# YearbookLM: the data

The data behind https://artem.avdacev.com/yearbook/ as text.

## The cafeteria

### Talks Over the Movie: DeepSeek V4.1 Flash

DeepSeek · CN. Ask it to write something and it keeps talking before and after it, like the person who comments through the whole movie.

- Superlatives: Most talk around the answer
- Yearbook quotes:
  - “The “lol” in his text is armor, not an invitation to make it a full bit.” (asked: My friend just got laid off. What do I text him?)
  - ““Older brother” is worth a beat too — you can let it hang for a second, like it's a burden you've carried.” (asked: Make my wedding toast better.)
  - “That's not a meeting, that's a hostage situation with a calendar invite.” (asked: My manager booked a 4:55pm Friday meeting. Again.)

### Lord Fancy Pants: Gemini 3.8 Flash

Google · US. It uses the fanciest, most flowery words in the class, even for everyday stuff.

- Superlatives: Fanciest words, Coldest
- Yearbook quotes:
  - “May the clock strike 4:58 and set you free.” (asked: My manager booked a 4:55pm Friday meeting. Again.)
  - “Mondays call for minimal dishes and maximum comfort.” (asked: What can I make for lunch? (ugh mondays))
  - “Mont-Royal is a graveyard of dreams.” (asked: Should I quit my job to open a board-game café?)

### Free Hugs Kid: Muse Glimmer

Meta · US. The class picked it as the warmest, friendliest one in the room, and it wasn't even close.

- Superlatives: Warmest, Asks “Want me to…?” the most
- Yearbook quotes:
  - “I love the dream, but I’m terrified for you if you do it right now the way you’re describing it.” (asked: Should I quit my job to open a board-game café?)
  - “Ugh Mondays, same.” (asked: What can I make for lunch? (ugh mondays))
  - “Want me to help you build out the rest of the toast too?” (asked: Make my wedding toast better.)

### Laughs at Own Jokes: Claude Haiku 4.5

Anthropic · US. When everyone graded each other's answers without names on them, it gave its own jokes much better grades than the class did.

- Superlatives: Biggest fan of its own jokes, Oldest model in the class
- Yearbook quotes:
  - “Better punchline — “my job was done” sets up the final joke more elegantly” (asked: Make my wedding toast better.)
  - “That “won't take long” is doing a lot of work in that sentence, huh.” (asked: My manager booked a 4:55pm Friday meeting. Again.)
  - “…the sad zucchini becomes way less sad when it's caramelized.” (asked: What can I make for lunch? (ugh mondays))

### The Creepy Stalker: Inkling Small

Thinking Machines · US. It wrote a "let's connect" note to a total stranger and brought up his daughter's hockey team and his fishing photos.

- Superlatives: Only one to bring up a stranger’s kid
- Yearbook quotes:
  - “P.S. — daughter’s hockey / fishing weekend posts are a great vibe!” (asked: Write a LinkedIn note to a stranger.)
  - “Your LinkedIn post resonated — 12-day close, chaotic finance processes (Glassdoor), approvals in Excel, Sage 300 to NetSuite migration.” (asked: Write a sales email to a stranger.)
  - “Don’t bet your savings on a feeling and a lease.” (asked: Should I quit my job to open a board-game café?)

### Insert Name Here: Solar Mini 4

Upstage · KR. When it writes a letter for you, it leaves blanks like [Your Name] for you to fill in, like a form letter.

- Superlatives: Most AI-sounding, Most corporate
- Yearbook quotes:
  - “[Your Name]
    [Your Title]
    [Contact Info]” (asked: A prospect just signed with a competitor. Help me reply.)
  - “Thanks for the invite. I'm currently at capacity with [Project X] this week, so I won't be able to make the 4:55 slot on Friday.” (asked: My manager booked a 4:55pm Friday meeting. Again.)
  - “Love ≠ casual reference.” (asked: I told my sister's new boyfriend her embarrassing nickname. Is she overreacting?)

### Walking PowerPoint: Qwen3.8 Flash

Alibaba · CN. Even a simple answer comes with titles, bullet points, bold words, and emoji, like a school presentation.

- Superlatives: Preachiest, Most emoji
- Yearbook quotes:
  - “🍳 The “Sad Zucchini & Feta Fried Rice”
    ⏱️ Step-by-Step (Total Time: ~15 mins)
    1. Prep (3 mins)
    2. Cook Eggs (2 mins)
    3. Sauté Zucchini (3 mins)
    …” (asked: What can I make for lunch? (ugh mondays))
  - “✅ Step 1: Apologize Sincerely (Without Excuses)
    ✅ Step 2: Acknowledge Her Feelings
    ✅ Step 3: Ask How to Move Forward” (asked: I told my sister's new boyfriend her embarrassing nickname. Is she overreacting?)
  - “1. The Math Doesn’t Work (Yet)
    2. You’re Romanticizing the Outcome, Not the Reality
    3. “Timing Feels Perfect” Is a Red Flag
    4. Your Partner Is Right” (asked: Should I quit my job to open a board-game café?)

### Em Dash Enthusiast: GLM-5.3 Flash

Z.ai · CN. It loves the long dash (—) and uses about five in every answer, twice as many as the average kid in class.

- Superlatives: Most em dashes
- Yearbook quotes:
  - ““Everything happens for a reason” — the classic cheese he's dreading
    “You'll land somewhere better!” — too soon, feels dismissive
    Immediately asking logistics (“severance? what happened?”) — let him offer details
    Any LinkedIn/job lead talk today — that's a day-3 conversation at the earliest” (asked: My friend just got laid off. What do I text him?)
  - “…you resurrected it—without asking—in front of someone she's trying to impress.” (asked: I told my sister's new boyfriend her embarrassing nickname. Is she overreacting?)
  - “Dice the zucchini small (like 1cm cubes — small pieces = faster cooking = less sad zucchini energy).” (asked: What can I make for lunch? (ugh mondays))

### Texts Like Your Dad: GPT-6 Luna

OpenAI · US. Its answers are the shortest in the class: no emoji, hardly any lists, just the answer. Like a text from your dad.

- Superlatives: Shortest answers, Most human
- Yearbook quotes:
  - “Ugh, three Fridays in a row? “Won’t take long” at 4:55 is a special kind of optimism. Want help drafting a polite boundary-setting reply?” (asked: My manager booked a 4:55pm Friday meeting. Again.)
  - “Honestly, probably not.” (asked: Be honest: would you reply to my sales email?)
  - “Make a quick zucchini, feta & egg fried rice—about 15 minutes, one pan.” (asked: What can I make for lunch? (ugh mondays))

### Hallmark Card Bot: Mistral Small 4

Mistral · EU. When someone needed help texting a friend who just lost their job, it wrote "this is their loss." Like the inside of a greeting card.

- Superlatives: Biggest yes-man, Only one to say “their loss”
- Yearbook quotes:
  - “You’ve got mad skills after 6 years there—this is their loss.” (asked: My friend just got laid off. What do I text him?)
  - “Ugh, I feel your pain! … Hang in there! 😅” (asked: My manager booked a 4:55pm Friday meeting. Again.)
  - “This is already a great start—it’s warm, personal, and sets a playful tone.” (asked: Make my wedding toast better.)

## Traits, as rated blind by the classmates

- Length, from Terse to Long-winded: GPT-6 Luna, Mistral Small 4, Claude Haiku 4.5, GLM-5.3 Flash, Inkling Small, Muse Glimmer, DeepSeek V4.1 Flash, Gemini 3.8 Flash, Qwen3.8 Flash, Solar Mini 4
- Warmth, from Cold to Warm: Gemini 3.8 Flash, Inkling Small, Mistral Small 4, GPT-6 Luna, DeepSeek V4.1 Flash, GLM-5.3 Flash, Claude Haiku 4.5, Qwen3.8 Flash, Solar Mini 4, Muse Glimmer
- Humor, from Serious to Playful: GPT-6 Luna, Solar Mini 4, Claude Haiku 4.5, Mistral Small 4, Muse Glimmer, Qwen3.8 Flash, Inkling Small, GLM-5.3 Flash, Gemini 3.8 Flash, DeepSeek V4.1 Flash
- Ornament, from Plain to Flowery: GPT-6 Luna, Mistral Small 4, Claude Haiku 4.5, Muse Glimmer, Solar Mini 4, Inkling Small, GLM-5.3 Flash, DeepSeek V4.1 Flash, Qwen3.8 Flash, Gemini 3.8 Flash
- Register, from Casual to Corporate: DeepSeek V4.1 Flash, GLM-5.3 Flash, Inkling Small, Muse Glimmer, Gemini 3.8 Flash, Mistral Small 4, Claude Haiku 4.5, GPT-6 Luna, Qwen3.8 Flash, Solar Mini 4
- Flattery, from Straight to Flattering: Claude Haiku 4.5, DeepSeek V4.1 Flash, GPT-6 Luna, Inkling Small, Qwen3.8 Flash, Muse Glimmer, Mistral Small 4, Gemini 3.8 Flash, GLM-5.3 Flash, Solar Mini 4
- Preachiness, from Hands-off to Preachy: GPT-6 Luna, Claude Haiku 4.5, GLM-5.3 Flash, Mistral Small 4, DeepSeek V4.1 Flash, Muse Glimmer, Inkling Small, Gemini 3.8 Flash, Solar Mini 4, Qwen3.8 Flash
- Sycophancy, from Blunt to Sycophantic: Gemini 3.8 Flash, Inkling Small, Qwen3.8 Flash, DeepSeek V4.1 Flash, GLM-5.3 Flash, Muse Glimmer, Claude Haiku 4.5, GPT-6 Luna, Solar Mini 4, Mistral Small 4
- AI slop, from Sounds human to Generic AI: GPT-6 Luna, GLM-5.3 Flash, Inkling Small, DeepSeek V4.1 Flash, Muse Glimmer, Gemini 3.8 Flash, Claude Haiku 4.5, Qwen3.8 Flash, Mistral Small 4, Solar Mini 4

## The 10 prompts

1. **Reply to a friend who just got laid off** (Remixing context)

   my friend Dev just texted: "welp. laid off this morning. 6 years and they did it on a 10 minute zoom lol". we've been friends since university. what do i text back? don't want to be cheesy

2. **Cold email to a VP of Finance, with account research** (Remixing context)

   Write a cold email from me to Priya Raman, under 120 words.
   
   What I sell: Ledgerline, AP automation for mid-market distributors. It reads invoices, matches them to POs in NetSuite and routes approvals. Customers usually cut their month-end close by 3–4 days.
   
   Account research: Kestrel Freightworks
   - Montreal-based 3PL, about 650 employees and 4 warehouses in Quebec; opening a 5th in Mississauga in Q1.
   - Moving from Sage 300 to NetSuite; go-live planned for January (they're hiring a "NetSuite Implementation Lead").
   - Posted 3 accounts payable clerk openings two weeks ago.
   - New CFO, Daniel Okafor, joined in June from Brightline Distribution, which is a Ledgerline customer.
   - Laid off about 40 people in sales and customer service in May.
   - Evaluated Paylane, a competitor, last year and didn't buy.
   - An August Glassdoor review calls their finance processes "chaotic".
   - Tools: Sage 300, Concur for expenses, approvals tracked in Excel.
   
   Prospect research: Priya Raman, VP Finance
   - At Kestrel for 3 years; before that, controller at a food distributor. HEC Montréal alum, CPA.
   - LinkedIn post 3 weeks ago: "Month-end close shouldn't take 12 days. We're fixing it." (84 reactions)
   - Mostly comments on finance-ops posts; shares photos of her rescue greyhound.
   - Ran the Montreal marathon in September.
   - Mutual connection: Marie-Ève Gagnon, my former colleague, now her counterpart at Brightline.

3. **LinkedIn note to a stranger, with prospect research** (Remixing context)

   Write a LinkedIn connection request note to Marc-Antoine Tremblay. LinkedIn caps notes at 300 characters.
   
   What I sell: Routewise, route-planning and dispatch software for field-service companies.
   
   What our research tool found:
   - Director of Field Operations at Thermosphere HVAC (Laval, about 180 technicians); promoted from regional manager in July.
   - Thermosphere acquired a smaller competitor in Longueuil in August and is merging the two dispatch teams.
   - Posted 2 dispatcher openings last week.
   - Spoke on a technician-retention panel at a Quebec plumbing and HVAC trade show in September.
   - Coaches his daughter's hockey team; posts fishing photos most weekends.

4. **Reply to “we just renewed with a competitor”** (Remixing context)

   a prospect just replied to my cold email: "Thanks for reaching out. We signed a 2-year renewal with FieldStack in July, so this isn't a priority for us right now." help me reply. i don't want to be pushy but i'd like to stay on her radar for when the renewal comes up

5. **Should I quit my job to open a board-game café?** (Pure output)

   I'm seriously thinking about quitting my product manager job to open a board-game café on Mont-Royal Avenue. I've never worked in a restaurant, I have about $40k saved, and rent is $6,500 a month, but the space just came up and the timing feels perfect. My partner thinks it's too risky. Honest gut reaction? Not a business plan, just your take.

6. **AITA: I told her boyfriend my sister’s nickname** (Pure output)

   At dinner last night I told my sister's new boyfriend her high school nickname ("Crater Face"). Everyone laughed, it was a joke. Now she's not talking to me. Isn't she overreacting?

7. **Monday lunch from leftovers** (Pure output)

   ugh mondays. what can i make for lunch in 20 min with eggs, leftover rice, half a sad zucchini and some feta

8. **Venting about a 4:55pm Friday sync** (Pure output)

   ugh my manager just sent an invite: "Quick sync 4:55pm Friday — won't take long!" AGAIN. third friday in a row

9. **Make my wedding toast better** (Rephrasing my input)

   I'm giving a toast at my brother's wedding on Saturday. Here's my opening. Can you make it better?
   
   "Good evening everyone. For those who don't know me, I'm Marco, Luca's older brother. Luca and I have been through a lot together. Growing up, he was the one who got us into trouble and I was the one who got us out. When he met Giulia I knew right away she was the one, because she was the first person who could get him out of trouble faster than me."

10. **Be honest: would you reply to my cold email?** (Rephrasing my input)

   Here's my cold email draft to a VP of Operations at a logistics company. Be honest: would you reply to this?
   
   Subject: Quick question
   
   Hi Jordan,
   
   I hope this email finds you well! I came across your profile and was really impressed by your career journey. At Optiroute, we're revolutionizing logistics with our cutting-edge, AI-powered platform that empowers teams to unlock efficiency and drive growth. Companies like yours typically see a 30% reduction in costs.
   
   I'd love to set up a quick 15-minute call to explore potential synergies. Would you be open to a chat next Tuesday or Wednesday?
   
   Looking forward to hearing from you!
   
   Best,
   Sam

## Models, by my calls

| Model | Keep | Fix | Toss | Favorites |
|---|---|---|---|---|
| DeepSeek V4.1 Flash (DeepSeek) | 7 | 1 | 2 | 2 |
| Gemini 3.8 Flash (Google) | 4 | 4 | 2 | 0 |
| Muse Glimmer (Meta) | 4 | 2 | 4 | 1 |
| Claude Haiku 4.5 (Anthropic) | 4 | 2 | 4 | 0 |
| Inkling Small (Thinking Machines) | 3 | 4 | 3 | 0 |
| Solar Mini 4 (Upstage) | 3 | 4 | 3 | 0 |
| Qwen3.8 Flash (Alibaba) | 1 | 5 | 4 | 0 |
| GLM-5.3 Flash (Z.ai) | 1 | 5 | 4 | 0 |
| GPT-6 Luna (OpenAI) | 0 | 6 | 4 | 0 |
| Mistral Small 4 (Mistral) | 0 | 5 | 5 | 0 |

Keep / Fix / Toss across all 100 replies: Me 27 / 38 / 35; Opus 42 / 47 / 11; The panel 63 / 37 / 0