Overnight, Everyone's Using a Bot That Writes Code, Draws Pictures, and Chats

A little while ago, I opened my laptop and found my social media feed had been completely taken over by one thing.

It wasn't news. It wasn't some tech mogul's quotable line. It was everyone posting screenshots of their conversations with a bot.

One person had it write a business email — done in three seconds. Another had it write Python code — and it actually ran without errors. Someone had it draw a picture, and it came out looking better than what an art student could do.

I'm talking about ChatGPT.

In February 2023, just two months after launch, it had 100 million users. Think about that — a product reaching 100 million people in two months. That's only happened a handful of times in business history.

The bigger deal this time is that it isn't just the tech crowd getting hyped. It's ordinary people — people who don't know code, don't know models, don't know what a Transformer (the architecture behind modern language models) is — all playing with it, using it, talking about it.

What does that tell us?

It tells us that this time, AI has genuinely walked onto everyone's workbench.

What Exactly Is This Thing?

A lot of people ask me: what even is generative AI? How is it different from the AI we had before?

One-sentence answer: the old AI was about "analysis" — you feed it a pile of data, and it tells you the patterns. Today's AI is about "generation" — you give it a single prompt, and it creates something brand new.

Articles, emails, images, videos, code — it can create all of them.

That's the biggest difference.

ChatGPT is one of its most famous representatives. Under the hood sit large models like GPT-4. What makes them "large"? The GPT-3.5 version alone has 175 billion parameters, trained on over a million datasets and 500 billion tokens. You can think of it as having "read" half the internet.

How well did it read? When GPT-4 came out in 2023, it sat the mock bar exam and scored in the top 10%. Its predecessor, GPT-3.5, taking the same exam, had landed in the bottom 10%.

In a single year, from dead last to top of the class. That's honestly a little scary.

Five Use Cases That Cover Most of What a Company Does

What can generative AI actually do? I've grouped it for you. The common ones on the market fall roughly into five buckets.

First, content generation. Writing blog posts, emails, and ad copy; drawing images; editing video. What used to take a marketing team a week now takes a morning.

Second, information extraction. Hand it a 200-page legal contract, and ten seconds later you get a summary. Which clauses have traps — flagged clearly.

Third, AI customer service. The old customer-service bots would faceplant after three questions. The new ones can handle follow-ups, admit when they're wrong, and figure out what you're actually asking.

Fourth, language translation. Not just Chinese-to-English — it can localize an entire website interface and marketing copy. The same ad, rewritten as a Japanese version, a German version, a Brazilian version.

Fifth, code generation. You describe a feature in plain English, and it writes the code. It can even help you hunt down bugs and convert SQL into Python.

Think about it — one tool that can walk into the IT department, marketing, legal, HR, and customer service all at once. KPMG ran its own survey: 85% of companies surveyed expect their use of AI and predictive analytics models to grow, and half said they've already earned real, money-in-the-bank returns on their AI investments.

This isn't a "future trend." This has already happened.

But Then Things Started Going Sideways

By this point you're probably a little excited.

Hold on. Let me tell you three "faceplant" stories first.

Story one. In 2022, Meta released an AI called Galactica, built to organize scientific papers. The idea was to help researchers find sources faster. It got ripped offline three days after launch — because it fabricated content on a massive scale, citing research that didn't exist with deadpan earnestness, even stamping real scientists' names on the fake material. Experts were livid.

Story two. That same year, another Meta chatbot, BlenderBot 3, started spouting racist and politically biased remarks shortly after release.

Story three. The most expensive one. Google's Bard, in its debut demo, got a question about astronomy wrong. Just like that, parent company Alphabet saw $100 billion wiped off its market cap that same day. A hundred billion dollars — over a single wrong answer.

Notice a pattern? These AIs all share one flaw: they speak with enormous confidence, and they're very often wrong.

The academic world calls this "hallucination." In plain English, it's making things up. It makes things up with such style, citing sources and authorities so plausibly, that even expert readers can be taken in.

This is the first pit — and the deepest one.

Confidentiality, Copyright, Hackers — Three More Pits Are Hiding Here

"Hallucination" is just the tip of the iceberg. Let me lay out three more pits you absolutely need to know about.

First, confidentiality. Many of these AIs "learn as they go" — the data you feed in may get absorbed and become part of the model. Next time someone asks a similar question, your trade secrets or customer data could come spilling back out.

This isn't a hypothetical. Once your company's core data goes into a public AI tool, it's effectively pinned on a public bulletin board. Good luck getting it back — you can't.

Second, copyright. Who owns the copyright on something an AI generates? There's no unified answer anywhere in the world right now. If you take AI-written copy straight into a product brochure and it turns out to have lifted protected content from someone else, the one getting sued is you. A tech outlet called CNET quietly used AI to write over 70 articles — and when errors started showing up, readers caught them, and CNET's reputation went into freefall.

Third, cybersecurity. AI models can be "poisoned" — mess with the training data, and the model goes bad right along with it. They can also be hit with adversarial attacks — carefully crafted inputs that trick the model into making absurd judgments. The moment your AI system is exposed to the public internet, all of these come knocking.

So What Should Companies Do?

At this point you might be thinking: with this many risks, maybe just don't use it?

My advice is the exact opposite. Don't pretend you can't see it — it's already here.

Think about it. Gartner has predicted that by 2025, 30% of the outbound information from large enterprises will be AI-generated. OpenAI took a $1 billion investment from Microsoft, then doubled down with billions more in 2023. Google and Meta are placing heavy bets too. This isn't one or two companies gambling — it's the entire industry chain going all in.

It's going to be stuffed into the browser, the office software, the chat tools you use every day. You can't hide from it even if you tried.

So how do you use it? KPMG's answer boils down to one phrase: use it responsibly. Broken down into a few rules, in plain English:

One, set the rules. The company needs an internal AI-usage policy. Who can use it, what they can do, what they can't do, and that classified information absolutely must not be fed in — all spelled out in black and white. Manage AI the way you'd manage any other technology, not as a toy.

Two, train the people. The same AI, in the hands of someone who knows what they're doing versus someone who doesn't, produces wildly different output. Asking the right question is a craft in itself. You're not just using a tool — you're training it. The better the question, the better the answer.

Three, always have a human watching. This one matters most. AI is an interface, not an oracle. Whatever it produces has to be reviewed, edited, and owned by a person. Machines can churn out work fast, but the final gate has to be a human.

Why? Because AI doesn't have human judgment. It can write a beautiful paragraph — but whether that paragraph fits your brand, whether it might harm a particular group, whether it'll hold up legally — those are calls it can't make.

A human in the loop isn't a burden. They're a guardrail.

This Round Is Only Just Beginning

When it comes to what happens next, I have to admit — I'm not entirely sure either.

But there are a few threads where you can already see the direction.

The software industry could be in for an upheaval. Writing code, fixing bugs, maintaining systems — AI can already do a big chunk of that work. The programmer's job may shift from "writing code" to "watching over code."

The metaverse and virtual reality need enormous volumes of 3D assets, virtual humans, and video content — and AI can produce them in bulk. Work whose costs used to scare off a whole wave of founders is getting dramatically cheaper by the day.

Even cybersecurity stands to benefit. AI can be turned around and used to teach people how attacks work, and to help companies find vulnerabilities before anyone else does.

See — a tool itself has no morality. The same hammer can build a house or smash a window. What decides the outcome is always the hand holding the hammer.

Coming back to the company level, I have a very plain judgment: this round of AI won't make you redundant overnight. But the people who use AI will, little by little, pull ahead of the people who don't. The gap doesn't open up in a day — but it widens, a tiny bit, every single day.

This is something worth taking seriously, right now.