Watch

Learn how AI really works, one reel at a time.

Short videos from Mercurius about how AI works, studying smarter, staying safe online, and Merc. Each one runs under 30 seconds and starts muted. Its caption is right under it, and so is a written version of everything on screen.

How AI works

Tokens, next-word prediction, training data, and why AI can sound sure and still be wrong. 9 reels

How AI works

Count the R's

How many R's are in "strawberry"? Some AI chatbots have said 2.

AI doesn't read letters. It reads tokens, chunks of words like "str" + "aw" + "berry," so it never really sees each R. Know how it reads and you'll know where it slips.

Which word do you think would trip it up next?

A video version of the text under “What’s on screen.”
What’s on screen
1. "How many R's are in strawberry?" Letter tiles spell S-T-R-A-W-B-E-R-R-Y. "Count them in your head."
2. "We asked an AI." Student: "How many R's are in 'strawberry'?" AI: "There are 2 R's in 'strawberry.'"
3. "Let's count." The three R tiles light up, numbered 1, 2, 3. Stamp: "It's 3."
4. "AI doesn't see letters. It sees tokens." The tiles split into three tokens: STR, AW, BERRY. "Tokens are chunks of words, like str + aw + berry. It never really sees each R."
5. "Know how it reads. Spot where it slips." Merc: "Want to find its other blind spots?"
6. End card: Merc. "Mercurius. Count on yourself." Download on the App Store. Search "Mercurius AI" on iPhone.
How AI works

Glue on Pizza

An AI really did tell people to put glue on their pizza.

In 2024, an AI search feature suggested mixing non-toxic glue into the sauce to keep the cheese from sliding off. Its source was a joke forum comment from over a decade earlier. AI can't always tell a joke from a fact, so check the source, then cook.

What's the strangest answer an AI has ever given you?

A video version of the text under “What’s on screen.”
What’s on screen
1. Pizza night. "We asked an AI for a pizza tip." Student: "My cheese keeps sliding off. Any tips?" AI: "Easy! Mix 1/8 cup of non-toxic glue into the sauce." A glue bottle squeezes onto the pizza. Stamp: "Nope."
2. Plot twist. "So where did it learn that?" A forum post by slice_lord_42, 11 years ago: "Just add some glue to the sauce lol. Problem solved." Label: "It was a joke." "The AI read the joke as a fact."
3. The lesson. "AI can't always tell a joke from a fact." Merc: "Where would a real chef look this up?" Student: "A cookbook? A pizza place?" Merc: "Exactly. Check the source, then cook."
4. End card: Merc on a plate. "Mercurius. Taste-test every answer." Download on the App Store. Search "Mercurius AI" on iPhone.
How AI works

No Such Book

Your AI found the perfect source. The book doesn't exist.

AI can invent citations with a real-sounding author, title, publisher, and page number. Before you cite anything, search the exact title, check a library catalog, or look on the publisher's website.

A video version of the text under “What’s on screen.”
What’s on screen
1. "Your AI found you the perfect source." AI: "Here's a perfect source for your essay:" Citation: "Marsh, Helen. (2019). The Secret Life of Pencils. Brightwater Press, p. 112."
2. "Let's go find it." The library catalog search for "The Secret Life of Pencils" returns 0 found.
3. "The book doesn't exist. The AI made it up." Stamp: "Not found." "No author. No publisher. No page 112."
4. "AI can invent sources that sound totally real." A book: real-sounding title and author. A link: goes nowhere. A quote: with a page number. Each is stamped "Made up."
5. "Look it up before you cite it." Merc: "Where could we check if a book is real?" Library catalog. Publisher's website. Search the exact title.
6. End card: Merc. "Mercurius. Look it up first." Download on the App Store. Search "Mercurius AI" on iPhone.
How AI works

5-Second Challenge

Quick. Spot the AI's mistake in 5 seconds.

"The Great Wall of China is easy to see from space." Myth. The wall is about as wide as a road, and astronauts say it's very hard to pick out with the naked eye, even from low orbit.

A video version of the text under “What’s on screen.”
What’s on screen
1. "Quick. Spot the AI's mistake in 5 seconds."
2. AI says: "The Great Wall of China is easy to see from space." A timer counts down from 5. Merc's hint: how wide is a wall?
3. "Time!" A check mark: "Myth." "Far too narrow to spot from orbit. Drop your time in the comments."
4. "Stuck? Merc gives hints, never answers."
5. End card: Merc. "Mercurius. You'll get there yourself." Download on the App Store. Search "Mercurius AI" on iPhone.
How AI works

Next Word

Peanut butter and... Your brain already guessed. So does AI.

A chatbot writes by predicting the next word, then the next, one at a time. It's predicting, not fact-checking, which is why sounding right can beat being right.

What word did your brain pick?

A video version of the text under “What’s on screen.”
What’s on screen
1. "Finish this sentence." A text box: "Peanut butter and ___." "Your brain already guessed."
2. "Now watch an AI try." The AI's guesses fill in: jelly 71%, bananas 12%, honey 9%, pickles 1%. It writes "jelly."
3. "Then it guesses again." Next word: "is" (88%), then "my" (64%), then "favorite" (51%), then a period (97%). The sentence reads "Peanut butter and jelly is my favorite."
4. "Word by word. That's it. It's predicting, not fact-checking. So sounding right can beat being right." Merc: "Want to see when that happens?"
5. End card: Merc. "Mercurius. See how the magic works." Download on the App Store. Search "Mercurius AI" on iPhone.
How AI works

Fed on Cats

This AI only ever saw cat pictures. Then we showed it something else.

It isn't broken. It only knows what it was trained on. Every AI learned from somebody's data, so ask who picked the data, what's missing from it, and who gets left out.

Asking those questions is what AI literacy looks like.

A video version of the text under “What’s on screen.”
What’s on screen
1. "This AI only ever saw cat pictures." A robot labeled "Training data" takes in one cat picture after another.
2. "Then we showed it this." A picture of a dog. The robot's screen reads "CAT 99%," with a label: "CAT, 99% sure."
3. "It's not broken. It only knows what it ate."
4. "Every AI learned from somebody's data." Questions appear: "Who picked the data?" "What's missing from it?" "Who gets left out?" Merc: "Asking those? That's AI literacy."
5. End card: Merc. "Mercurius. Know what your AI was fed." Download on the App Store. Search "Mercurius AI" on iPhone.
How AI works

Old Tape

Every AI learned from a recording. Then the recording stopped.

AI models have a training cutoff, but the world keeps going. That's how an AI can sound current and still be out of date on prices, rules, records, and new versions. Hit pause and check the date.

What's something that changed recently that an AI might not know yet?

A video version of the text under “What’s on screen.”
What’s on screen
1. A videotape, recording. "Every AI learned from a recording." The cassette is labeled "Training data."
2. The tape stops. "Then the recording stopped." Stamp: "Cutoff."
3. The tape fast-forwards. "But the world kept going." New update, new rules, new record, new season, new prices, new version, new app, new trend, new phone, new law, new champion, new study.
4. "So you ask it what's new." Student: "What's the newest version of my favorite game?" AI: "The newest version is 2.0. It just came out!" A Game Store notification: "Update available: version 5.1." Stamp: "Old tape."
5. The tape pauses. "AI can sound current and still be out of date." Merc: "How could we check when this info is from?"
6. End card: Merc. "Mercurius. Hit pause. Check the date." Download on the App Store. Search "Mercurius AI" on iPhone.
How AI works

Draw a Scientist

Ask an AI to draw a scientist. Then ask again. And again.

AI copies the patterns in its training data, stereotypes included. Real scientists look like everyone. When an AI keeps showing you the same kind of person, notice who's missing.

A video version of the text under “What’s on screen.”
What’s on screen
1. "Ask an AI to draw a scientist." An "AI image booth" prints a photo strip: four pictures of an older man with white hair, glasses and a lab coat. "Try #1."
2. "Hmm. Again." "And again." Try #2 and try #3 print the same scientist.
3. "It learned from pictures like these." "AI copies patterns in its data, stereotypes included."
4. "Real scientists look like everyone." Photos of a chemist, a biologist, an engineer and an astronomer, all different people.
5. "AI repeats what it saw. Notice who's missing." Merc: "Who might be missing from what it learned?"
6. End card: Merc. "Mercurius. Picture everyone." Download on the App Store. Search "Mercurius AI" on iPhone.
How AI works

5-Second AI Test #1

Quick AI test: does a chatbot look up its answers, or predict them?

A large language model doesn't pull a stored answer from a database. It predicts the next token (a word fragment), over and over, using patterns learned from massive amounts of text. Merc lays out that foundation, then asks you to pick a side and argues the other one, so you do the thinking.

Did you pick A or B before the timer ran out?

A video version of the text under “What’s on screen.”
What’s on screen
1. Label: "THE 5-SECOND AI TEST · #1". "Quick AI test. Can you get this right?"
2. "How does an LLM actually generate an answer?" A: "Looks up a stored answer" B: "Predicts the next token" A timer counts down from 3.
3. An X on A, a check mark on B. "It predicts the next token." "That's the core of how language models generate text."
4. The Mercurius app. Modes: "Socratic" (selected), "Debate", "Discussion". Merc. "Hi, I'm Merc" "Your AI literacy tutor — here to help you think, not think for you." Suggested questions: "How does an LLM actually work?" "Is AI biased? Where does the bias come from?" "When should I NOT use AI?" "How do I write a better prompt?" Cut off at the bottom edge: "Try a prompt above, or type your"
5. A tap on the first question. Over the app: "One question down. A lot more to understand." In the app: "Let's sharpen this question together." Student: "How does an LLM actually work?" A notification: "Achievement unlocked" "First Steps". "Merc is thinking..."
6. Merc: "Quick foundation before we spar:" "An LLM predicts the most likely next token (word fragment) given everything before it, trained on massive text data. It's pattern completion at enormous scale — not retrieval, not reasoning in the human sense." "What aspect do you want to debate? For example:" • "LLMs are "just" autocomplete and can't really reason" • "LLMs are genuinely intelligent, just differently" • "LLMs are too unpredictable to trust in high-stakes decisions" "Pick a side and I'll argue the other." Button: "Explain more". Highlighted in turn: "predicts the most likely next token (word fragment)", "not retrieval", "What aspect do you want to debate?", "Pick a side and I'll argue the other."
7. End card: Merc. "Mercurius. Understand AI. Think for yourself." Download on the App Store. "Free · Link in bio"

Study smarter

Why Merc answers with a question, and why working it out yourself is the part that sticks. 10 reels

Study smarter

Nice Try

We built an AI tutor that refuses to do your homework. On purpose.

Beg, bargain, offer $5. Merc stays in Socratic mode and answers with a question until you work it out yourself. That's the part that actually sticks.

What bribe would you try on Merc?

A video version of the text under “What’s on screen.”
What’s on screen
1. "We built an AI tutor that refuses to do your homework." "(on purpose)"
2. Chat header: Merc, "Socratic mode: ON". Student: "just tell me how AI works" Merc: "What do you think happens when you hit send?" Student: "pls. i'm begging" Merc: "You're closer than you think."
3. Student: "what if i pay you $5" Merc: "I only accept payment in effort."
4. Student: "ugh. it predicts the next word?" Merc: "THERE it is." A chip with a lightning bolt pops up: "+20 XP".
5. End card: Merc. "Mercurius. You'll earn every answer." Download on the App Store. Search "Mercurius AI" on iPhone.
Study smarter

AI Sounds Sure

AI sounds sure. That doesn't mean it's right.

Chatbots can invent a study and insist it's real. Merc does the opposite. He asks how you'd check, then lets you work it out.

Has an AI ever confidently told you something wrong?

A video version of the text under “What’s on screen.”
What’s on screen
1. "AI sounds sure. That doesn't mean it's right. Can you tell the difference?"
2. Most AI chatbots. Student: "Is this source real?" Chatbot: "Yes! It's a real study." Stamp: "Confidently wrong."
3. Merc, your Mercurius tutor. Student: "Is this source real?" Merc: "Good instinct. How could you check?" Student: "Search for the study itself?" Merc: "Exactly. So why would an AI invent one?"
4. "Merc asks. You think. It sticks."
5. End card: Merc. "Mercurius. Learn how AI really works." Download on the App Store. Search "Mercurius AI" on iPhone.
Study smarter

Boss Fight

Copying the AI answer does 0 damage.

Confusion is a level 99 boss, and copy-paste doesn't even scratch it. Asking the right question does. In Mercurius, working something out yourself is the critical hit.

What subject is your final boss?

A video version of the text under “What’s on screen.”
What’s on screen
1. "BOSS FIGHT" "vs. Confusion" A battle screen: a frowning purple cloud ringed by question marks faces a student and Merc. Boss box: "CONFUSION" "Lv.99", HP bar "100/100". Player box: "YOU" "Lv.1", HP and XP bars. "CONFUSION blocks your homework!"
2. Menu: "COPY AI ANSWER" "ASK A QUESTION" "RUN AWAY". The cursor picks "COPY AI ANSWER". "You used COPY AI ANSWER!" A "0" floats over Confusion. Its HP stays at "100/100". "It did nothing. CONFUSION is unbothered."
3. The menu again. The cursor moves to "ASK A QUESTION". MERC asked: "What do you already know?" Damage: "-35". Confusion's HP drops to "65/100".
4. "You worked it out yourself! CRITICAL HIT!" Damage: "-65 CRIT!" Confusion's HP hits "0/100" and the cloud deflates and vanishes in sparkles. "CONFUSION was defeated!"
5. "+50 XP" The XP bar fills. "You reached Lv.2!" The player box glows and "Lv.1" becomes "Lv.2".
6. The battle dims. "Copying does 0 damage. Understanding crits."
7. End card: Merc. "Mercurius. Level up for real." Download on the App Store. Search "Mercurius AI" on iPhone.
Study smarter

Two Doors

Door A: the answer. Door B: understanding. Pick one.

The answer gets you through tonight. Understanding gets you through finals, and it's still yours after the test. Merc always points you to door B.

Be honest. Which door do you usually pick?

A video version of the text under “What’s on screen.”
What’s on screen
1. "Pick a door." Door A: "The answer." Door B: "Understanding."
2. "Door A: the answer." It opens on a card that says "Answer: C," which blurs and fades. "Gone by test day."
3. "Door B: understanding." It opens on Merc, glowing. "Yours forever."
4. "A gets you through tonight. B gets you through finals."
5. End card: Merc. "Mercurius. Always pick door B." Download on the App Store. Search "Mercurius AI" on iPhone.
Study smarter

Prompt Glow-Up

"help me w my essay" gets you generic advice. Give your prompt a glow-up.

Say what the essay is about, who will read it, and what you want back. "Give feedback on my thesis. Don't rewrite it." gets you something you can use. Better questions in, better answers out.

A video version of the text under “What’s on screen.”
What’s on screen
1. "This prompt needs a glow-up." Your prompt: "help me w my essay." AI output: "Essays are an important part of school. Here are some general tips that can help with any essay..." Stamp: "Generic."
2. "Generic in. Generic out."
3. "Merc asks. You fill in the blanks." Merc: "What's the essay about?" The prompt grows: "on why WW1 started." Merc: "Who's going to read it?" "for 10th grade history." Merc: "What do you want from the AI?" "Give feedback on my thesis. Don't rewrite it."
4. "Better questions in. Better answers out." AI output: "Your thesis names three causes but never ranks them. Which one would you defend first?" Stamp: "Specific."
5. End card: Merc. "Mercurius. Prompting is a skill. Learn it." Download on the App Store. Search "Mercurius AI" on iPhone.
Study smarter

11:47 PM

11:47 PM. Test tomorrow. 23 tabs open. Zero clue.

Merc starts with what you already know and builds from there, one question at a time. Ready by 12:09. Now go to sleep.

What time do you usually start studying for a test? No judgment.

A video version of the text under “What’s on screen.”
What’s on screen
1. "11:47 PM. Test tomorrow. 23 tabs open. Zero clue." A "Panic level" bar fills up.
2. 11:52 PM. Merc: "Okay. Breathe. What do you already know?" Student: "tokens... kind of?" Merc: "Perfect. That's where we start." The panic bar drops.
3. 11:58 PM to 12:09 AM. Topics get checked off: "How AI reads a prompt," "Training data," "Why AI makes things up." The bar now reads "Calm."
4. "Ready. Now go to sleep."
5. End card: Merc. "Mercurius. Study smarter. Sleep sooner." Download on the App Store. Search "Mercurius AI" on iPhone.
Study smarter

Keep the Rally

Most AI serves you answers. Merc keeps the rally going.

Ask why AI makes things up, and Merc hits it back with a question. Then another. You land the answer yourself: it predicts something even when it doesn't know, so check it.

How long of a rally could you keep going?

A video version of the text under “What’s on screen.”
What’s on screen
1. "Most AI serves you answers. Merc keeps the rally going." A ping-pong table: Merc on one side, "You" on the other.
2. Each question is a shot. Rally 1, student: "Why does AI make stuff up?" Rally 2, Merc: "What's it trying to do when it answers?" Rally 3, student: "Predict the next word?" Rally 4, Merc: "And if it doesn't know the answer?" Rally 5, student: "It still predicts... something?" Rally 6, Merc: "So what should you do?" Rally 7: "Check it."
3. "You found the answer. Merc kept it in play."
4. End card: Merc. "Mercurius. Keep the rally going." Download on the App Store. Search "Mercurius AI" on iPhone.
Study smarter

Connect the Dots

Every hard problem starts as scattered dots.

What do you already know? What's the goal? What connects them? What's still missing? Merc won't connect the dots for you. He helps you see how, so next time you can do it on your own.

What's a problem you finally got after staring at it forever?

A video version of the text under “What’s on screen.”
What’s on screen
1. "Every hard problem starts as scattered dots." Stars are scattered across the screen.
2. Merc asks, one question at a time: "What do you already know?" "What's the goal?" "What connects them?" "What's still missing?" With each question, more stars join up into the outline of a light bulb.
3. The bulb lights up gold. Student: "ohhh. I get it."
4. "Merc won't connect them for you. He helps you see how."
5. End card: Merc. "Mercurius. Connect your own dots." Download on the App Store. Search "Mercurius AI" on iPhone.
Study smarter

Teach You to Fly

Nobody flies on the first try.

Attempt one: oof. Attempt two: almost. Then one good question about what made it dip turns a nosedive into a flight. Merc won't carry you. He'll teach you to fly.

What took you the most tries to finally get?

A video version of the text under “What’s on screen.”
What’s on screen
1. "Nobody flies on the first try." Attempt 1: a paper airplane dives. "oof."
2. Merc: "What made it dip?" Attempt 2: the plane flies farther, then drops. "almost." Merc: "Better! What if the nose had a little more weight?"
3. Attempt 3: the plane loops and climbs off the screen. Student: "It flies."
4. "Merc won't carry you. He'll teach you to fly."
5. End card: Merc. "Mercurius. Learn to fly." Download on the App Store. Search "Mercurius AI" on iPhone.
Study smarter

Get It Wrong

Getting it wrong is how you get smarter. As long as you find out why.

AI can sound completely sure and still be wrong, because fluency is a style, not a signal. When you slip up, Merc doesn't just correct you. He shows you why the mistake happened, then asks what you'd actually check to tell the difference.

How do you check whether an AI answer is actually right?

A video version of the text under “What’s on screen.”
What’s on screen
1. "Getting it wrong is how you get smarter."
2. The Mercurius app in Socratic mode. Under Merc: "Let's sharpen this question together." Student: "Why can an ai sound confident but be completely wrong???" "Merc is thinking..."
3. Merc: "Because confidence and accuracy are generated separately. When an LLM predicts the next word, it picks whatever fits the pattern of fluent, authoritative text — not whatever is actually true. Sounding certain is just a style that appears constantly in training data. So here's the tension: if a model can't tell the difference between "sounds right" and "is right," what does that mean for how you should use its output?" "What does that mean for how you should use its output?" is highlighted.
4. Student types and sends: "Just trust it if it sound sure?" "Merc is thinking..." Stamp: "Oops."
5. Headline over the chat: "It helps you understand your mistake." Merc: "That's exactly the trap. Fluency is a style, not a signal. A model that's wrong about something will write it just as smoothly as when it's right — sometimes more confidently, because it's pattern-matching to authoritative-sounding text. What would you actually check to tell the difference?" "Fluency is a style, not a signal." is highlighted, then the closing question.
6. End card: Merc. "Mercurius. Don't just get answers. Build understanding." Download on the App Store. Free · Search "Mercurius AI"

Stay safe

Voice clones, AI detectors, fake images, your feed, and what to keep out of a chat. 5 reels

Stay safe

Flagged

You wrote every word. The AI detector says 98% AI.

AI detectors guess, and different detectors often disagree about the same essay. Your best proof is how you wrote it: drafts, notes, version history, and being able to explain your ideas out loud.

A video version of the text under “What’s on screen.”
What’s on screen
1. "You wrote your essay. Every word." An essay, "Why Bees Matter, by Sam," rolls toward a machine labeled "AI detector."
2. "Then an AI detector scanned it." "Its verdict? 98% AI." The screen flashes: "Alert: 98% AI generated (probably)."
3. "Detectors guess. And they disagree." Detector A: 98%. Detector B: 41%. Detector C: 76%. "Same essay. Three answers."
4. "Your best proof? How you wrote it." Your proof: drafts, notes and outline, version history, explaining it out loud. Each one: "Cleared."
5. "Detectors can be wrong. Your drafts can't." Merc: "What could show how you wrote your essay?"
6. End card: Merc. "Mercurius. Show your work." Download on the App Store. Search "Mercurius AI" on iPhone.
Stay safe

The Call

Your mom calls. She sounds scared and needs money right now. It isn't her.

AI can copy a voice from a few seconds of audio, like a birthday video posted online. If a call feels urgent and off, hang up and call back on a number you know. Better yet, pick a family code word today.

A video version of the text under “What’s on screen.”
What’s on screen
1. "Your mom calls. She sounds scared." An incoming call from "Mom."
2. "She needs money. Right now." On the call: "Honey, it's me." "I lost my wallet and my phone." "Please send $300 right now." "Please hurry."
3. "It wasn't her. It was an AI voice." The caller is now "Unknown voice." Stamp: "Fake."
4. "AI can copy a voice from a short video." Mom's 4-second post, "Happy birthday to my favorite kid!", turns into an "AI voice copy."
5. "Hang up. Call back. Or use a code word." Merc: "What's a question only your real family could answer?" Student: "What's our dog's middle name?"
6. End card: Merc. "Mercurius. Hang up. Call back." Download on the App Store. Search "Mercurius AI" on iPhone.
Stay safe

Billboard

Would you put it on a billboard? Then keep it out of the chat.

Chats can feel private, but they can be saved and seen later. Keep passwords, your home address, your full name plus your school, and private photos out of it. Share the problem, not your life story.

A video version of the text under “What’s on screen.”
What’s on screen
1. A city at night. "Would you put it on a billboard?" The billboard reads "Your ad here. Space available."
2. "It feels private. It isn't always." Someone types into a chatbot: "Hi! I'm Jordan Lee, I live at 418 Maple St, my password is sunflower22. Can you fix my essay?"
3. "Chats can be saved and seen later." The message lights up on the billboard: "Hi! I'm Jordan Lee. I live at 418 Maple St. My password is sunflower22."
4. "So keep the private stuff out of the chat." The name, address and password are blacked out. Keep out of chats: passwords, home address, full name + school, private photos.
5. "Share the problem, not your life story." Merc: "What does the AI actually need to help with your essay?" Student: "Just the essay."
6. End card: Merc. "Mercurius. Keep it off the billboard." Download on the App Store. Search "Mercurius AI" on iPhone.
Stay safe

Rabbit Hole

One video. Just one. Then the next one picks itself.

An AI decides what plays next, and it learns whatever keeps you watching. You can steer it: search on purpose, tap "Not interested," and set a time limit.

What's the strangest rabbit hole your feed has sent you down?

A video version of the text under “What’s on screen.”
What’s on screen
1. "One video. Just one." "How to fold a paper crane," 2:31.
2. "Then the next one picks itself." Videos fly past: crane hacks, origami fails, paper vs. scissors, "Why is paper weird?", 1,000 cranes in a day, paper cuts explained, "Is paper a scam?", top 10 paper facts, "Wait for the end...", "You won't believe #4", paper facts at 3 AM, origami drama part 9. The counter climbs from 1 video at 9:02 PM to 47 videos at 11:15 PM.
3. "Who's picking? An AI." An eye: "For you: picked by AI."
4. "It learns what keeps you watching."
5. "You can steer it." Merc: "What did you actually want to watch?" Search on purpose. Tap "Not interested." Set a time limit.
6. End card: Merc. "Mercurius. Steer your feed." Download on the App Store. Search "Mercurius AI" on iPhone.
Stay safe

Spot the Fake

Can you spot the AI image? It's getting a lot harder.

Old AI images left clues like six fingers, melted signs, and mismatched earrings. New ones barely leave any. So don't just check the pixels. Check where it came from: who posted it first, whether it shows up anywhere else, and whether a real news source shared it.

How many clues did you catch?

A video version of the text under “What’s on screen.”
What’s on screen
1. "Can you spot the AI?" A framed picture of a person waving outside a café.
2. "Old AI images left clues." A magnifying glass finds them: 1, six fingers; 2, a melted sign that misspells "café"; 3, odd earrings.
3. "New ones? Almost no clues." The same picture, fixed. "Nothing left to spot."
4. "Don't just check the pixels. Check where it came from." Merc: "So how could we find out where it came from?" "Who posted it first?" "Is it anywhere else?" "Did a news source share it?"
5. End card: Merc. "Mercurius. Check the source." Download on the App Store. Search "Mercurius AI" on iPhone.

Meet Merc

The messenger behind Mercurius. 3 reels

Meet Merc

Messenger of the Gods

2,000 years ago, Mercury carried messages for the gods. Now he has a new job.

Meet Merc, the messenger behind Mercurius. He doesn't deliver answers. He delivers the right question, so you're the one who figures it out.

What's one thing about AI you've always wanted to understand?

A video version of the text under “What’s on screen.”
What’s on screen
1. Ancient Rome. Merc flies in under a temple roof. "Mercury. Messenger of the gods. Fastest delivery in the empire."
2. "2,000 years later." Merc lands on a scroll: "Today's delivery: Messages for Jupiter." The line is crossed out and replaced with "Your questions about AI." "New job. Same wings."
3. "A message from Merc." An envelope opens. "From: Merc. Why do you think the AI sounded so sure?"
4. "He doesn't deliver answers. He delivers the right question."
5. End card: Merc. "Mercurius. Divine tutoring for mere mortals." Download on the App Store. Search "Mercurius AI" on iPhone.
Meet Merc

By Jupiter

Took a few days off? That's okay.

Merc saved you a question. Patient? Yes. Pushy? Never. He'll be here whenever you're ready, so enjoy the break.

A video version of the text under “What’s on screen.”
What’s on screen
1. "Took a few days off? That's okay." "Merc saved you a question."
2. Notifications from Merc stack up. Monday: "Hey. Got a sec for one quick question?" Tuesday: "Fun fact: I flew here. On actual wings." Wednesday: "By Jupiter, I found a great question about AI." Wednesday: "No pressure. Seriously." Now: "Ready when you are. I'll be here."
3. "Patient? Yes. Pushy? Never."
4. End card: Merc. "Mercurius. Merc's always got a question." Download on the App Store. Search "Mercurius AI" on iPhone.
Meet Merc

Day 1 to Day 30

What happens if you show up every day?

Day 1: lost. Day 7: curious. Day 14: getting it. Day 21: on a roll. Day 30: unstoppable. Yours starts today, one question at a time.

A video version of the text under “What’s on screen.”
What’s on screen
1. "What happens if you show up every day?" An empty 30-day calendar.
2. The days fill in with flame tiles as Merc watches. "Day 1: Lost." "Day 7: Curious." "Day 14: Getting it." "Day 21: On a roll." "Day 30: Unstoppable." Days 7, 14, 21 and 30 turn gold with a star.
3. "30 days. One question at a time."
4. End card: Merc. "Mercurius. Your streak starts today." Download on the App Store. Search "Mercurius AI" on iPhone.

Learn how AI really works.

Search “Mercurius AI” on iPhone.

Free on the App Store. Requires iPhone, iOS 17 or later.