😺 "CheatGPT" is breaking education

PLUS: ChatGPT is a dangerous therapist...

Welcome, humans.

Don’t say “vibe-coding” can’t pay off. Apparently, a six month old vibe-coded AI startup called Base44 sold to Wix for $80M…cash.

In case you need a refresher, vibe-coding = letting the AI write the code for you. In this case, Base44 helps regular people type in a software idea, and codes the whole thing for them. Here’s an example we made with only 3 words: “a myspace lookalike.”

“Glam vibes”

This type of acquisition is probably very useful for a cookie cutter website developer like Wix, as AI (applied well) has made creating websites even easier than using drag and drop tools (don’t drag and drop, just “vibe). Hence the $80M.

The developer, Maor Shlomo, had only 8 employees, and will divvy up $25M of his sale price amongst them (Split evenly, that’d be like ~$3M a piece. Hey, that’s almost enough to finally afford to live in San Francisco without seven roommates!).

As Bloomberg wrote about recently, we’re in the “tiny team” era of Silicon Valley, where “revenue per employee” matters more than multiple. If you can scale and build something with the least amount of people possible, and get it to profitability? Who knows, you might be able to flip it to an upstart trying to not get disrupted.

Here’s what you need to know about AI today:

Some schools embrace AI teachers as trad edu faces AI chaos.
OpenAI started a $10M consulting service for model fine-tuning.
Claude failed at running a vending machine business.
ChatGPT triggered involuntary psychiatric commitments due to psychosis

Advertise in The Neuron here.

In a world where ChatGPT = CheatGPT, wtf is a college degree anymore?

We’re sure you’ve seen the video of the UCLA student flexing his use of ChatGPT at graduation, but if you haven’t yet, here it is:

Now, as it turns out, the UCLA flex wasn’t cheating, according to Andre Mai (the student himself): He was actually using AI for an AI class, where he was summarizing the key equations he used for his final, critically, with his teacher’s blessing.

But the viral moment exposed a raw nerve: schools are in chaos, and nobody agrees on the rules anymore. According to a recent Axios report, 90% of college students used ChatGPT on assignments within two months of its launch. Now, 1 in 4 teens use it for schoolwork.

This has created a nightmare for educators:

Teachers are now forced to be both educators and AI cops.
AI detection software is notoriously unreliable, leading to false accusations against honest students while cheaters get away with it.
Some professors encourage using AI for outlines; others in the same university will fail you for it. There’s no consensus.
Students are now using AI to write apologies for papers they wrote with AI.

It’s without question that AI has infected schools everywhere, so the question now is what to do about it?

For starters, this chaos has opened the door for radical new models. Enter Alpha School, a private school in Austin, Texas for pre-K through 8th grade where AI IS the teacher, and the results are wild:

Students complete all core academics in just 2 hours a day using AI tutors.
The results? Kids are allegedly learning at least 2x faster than in traditional schools.
The rest of the day is spent on real-world “life skills” like wilderness survival camps, financial literacy, and even starting businesses.

Austen Allred shared a story about the school, which put it on our radar (definitely read his anecdote; it’s wild). They’re apparently using a system called 2hourlearning to do this. This could be a pilot of more things to come, as even the US government has made a push to get AI in schools.

But it begs the question: What about that MIT study that showed when you use AI to write essays, you don’t actually learn the material? Well, it’s all about how you use AI to learn. We wrote a whole deep dive on this topic on the website here. Check it out!

FROM OUR PARTNERS

As a founder, you’re used to sprinting—toward product-market fit, your next round, or that first enterprise contract.

But with AI accelerating how quickly startups build and ship, security expectations are higher earlier than ever. Getting security and compliance right can unlock growth—or stall it if you wait too long.

That’s why fast-growing startups like LangChain, Writer, and Cursor trusted Vanta to build a scalable foundation from the start.

With deep integrations and automated workflows built for fast-moving teams, Vanta gets you audit-ready fast—and keeps you secure with continuous monitoring as your models, infra, and customers evolve.

Join Vanta's Live Demo on July 10 to see the product in action and ask questions in real time.

Save your spot.

Prompt Tip of the Day

It’s time to stop trying to craft a “perfect prompt” and start letting AI gather the context it needs by asking you questions first.

The technique is simple: add “ask me any clarifying questions before you begin” to the end of your prompts. Instead of guessing what information your AI needs, let it interview you.

This aligns with what industry leaders are calling “context engineering”—the shift from perfect instructions to providing AI with the right background information.

Why? Because even the most perfectly worded prompts fail without proper context.

Example:

Before: “Help me meal prep” = generic tips.
After: “Help me meal prep, but ask clarifying questions first” = AI asks about cooking skill, dietary restrictions, time constraints, to create a personalized plan.

Now, if you want a little help in the prompt department, someone on Reddit shared their “perfect prompt” for implementing this process in action. It’s a bit over-written for our tastes, but try it out and see if it helps you!

Treats To Try.

*Anything marked with asterisks is sponsored content. Advertise in The Neuron here.

*Guidde turns your screen recordings into professional video tutorials with AI-generated step-by-step narration and voiceover in 100+ languages.
Twenty is an open-source CRM alternative to Salesforce that you fully own and customize—free to try.
Clarify automatically reads your emails/calls to update your CRM and prep for meetings automatically—paid only rn.
Optibot reviews your GitHub pull requests with actual judgment, finds security issues, and maintains your codebase like a senior engineer would—free to try.
MoodGallery tracks your daily emotions and turns them into AI art—building a visual mood diary you can share with friends (iOS only rn).
Dyad builds full-stack apps locally on your computer like Lovable or v0, but free and open-source.
Byterover remembers how you code, so your coding agent learns your patterns and shares knowledge across projects and teams—free tier, then $19.99 a month.
Lettermeme from Sam Lessin of the More or Less pod (which we shared on Friday) summarizes Sam's top newsletter picks based on a tool he built to manage his own newsletter overload and find the signal worth his time… now you can read them too!

See our top 51 AI Tools for Business here!

Around the Horn.

iRonCub3: First Liftoff of a Jet-Powered Humanoid Robot

Friendly reminder to all you would-be robot developers out there: when you make something as cool as a jetpack powered robot, DON’T give it a creepy baby face… sincerely, the uncanny valley.
Meta hired four more OpenAI researchers, bringing the total to at least 8 hires (that have been reported so far)—keep in mind, about ~5.6K people work at OpenAI now. Oh, and P.S: at least one said they did not receive $100M.
OpenAI is getting into the consulting business and will charge ~$10M to help you fine its models on your own “proprietary corporate data.”
Denmark planned to give its citizens the right to copyright their own face and likeness, so that “everybody has the right to their own body, their own voice, and their own facial features.”
Timothy B. Lee at Understanding AI tested seven coding agents as well as computer-use agents, and found computer use agents were a dead end, while Claude Code, Cursor, and OpenAI Codex were the best vibe coders.
Here are last week’s top AI papers according to Dair AI, and the top papers according to The AI Timeline.
Toyota created a “software designed city” called Woven City, just south of Mt. Fuji, where it tests self-driving cars and delivery robots, many of whom work in waste management in a series of underground tunnels.
Anthropic published new research that studied if its AI Claude could run a vending machine business (spoiler alert: it failed, lost money, and had an identity crisis) AND launched a new program offering $50K research grants to study AI's economic impacts (apply here).
Another reminder: do not trust ChatGPT with your mental health; multiple people have been involuntarily committed due to psychosis triggered by chatting with GPT, and a recent Stanford study showed that both ChatGPT and commercial therapy bots are bad at recognizing and pulling you out from paranoid delusions and should NOT be used instead of human therapists.