students might be working hard, AI is outworking them all. According to data from Turnitin (just reading the name gave me chills), over 22 million high school and college essays that were turned in were found to have at least some form of generative AI.

It’s a response to a lingering problem

Since ChatGPT took over the internet in late 2022, it’s been dedicated to a plethora of uses: content creation, writing your travel itinerary, generating ideas, and of course, essays.

In response to the flood of students who suddenly started capitalizing off of the ease of writing AI papers, Turnitin and other education-related companies started creating their own defense: AI detection software.

• Turnitin says its detector has reviewed more than 200 million papers in total.

• Of those 200 million, 11% of papers (might) have been 20% written by AI, while 3% of papers have at least 80% AI-generated writing.

Hit or miss

Experts say AI detection software isn’t very reliable and that false positives (which happen frequently) can hurt a student’s academic career. Even AI software was found to contain bias about what it thought was written by its robot peers:

• A 2023 study found that seven AI-detecting tools had over a 60% false positive rate when reviewing exams taken by those who speak English as a second language.

• Turnitin claims its false positives rate is around 1%, but wasn’t included in the study.

  • A different study in October found Turnitin was the most accurate out of 16 AI detection softwares.

The education company has also stressed that its AI detection software isn’t the end-all be-all regarding AI plagiarism. Schools like Vanderbilt, Northwestern, and Montclair have told professors to stop using the AI detector completely, while Purdue University told its educators not to use the detector as proof of cheating on essays.

• The software isn’t transparent in how it detects AI content, and administrators have called it a “black box”.

Regardless, AI regulation in the education system (like everywhere else) is playing a game of catch up that can’t quite stay with the rate in which the technology is advancing. So, however this plays out could be anyone’s guess.

Tech

One man stopped an international cybersecurity attack

The random computer guy you never knew you needed came to the rescue. A German software developer who works for Microsoft, Andres Freund, found an popular open-source software was deliberately sabotaged by its developer.

Programmers rushed to find a solution and said that if it hadn’t been detected, it could have led to hundreds of millions of compromised devices and a catastrophic cybersecurity breach.

The discovery shook the software world and grabbed the attention of tech executives and politicians alike.

How did he find it?

Freund initially saw a strange error message while doing a regular check up on the Linux operating system, which is a crucial software used by governments, banks, and corporations around the world, per the NYT.

He didn’t think anything of it at first, but after a couple weeks, Freund noticed an application used to log into computers remotely was using exponentially more power in the system than it was supposed to.

Who is responsible?

So far, it looks like a user with the pseudonym Jia Tan. Open-source software is mostly maintained by volunteer developers, but it takes time and effort to become the one who finalizes the product.

Tan spent over three years gaining the trust of those in charge of the code and working with them on a daily basis. It’s unclear who Tan really is, but cybersecurity experts say that because of the project’s intricate nature, it is likely a gang of hackers working for a powerful group or possibly even a nation.

Looking forward: A handful of volunteer developers were the only thing that stood in the way of a potentially devastating attack on global cyber infrastructure. Giant corporations like Microsoft build these open-sourced systems (and make billions doing so), so this incident may serve as a wake-up call to those wanting to prevent a similar situation from unfolding in the future.

Grab Bag

Google might start charging users

After a quarter century of operating a free search engine, Google might start charging. More specifically, Google is thinking about charging for AI-powered search features.

This would be the first time in history that Google has made anyone pay to search, which has some analysts thinking even one of the most powerful tech companies in the world doesn’t know how to compete against ChatGPT.

How would it work?

More than likely, Google as you know it wouldn’t be any different, but you could get a paid subscription to a different version of Google that’s powered by the company’s AI software, Gemini.

AI powered search engines are becoming increasingly competitive in the search engine space, offering an alternative (that’s only getting more popular) that can provide a complete, well researched, and fact-checked answer (with sources) to any question you would usually have to have to go through pages to find.

However, this tech isn’t cheap:

• A former Google employee said that using AI to power a search engine requires insane amounts of data and computing power, and thus is “eye-wateringly” expensive.

If Google’s goes through with this idea, it could have massive implications for the next generation of search engines. A mainstream AI-powered engine might disarrange the ad market, making it harder for websites to gain traffic (and upending the whole industry).

Amazon pulled it’s “just walk out” grocery stores

While the tech giant wanted to revolutionize the grocery game, it ended in a rather disappointing way. Amazon is getting rid of its “Just Walk Out” checkout system from Amazon Fresh stores in the US and instead investing into its Dash Carts division.

• “Just Walk Out” uses overhead cameras and shelf sensors to track customers’ baskets and charge them once they leave the store, in theory, making it more convenient for the customer.

If it’s so convenient, why scrap it?

Due to the contactless nature of the store, it requires a vast amount of technology to simply open the store, which is extremely time-consuming and expensive to install and operate. Plus, there are issues:

• The tech would get confused if customers put items on different shelves than they got them from.

• In addition, it would take time to send customers their digital receipts, with some reporting hour long delays.

This is mostly because the “Just Walk Out” system needs around 700 human checks for every 1,000 sales (as of 2022), which was insanely more than Amazon wanted… the company’s goal was 20 to 50 checks per 1,000 sales, according to The Information.

However, if you want to “Just Walk Out”, there’s still some hope. Consumers can still shop at US Amazon Go locations, smaller UK Amazon Fresh stores, along with some airport stores and stadium concession stands that have the contactless technology.

Maybe they should just walk out on the physical side of things: Amazon hasn’t been doing great in the physical store space. In Q4 of 2023, its revenue was only up 14% compared to the first full quarter after Amazon bought Whole Foods in 2017, and only accounted for 3.4% of the company’s business.

Fast Facts

Antipode Analyzer: Find the exact opposite spot on Earth from where you are, which is called your antipode.

Battery Ram: Apparently, a chunk of old batteries from the International Space Station fell from outer space and crashed through someone’s roof in Naples, Florida.

Job Generator: About six out of every 10 jobs people do today didn’t exist in 1940, according to a new analysis of US census data from 1940 to 2018 led by an MIT economist.

Music Muzzle: Chechnya banned music that’s too fast or too slow, stating all music should fit into a tempo ranging from 80 to 116 beats per minute, Russian media reported.

24-Hour Power: The number of restaurants that stay open 24 hours dropped 18% from 2020 to 2024, with Los Angeles experiencing a 35% plunge, according to Yelp data.