📟 Before LLMs — Rules and Narrow ML
For decades, “AI” meant rule systems and machine learning models narrow: each one trained for a single task—detecting spam, recommending products, recognizing a face. It looked promising, but required specialists and labeled data, and didn’t generalize: a spam model couldn’t translate, and a translation model couldn’t summarize.
🧠 The central idea
“Old” AI worked—but it was like a toolbox where each tool fit only one specific screw. Switching tasks meant practically starting from scratch.
- •One-time task: one model, one problem. Don’t ask for “anything.”
- •Expensive to train: needed labeled data and specialized teams.
- •No fluff: you didn’t speak to it in Portuguese—you programmed it.
💥 The ChatGPT turning point
In November 2022, ChatGPT opened AI up to everyone: all you had to do was talk in natural language, and it responded. In about 2 months reached ~100 million users — a record for adoption at the time. It was no longer a single task: the same system wrote, summarized, explained, and coded.
📊 Why it was a milestone
It was estimated that ChatGPT reached ~100 million users in about 2 months after its launch (Nov/2022), the fastest-growing consumer app ever recorded up to that point. The leap wasn't just technical — it was one of access: AI has gone from a lab tool to something your aunt uses too.
💡 Why this matters to you
The barrier to entry has fallen to zero. You don’t need to “know how to program AI”—you need to know talk with it and direct it. This is the skill the rest of the course develops.
🖼️ From Text to Multimodal
In March 2023 GPT-4 arrived, and AI began to see and hear, not just read. Multimodal means understanding and producing multiple formats: image, voice, and video, as well as text. Along with it came long context — the ability to "read" entire documents at once, not just short sentences.
👁️ Vision — understands images
Send a photo, chart, or screenshot: AI describes it, extracts data, and answers questions about what it sees.
🎙️ Voice — speech and listening
Real-time audio conversation. You speak; it transcribes, understands, and responds out loud.
📄 Long context — reads everything at once
Paste an entire contract, report, or book: it keeps everything in mind so it can summarize, compare, and cross-reference.
💡 Practical tip
Stop describing a screen when you can send the screenshot. Stop transcribing when you can paste the document. Multimodal only helps if you make use of it: give AI the raw material, not your summary of it.
🧰 From Chatbot to Tool
Go in 2023 e 2024, AI moved out of the isolated browser tab and into the workflow. Assistants began living inside email, document editors, spreadsheets, and code editors — where work already happens. They stopped being a place you went to and became a layer that’s everywhere.
✓ Chatbot in a tab
- ✓You leave work, open the chat, copy and paste it back.
- ✓Great for asking questions and drafting one-off things.
- ✓But the context of your work is left out.
→ Tool in the workflow
- →AI is built into email, docs, and code editors.
- →It sees what you’re doing and acts right there.
- →Less copying and pasting; more real work.
📊 The leap in code
The integrated assistant first became standard in programming—suggesting and completing code inside the editor. It was a preview of what would spread to every profession: AI not as a destination, but as part of the environment where you already work.
🦾 From Tool to Agent
Go in 2024 e 2025 the key breakthrough came: AI stopped just answering and started to act. An agent uses tools, browses, takes steps, and completes multi-step tasks. The family Claude (Sonnet and Opus) made progress in this area with what's called "computer use" — AI operating a computer, clicking and typing the way you would.
Each step keeps the previous one and adds autonomy — from who you use to who works with you. Illustrative diagram.
💡 What changes in practice
With an agent, you stop doing every step and start to describe the result: “research these 5 suppliers, put the table together, and send it to me.” AI chains the steps. Knowing when to trust it is the subject of Track 3.
🔮 Where It’s Headed — The Always-On Partner
The next step isn’t for you to launch AI every time — it’s for it to run on background, notice what needs attention, and anticipate. An always-on partner that works while you sleep. That’s the idea behind a personal "Jarvis"—exactly what you’ll build in Track 3.
🧭 The full timeline
💡 The reading that matters
The direction is clear: from responding to acting, from acting to anticipating. Whoever understands this stops treating AI like a little question box and starts shaping it into a partner. That’s where the course is headed—and the bridge to your Jarvis.
Self-check (optional): What major transition is underway in AI?
✅ Module Summary
Next step
Track 1 · Module 1.3 — The Evolution of Professions: how each wave of technology reshaped work, and what AI changes.