Workers of the world unite, Meta ups the AI video ante, cloning Duolingo using ChatGPT, and what happens when ALL internet content is AI-generated?

Written by Fola Yahaya

Thought of the Week: The internet is dead. Long live AI?

I've written extensively about Dead Internet Theory — the idea that the internet is slowly being overrun by AI-generated content, to the point where human-authored work becomes a rarity. This theory, which gained traction a few decades ago as companies engaged in an SEO arms race to outrank each other on Google, is even more relevant today given the ease with which tools like ChatGPT can churn out content. So much so that predictions that the majority of online content will be AI-generated by 2025 now seem like a massive understatement.

This makes finding nuggets of great content amid the detritus of shallow, SEO-driven, sponsored material increasingly tedious. Nor can we assume that AI search agents will come to the rescue. Two of the biggest players in the race to dethrone Google, Perplexity and OpenAI's upcoming SearchGPT, have both been clear that their AI-generated search results will be ad-supported. So we'll be doubly screwed: crappy AI-generated answers and ads!

Ok, so you may think this doesn't matter as soon we'll all have personal AI agents who find things for us and obviate the need to search for stuff in the first place. I foresee six key trends that will shape our online experiences:

1. Machines as curators: Given all my personal data, it's easy for an AI (and what or who controls it) to fully shape my opinions and perceptions of reality. Algorithms, such as the ones that have made Instagram, TikTok and Netflix so powerful, already create content based on what we watch, but soon they'll dictate what we value, how we think and, possibly, who we become.
2. Fracture and retreat: This is already the case for many internet users on social media. A general withdrawal from the 'vanilla' internet – the stuff where most content is advertorial – to private, authentic and critically human online spaces; the online equivalent of gated communities.
3. A massive premium on well-written, human content: The content creation economy changes dramatically, as certain types of content are devalued while seeing an increased premium on demonstrably human-created content. A good example of this is the success of publishers like The New York Times. On its knees only a few years ago, the NYT now has more than 10.8 million subscribers and over $600 million in revenue.
4. The need to pass the AI sniff test: Whether you're handing in your homework or posting pictures of your family holiday, all content will be assumed to be AI-generated until proven otherwise.
5. Offline is the new online: As AI-generated content becomes more prevalent, we may see a shift in how we value and consume information, potentially leading to a renaissance of offline, in-person experiences and analogue content.
6. Global groupthink: The global soft power that the West has exerted over the rest of the world will be massively amplified as (poor) countries fail to create AI in their own image, and instead rent their 'intelligence' from Big Tech.

Workers of the world unite (for now)

As automation and artificial intelligence reshape industries, workers are increasingly uniting to resist the rising tide of job-killing automation. Following similar action by Hollywood writers in 2023, last week it was the turn of the International Longshoremen’s Association (ILA) to strike and effectively shut down all East Coast U.S. ports. Their grievance? Automation. U.S. dockworkers are worried that robots are coming for their well-paid but highly labour-intensive work of unloading and transporting goods, a scenario they believe could see them suffer the same fate as Detroit autoworkers.

Dig a little deeper, however, and sympathy for the ‘plight’ of the 45,000 or so U.S. dockworkers starts to evaporate. Firstly, according to a 2019-20 annual report from the Waterfront Commission of New York Harbor, about one third of local longshoremen make $200,000 or more a year. The head of their union, Harold Daggett, drives a Bentley, owns a yacht and earns a staggering $728,000! Given that the strike is costing the economy an estimated $4.5 billion per day, the ILA literally have the U.S. economy over a barrel. So really, this strike is little more than a shakedown.

Ironically, these kinds of protests often end up accelerating automation as corporations reassess the inherent inefficiencies of keeping humans doing physically demanding, dangerous and often gruelling work. Moreover, because they are heavily and expensively unionised, U.S. ports are among the least automated in the world. While ports in countries such as Japan, South Korea and the Netherlands use advanced robotics and remote-controlled cranes to increase efficiency, American docks have fallen behind. In fact, none of the U.S. terminals ranked in the World Bank's top 50 performing ports in 2023.

Despite Daggett’s efforts, the fight against automation is a losing battle. While ILA members have won a 61.5% raise over six years in a tentative agreement, the broader effort to halt technological advancements at U.S. ports is futile.

The ruthless job-killing efficiency of AI needs to be top of the government agenda before employment becomes a dystopian, automated wasteland.

How to use AI to build a Duolingo clone in 1 hour

My son received a terrible predicted grade for his upcoming French GCSE. This was both surprising and unsurprising. He speaks Danish fluently, is a bright kid and only has six other kids in his class, so I know he should be able to pass with flying colours. However, I’m not entirely shocked because he dislikes the subject, and let's face it, the English are notoriously bad at teaching French. Even though I'm a lifelong Francophile, I distinctly remember my early love for the language being slowly conjugated out of me.

Given that I can’t ship him off to France, as an AI enthusiast, I naturally thought, "How can I use technology, specifically AI, to help him turn things around?"

Most people immediately reach for Duolingo, but I have my reservations about it. As a major believer in learning efficiency, I instead downloaded the GCSE curriculum, cut and pasted the vocabulary list into a Google Sheet, and uploaded it to ChatGPT to create a nicely formatted table. I then planned to paste this into Anki, a spaced repetition.

If you’ve never heard of spaced repetition (SR), then you’re missing out on a ruthlessly efficient method for brute-force cramming of practically any information. SR is based on research into the forgetting curve, which maps out when people typically forget new information. By reviewing what you’ve learned – usually in flashcard format – just before it fades from memory (at intervals like one minute, one day, one week, one month and six months later), you can commit the material to your long-term memory.

You can download dedicated SR apps like Anki, which provides a home for your flashcards and a preset 'remembering schedule' that will periodically remind you to review. However, where’s the fun in that? Instead, I saw this as a great opportunity to test how well the new Strawberry version of ChatGPT handles coding.

Get in touch if you want to see my full workflow, but in summary, after about 60 minutes of chatting with Strawberry, I managed to create a fully functioning, basic Duolingo clone. This is insane! I’ll repeat myself. This is insane!

What blew me away was that I could:

• set limitations like, “I can’t code for toffee, so make me something that I can cut and paste into a Notepad file and run in any browser”
• create something that runs locally and stores my progress
• add gamification elements such as streaks to motivate me, and, the kicker
• get it to “think of improvements and update the code.”

Use cases for Strawberry’s coding chops include:

• Prototyping apps: What tech entrepreneurs call creating a minimum viable product (MVP) can now be done in minutes to test an idea.
• Building personal productivity tools, such as planners that don’t sell on your data.
• Cloning existing apps: It’s only a matter of time before you can input a URL of an existing app and ask AIs like Strawberry to clone it.

The pace of change in the power of these AI tools is truly breathtaking, and I urge you to experiment with them.

Meta unveils advanced AI video model: Movie Gen

All the major software and tech players are racing to prove they are at the forefront of AI development. From Intel to Salesforce, the rule is simple: if you're not aggressively promoting your AI capabilities, you're risking a dip in your share price. Perhaps the most blatant example of this is Meta, which seems to be caught in a constant cycle of hyping its AI prowess. Last week, it was the much-publicised Orion Ray-Ban AR glasses, which promise to – eventually – do a lot of, well, "stuff." This week, Meta has teased yet another unreleased tool: a new AI text-to-video generator they claim surpasses Sora.

What supposedly sets this awkwardly named Movie Gen apart from its competitors are two key features:

1. Audio generation: It can automatically create audio to match video content, whether it's engine noises, natural sounds or even background music.
2. Natural language video editing: Users will be able to modify existing videos simply by talking to the AI. This means altering someone's clothing, adjusting backgrounds or changing camera angles will all be possible through natural language commands. Movie Gen aims to address a major challenge in generative AI video tools: the difficulty of editing them. Meta is promising that users will be able to verbally control everything from actor types to camera movements by, essentially, talking to the AI.

While this might sound like more tech hype, Meta's plan to integrate Movie Gen into Instagram by 2025 is significant. Coupled with similar AI tools being added to platforms like TikTok and YouTube, this suggests that in the not-so-distant future, much of social media video content will likely be AI-generated.

The real game changer here, though, is not just an upgrade to existing tools. What Movie Gen offers is the ability to create professional-grade content with minimal effort, potentially revolutionising video production across digital platforms.

AI video of the week

A good example of how far text- (and soon voice-) to-AI video creation has come. This may usher in an era of low-cost, home-made ‘direct-to-YouTube’ movies that take market share away from big-budget movie studios.

What we’re reading this week

• Great week for AI as researchers at Google’s DeepMind and Geoffrey Hinton (also a former Googler) won Nobel Prizes respectively for Chemistry and Physics. Interestingly Hinton, often described as the godfather of AI, is now also the face of the ‘AI doomer’ movement and repeated his concerns that AI will at some point be impossible to control.
• The US Department of Justice ruled that Google is a big, fat monopolistic cheat and should be broken up.
• Another week and another major publication sells out to AI – this week it’s the turn of Hearst (publisher of over 20 magazine brands such as Elle and Esquire and 40+ newspapers) which has agreed to let OpenAI loose on its products.
• Good article in the FT (paywalled) on OpenAI’s bet that AI agents will become mainstream by 2025 – summary.

Tools we’re playing with

• ChatGPT Strawberry – The most accessible way to really get a grip on where AI is going. Play with it and build something!

That's all for this week. Subscribe for the latest innovations and developments with AI.

So you don't miss a thing, follow our Instagram and X pages for more creative content and insights into our work and what we do.