The research, translated
What we're learning about AI and you
Four things researchers want you to know. Tap a card to flip it.
under the hood
Does AI actually "know" things?
tap to flip
No. It's essentially advanced predictive text—it guesses the next most likely word based on its training, without real comprehension.
Computer Science Basicson privacy
Where do your chats go?
tap to flip
Many free tools use your conversations as training data for future models. Assume anything you type could be read by others.
Data Privacy Guidelineson environment
Is AI bad for the planet?
tap to flip
Generating an AI image or text takes vastly more computing power—and water for cooling servers—than a standard search.
Nature · MIT Tech Reviewon connection
Can a chatbot replace a friend?
tap to flip
Not really. A friend who never disagrees isn't practice for the real thing.
Stanford News · Science News · OPBon your mind
Is it doing your thinking for you?
tap to flip
It can. Treat every answer as a first draft — not a final one.
APA adolescent wellbeing advisoryon real life
How do we actually use it?
tap to flip
Mostly like a calculator — homework help, not a best friend.
Pew Researchon limitations
Does AI have blind spots?
tap to flip
Always. It repeats the gaps and biases in what it learned from.
AILit frameworkThe Ultimate AI Glossary
Speak the language of the technology.
LLM (Large Language Model)
A type of AI designed to understand and generate human language. It is trained on massive amounts of text data to predict the most likely next word in a sequence. (e.g., ChatGPT, Claude)
Hallucination
When an AI confidently presents false or invented information as absolute fact. This happens because it predicts words, not truths.
Token
The basic unit of data an AI processes. A token can be a word, part of a word, or just a single character. AI models have "token limits" dictating how much text they can remember in one chat.
Training Data
The massive dataset (books, articles, reddit posts) fed into the neural network during its creation. The AI learns patterns, grammar, and human biases from this data.
Neural Network
A computer architecture loosely inspired by the human brain, using interconnected nodes to process complex patterns in data.
Deepfake
Synthetic media (video, audio, or images) generated by AI to look or sound exactly like a real person, often used to spread misinformation.
Myths vs. Facts
MYTH: AI has access to a massive database of verified facts and looks up answers when you ask a question.
FACT: Standard LLMs do not "look things up" (unless connected to a web-search plugin). They generate text probabilistically based on their training. They are creative writers, not encyclopedias.
MYTH: If I ask an AI if it is lying, it has to tell me the truth.
FACT: An AI does not know if it is lying. If it hallucinates a fact, and you ask "Are you sure?", it will often confidently double-down on the lie because it statistically predicts that confidence follows assertions.
MYTH: AI is completely objective and free of human bias.
FACT: AI learns from humans. Because human history and the internet are deeply biased, AI models often replicate racial, gender, and political biases found in their training data.
Quick check
Two different AI tools always give the same answer.
They often don't — that's proof neither is "the truth," just a best guess.
Feeling closer to a friend after a long AI chat is a good sign.
If you feel more like reaching out to people, the AI's helping, not replacing.
You should be able to explain how you used AI on schoolwork.
If you can explain it clearly, you probably used it well.
AI chatbots 'hallucinate' because they want to lie to you.
They don't have intent. They hallucinate when they confidently guess the next word wrong, prioritizing sounding plausible over being factual.
It's safe to paste personal journal entries into AI for feedback.
Never paste private data. It might be used as training data and potentially leaked to others in the future.
The AI Boom: How We Got Here
It didn't happen overnight. It took decades of math.
1956: The Term is Coined
Scientists at Dartmouth College officially coin the term "Artificial Intelligence," dreaming of machines that can simulate every aspect of learning.
1997: Deep Blue Wins
IBM's Deep Blue defeats world chess champion Garry Kasparov, proving machines could out-calculate humans in narrow, complex games.
2017: "Attention Is All You Need"
Google researchers publish a paper introducing the "Transformer" architecture. This breakthrough mathematical structure allows models to process entire sentences at once, unlocking modern AI.
2022: ChatGPT Launches
OpenAI releases ChatGPT to the public. It isn't the first LLM, but it's the first with an easy-to-use chat interface, triggering the massive AI boom we are living through today.
AI in Everyday Life
The truth about "AI Detectors" and False Accusations.
Why Detectors Fail
Many teachers and organizations use software to detect if content was written by AI. These detectors are fundamentally unreliable. They guess based on sentence complexity ("perplexity"). Because of this, they frequently flag students or professionals who write with straightforward, predictable vocabulary—especially non-native English speakers or neurodivergent individuals—as "AI".
How to Protect Yourself
- Use Google Docs Version History: Always write in a cloud document. If you are falsely accused at school or work, you can prove you wrote it by showing your typing history.
- Cite Your Prompts: If you are allowed to use AI for brainstorming, include an appendix showing exactly what prompts you used.
- Advocate for Policy: If your school or employer has a zero-tolerance policy, point them to research from OpenAI and MIT showing that AI detectors do not work reliably.
Ready for the deep dive?
Master AI in our comprehensive 5-hour Academy course.