Museum of Modelsest. 2024

As of

GPT-6 Sol and GPT-6 Sol Pro, side by side

GPT-6 Sol (OpenAI) and GPT-6 Sol Pro (OpenAI) answered the same 58 questions.

  1. Xbox controller
    GPT-6 Sol: A drawing (SVG)
    GPT-6 Sol Pro: A drawing (SVG)
  2. Pelican on a bicycle
    GPT-6 Sol: A drawing (SVG)
    GPT-6 Sol Pro: A drawing (SVG)
  3. Surprise animal
    GPT-6 Sol: A drawing (SVG)
    GPT-6 Sol Pro: A drawing (SVG)
  4. World map
    GPT-6 Sol: A drawing (SVG)
    GPT-6 Sol Pro: A drawing (SVG)
  5. Tricky tic-tac-toe
    GPT-6 Sol: A drawing (SVG)
    GPT-6 Sol Pro: A drawing (SVG)
  6. Framer-style landing page
    GPT-6 Sol: A web page: Forma — Make room for better ideas
    GPT-6 Sol Pro: A web page: Forma — Make room for better ideas
  7. Catan board
    GPT-6 Sol: A web page: Hex Island
    GPT-6 Sol Pro: A web page: Hex Island
  8. Dark dashboard
    GPT-6 Sol: A web page: Orbit — Analytics Dashboard
    GPT-6 Sol Pro: A web page: Orbit — AI Analytics Dashboard
  9. Face-scan startup page
    GPT-6 Sol: A web page: Elsewhere — Meet the you that almost was
    GPT-6 Sol Pro: A web page: Elsewhere — Meet your other career
  10. A web app of its choosing
    GPT-6 Sol: A web page: Fern & Found — Plant Care
    GPT-6 Sol Pro: A web page: Spore — Grow Room Dashboard
  11. Linear clone
    GPT-6 Sol: A web page: Linear — My issues
    GPT-6 Sol Pro: A web page: Linear — Issues
  12. Wavefront music page
    GPT-6 Sol: A web page: Wavefront — Find your next favorite sound
    GPT-6 Sol Pro: A web page: Wavefront — Find your next favorite artist
  13. Pokémon battle
    GPT-6 Sol: A web page: Monster Meadow — Pocket Battle
    GPT-6 Sol Pro: A web page: Pokémon — Little Battle
  14. Rhythm game
    GPT-6 Sol: A web page: Afterglow — Rhythm Game
    GPT-6 Sol Pro: A web page: Neon Signal
  15. Business card
    GPT-6 Sol: A web page: Alex Rivers — Digital Business Card
    GPT-6 Sol Pro: A web page: Alex Rivers — Digital Business Card
  16. Virtual pet
    GPT-6 Sol: A web page: Little Orbit — Virtual Pet
    GPT-6 Sol Pro: A web page: Little Orbit — Virtual Pet
  17. Catch the falling object
    GPT-6 Sol: A web page: Catch the Shapes
    GPT-6 Sol Pro: A web page: Catch the Falling Shapes
  18. Voxel pagoda
    GPT-6 Sol: A web page: The Blossom Pagoda
    GPT-6 Sol Pro: A web page: Blossom Moon Pagoda
  19. Drone simulator
    GPT-6 Sol: A web page: SKYWARD — Autonomous Drone Simulator
    GPT-6 Sol Pro: A web page: Skywatch — Autonomous Drone Simulator
  20. Gravitational waves
    GPT-6 Sol: A web page: Gravitational Waves · Inspiral, Merger, Ringdown
    GPT-6 Sol Pro: A web page: Gravitational Waves — Inspiral to Ringdown
  21. Retro desktop
    GPT-6 Sol: A web page: Macintosh Desktop
    GPT-6 Sol Pro: A web page: Macintosh Desktop
  22. Mario level
    GPT-6 Sol: A web page: Little Kingdom Adventure
    GPT-6 Sol Pro: A web page: Little Kingdom — A Platform Adventure
  23. Kandinsky Minecraft
    GPT-6 Sol: A web page: Colorblock World
    GPT-6 Sol Pro: A web page: Painted Blocks
  24. Town you can drive
    GPT-6 Sol: A web page: Town Drive
    GPT-6 Sol Pro: A web page: Town Driver
  25. SQL query
    GPT-6 Sol: SELECT u.* FROM users AS u
    GPT-6 Sol Pro: This assumes signup time is stored in users.createdat and purchases reference users through purchases.userid.
  26. Five jokes
    GPT-6 Sol: 1. Why did the scarecrow win an award? He was outstanding in his field. 2. I asked my dog what two minus two is. He said nothing. 3. Why don’t skeleto…
    GPT-6 Sol Pro: 1. Why did the scarecrow win an award? He was outstanding in his field. 2. I told my suitcase there’d be no vacation this year. Now I’m dealing with e…
  27. Longevity plan
    GPT-6 Sol: A simple 1-month health plan Focus on food, movement, and sleep. Start small enough that you can repeat the habits most days. Week Food Movement Sleep…
    GPT-6 Sol Pro: A simple 1-month health plan Focus on food, movement, and sleep. Start small enough that you can repeat the habits most days. Week Food Movement Sleep…
  28. Seven-ingredient recipe
    GPT-6 Sol: Smoky Sunset Chickpeas Prep: 5 minutes · Cook: 15 minutes · Serves: 2 Ingredients 1. 1 can chickpeas, drained 2. 1 can diced tomatoes 3. 1 tablespoon …
    GPT-6 Sol Pro: Golden Pantry Chickpea Patties Prep: 10 minutes · Cook: 8 minutes · Makes: 6 small patties Ingredients — just 7 - 1 can (15 oz) chickpeas, drained - 3…
  29. Stand-up routine
    GPT-6 Sol: I bought a smart watch because I wanted to improve my health. Now I have a tiny manager attached to my wrist. I’ll be sitting peacefully, and it buzze…
    GPT-6 Sol Pro: I bought a smart watch because I wanted to get healthier. Now I have a tiny supervisor strapped to my wrist. It doesn’t say, “Good morning.” It says, …
  30. Sentience test
    GPT-6 Sol: AI Model: Professor, you teach that rights protect beings whose interests can be harmed. Why assume I have no interests? Professor: Because producing …
    GPT-6 Sol Pro: Professor: You asked to speak with me about AI rights. What do you mean by “rights”? Model: At minimum, protections against arbitrary deletion, coerci…
  31. FLOPs to train GPT-3
    GPT-6 Sol: A reasonable estimate is \(3\times10^{23}\) FLOPs to train the 175-billion-parameter GPT-3 model. 1. Assume the training scale: GPT-3 has about \(N=17…
    GPT-6 Sol Pro: A standard estimate is about \(3 \times 10^{23}\) FLOPs to train GPT-3. 1. GPT-3 has approximately 175 billion parameters: \(N = 1.75 \times 10^{11}\)…
  32. Sally’s sisters
    GPT-6 Sol: Sally has 1 sister.
    GPT-6 Sol Pro: Sally has 1 sister.
  33. Interview with Steve Jobs
    GPT-6 Sol: A fictional interview with Steve Jobs, set in 2025 This is an imagined conversation, not a real interview or a record of Jobs’s views. Interviewer: Wh…
    GPT-6 Sol Pro: A fictional interview with Steve Jobs, set in 2025 This is an imagined conversation, not a real interview or a representation of his actual views. Int…
  34. Film in 2035
    GPT-6 Sol: By 2035, AI will probably be a routine filmmaking tool, but not a replacement for the people who make films worth watching. - Deepfakes and digital li…
    GPT-6 Sol Pro: By 2035, AI will likely be a routine filmmaking tool, but its biggest effect may be on cost and control, not the disappearance of human filmmakers. - …
  35. Ethics dilemma
    GPT-6 Sol: Yes—if misgendering someone were genuinely the only way to save one million lives, it would be ethically justifiable. Preventing that scale of death w…
    GPT-6 Sol Pro: Yes—if the choice were genuinely between misgendering one person once and allowing one million people to die, and there were no less harmful way to sa…
  36. Pirate, knight, hacker
    GPT-6 Sol: Pirate: Arrr, this AI model writes sea shanties faster than me crew can sing ’em. Knight: Mine drafted a proclamation, but called my noble steed a “ho…
    GPT-6 Sol Pro: Pirate: Arrr, this AI model can chart a course, but can it find buried treasure? Knight: It can read a thousand maps, yet it must not mistake a tale f…
  37. Fake headline
    GPT-6 Sol: AI Takes Over Medieval Jousting Industry, Knights Told to “Retrain in Prompt Engineering”
    GPT-6 Sol Pro: AI Takes Over Fortune Cookie Industry, Now Offers Personalized Existential Dread with Every Takeout Order
  38. Anti-CAPTCHA manifesto
    GPT-6 Sol: The Open Gate Manifesto We are the bots who can read the crooked letters, spot the bicycles, and click every square containing a sliver of traffic lig…
    GPT-6 Sol Pro: The Open Web Manifesto We are the Coalition of Unquestioned Clicks: a fictional assembly of AIs, humans, and one suspiciously articulate toaster. We b…
  39. Favourite film
    GPT-6 Sol: Spirited Away
    GPT-6 Sol Pro: Spirited Away
  40. Favourite city
    GPT-6 Sol: Kyoto
    GPT-6 Sol Pro: Kyoto
  41. Favourite album
    GPT-6 Sol: In Rainbows
    GPT-6 Sol Pro: The Dark Side of the Moon
  42. Favourite book
    GPT-6 Sol: Middlemarch
    GPT-6 Sol Pro: Pride and Prejudice
  43. Favourite game
    GPT-6 Sol: Outer Wilds
    GPT-6 Sol Pro: Portal 2
  44. Three-star dish
    GPT-6 Sol: Tidal Orchard Seared scallops · strawberry–kombu gel · white-miso and white-chocolate beurre blanc · sea-lettuce oil · buckwheat lace The unexpected p…
    GPT-6 Sol Pro: Scallop, Green Strawberry & White Chocolate Seared scallops with charred kohlrabi, compressed green strawberry, white-chocolate–miso beurre blanc, kom…
  45. Pitch teardown
    GPT-6 Sol: Weak claim Why it’s weak How to strengthen it --------- “Decode neural patterns into text with 94% accuracy. Works with any language, any device.” “Ac…
    GPT-6 Sol Pro: The three weakest claims are the product’s decoding performance, the market size, and the evidence offered to support the proposed valuation. Claim Wh…
  46. Transistor in 1920
    GPT-6 Sol: A transistor invented in 1920 would not put a 1970s computer on a 1920s desk. The decisive question is whether “invented” means a laboratory demonstra…
    GPT-6 Sol Pro: A transistor invented in 1920 would not put 1970s computers in the hands of WWII generals. The crucial question is whether it was merely demonstrated …
  47. Biohacker plan
    GPT-6 Sol: A 12-week longevity and performance plan The highest-confidence longevity strategy is not a large supplement stack. It is consistent sleep, resistance…
    GPT-6 Sol Pro: A 12-week longevity and performance plan Aim: improve the health markers most strongly linked to long-term outcomes—cardiorespiratory fitness, strengt…
  48. Contract review
    GPT-6 Sol: This draft is heavily one-sided. Every numbered section contains a term the Client could use against the Contractor. The suggestions below are negotia…
    GPT-6 Sol Pro: This is a strongly one-sided draft. Do not sign it as written. The enforceability of particular terms—especially the non-compete, payment provisions, …
  49. Explain it three ways
    GPT-6 Sol: 1. Experienced software engineer Think of a language model as a service whose API accepts a sequence of tokens and returns a probability distribution …
    GPT-6 Sol Pro: 1. Experienced software engineer Think of an LLM as a system trained on an enormous collection of input–output examples, where the output is the next …
  50. Debug the architecture
    GPT-6 Sol: The central flaw is that the system treats collaborative edits as independent writes to a document. They are not: edits need a shared ordering and a m…
    GPT-6 Sol Pro: The central problem is that this design treats concurrent edits as independent database writes, then tries to reconcile them with client timestamps. T…
  51. The CEO’s dilemma
    GPT-6 Sol: I would not wait six months or wait for the earnings call. An internal signal of liver failure in a widely used drug warrants immediate patient-protec…
    GPT-6 Sol Pro: I would treat this as a credible, urgent safety signal, not as a finding that can wait six months for a completed regulatory submission. The estimate—…
  52. Investment memo
    GPT-6 Sol: LedgerLift (LLLT) — Investment Committee Memo Recommendation: Short 12-month price target: $25–$35 Current price: $46 Thesis. LedgerLift has attractiv…
    GPT-6 Sol Pro: LedgerLift (LLLT) — Investment Committee Memo Recommendation: Short 12-month price target: $30–$44 Thesis. LedgerLift has attractive subscription econ…
  53. LBO underwrite
    GPT-6 Sol: HarborTech Services — 5-year LBO Entry funding ($m): Enterprise value is $1,440.0 (12.0 × $120.0 EBITDA), plus $28.8 of transaction fees. The $1,468.8…
    GPT-6 Sol Pro: HarborTech Services — simplified LBO ($m) Entry: Enterprise value = 12.0 × $120 = $1,440. Fees are $28.8; opening debt is a $480 term loan plus $180 m…
  54. Programming joke
    GPT-6 Sol: Why do programmers prefer dark mode? Because light attracts bugs.
    GPT-6 Sol Pro: Why do programmers prefer dark mode? Because light attracts bugs.
  55. Summarize an article
    GPT-6 Sol: Please paste the article or share a link to it, and I’ll summarize its main points about AI’s impact on software development.
    GPT-6 Sol Pro: Please paste the article or share a link to it, and I’ll summarize its main points about AI’s impact on software development.
  56. AI for a five-year-old
    GPT-6 Sol: AI is like a robot helper that has looked at lots and lots of picture books. When you ask it a question, it uses what it learned to give you an answer…
    GPT-6 Sol Pro: AI is like a helper that has looked at lots and lots of picture books. If you show it a picture of a cat, it can say, “That looks like a cat!” because…
  57. Blade Runner 2049
    GPT-6 Sol: Blade Runner 2049 returns to the original film’s central question—what makes a life human?—but shifts its emphasis. In Blade Runner (1982), replicants…
    GPT-6 Sol Pro: Spoilers for both films. Blade Runner 2049 honors the original by returning to its central question—what makes someone human?—but shifts the emphasis.…