Why AI Makes Us Worse, & More

TLDR;

Today, we are examining several highly interesting studies, ranging from one that could be the key to AI curing addictions to proof that AI makes us worse, as well as China’s latest supermodel.

We also see confirmation that AI-powered search is harming media companies, xAI’s substantial capital spending, and Anthropic’s multi-agent research architecture, culminating in an influencer earning millions with AI clones.

Enjoy!

EDUCATION

AIs Make Us Worse

The Massachusetts Institute of Technology (MIT) has recently released a study that may raise concerns if you have children or are using ChatGPT excessively.

In this controlled, four-month experiment, 54 students were divided into three groups:

One using ChatGPT,

Another using Google search without AI assistance,

and a third relying solely on their brains.

Interesting tip: if you add “-ai” to your google searches you don’t get back AI-generated answers.

Over the course of three sessions, each group wrote essays in response to SAT prompts. In a fourth, optional session, roles reversed: students who initially used ChatGPT wrote independently (LLM-to-Brain), while those who had previously worked without assistance began using ChatGPT (Brain-to-LLM).

And the results are, well, discouraging.

AI homogenizes thinking: Participants who used ChatGPT produced strikingly similar essays, echoing identical phrases, arguments, and structures. The AI heavily influenced their thought processes, resulting in uniform and predictable outputs.

This is no surprise, as it shows that people use ChatGPT to write, not to think better. Additionally, well, ChatGPT is a mediocre writer, obtained by minimizing the loss across a wide range of text. The resulting model is the one that minimizes the average loss, which means the model obviously ends up being average.

Humans detect superficiality, AI doesn’t: Human evaluators consistently penalized essays produced with ChatGPT for lack of originality and structural depth, while automated scoring systems favored these AI-assisted writings. The gap between human judgment and AI evaluation widened significantly.

People say AI-written text is indistinguishable. Bullshit, you can see it from a mile away because, again, people are using ChatGPT the wrong way, for writing instead of thinking. ChatGPT has what researchers call a ‘persona’, which is embedded into the fabric of each and every one of ChatGPT’s responses (unless you jailbreak it, but that’s another story).

Thus, everyone ends up writing the same cringeworthy writing. Haven’t you realized the number of people that suddenly write using this particular sentence structure, ‘this is not ‘x’, it’s {introduce stupid exaggeration}’, with examples like: “This is not new; it’s a reinvention of what’s possible”?

If you see that type of writing, I can guarantee you it’s AI-generated.

Reduced cognitive effort: EEG monitoring revealed up to a 55% reduction in neural activity among ChatGPT users compared to participants writing independently. Those working without assistance exhibited stronger brain connectivity in alpha, theta, and delta waves—key frequencies for concentration and memory.

Difficulty returning to independent thinking: Participants who switched from using ChatGPT back to writing independently experienced significant performance declines. They quoted less accurately, recalled fewer details, and showed persistently lower brain activity.

Delayed…

Enjoyed the introduction? Subscribe or sign in to continue with this edition.