Shepherd's Dog: A Game by Fable
An AI model was prompted to generate a simple shepherding game in one shot, impressing many with how quickly it produced smooth controls, coherent mechanics and decent visuals, but also drawing criticism for bugs, poor mobile UX and the game’s derivative design. Commenters debate whether such outputs represent genuine creativity or just recombination of training data, especially given that nearly identical games already exist. The thread broadens into questions about how AI-assisted coding affects learning, software maintenance, originality, and the future role of human developers in game and app creation.
Game and UX Feedback
- Several users share a direct play link and note that the article’s link chain is annoying.
- Multiple reports of poor mobile UX: forced landscape rotation, overlays blocking instructions, top browser bars hiding half the play area, and unusable controls on some phones.
- Specific complaints: dog pathfinding problems, awkward barking controls, nonfunctional sound, low-contrast text on light backgrounds.
- Others find it “pretty great” for a quick browser game and say they had fun.
Originality and Prior Art
- Many point out the core mechanic is decades old, with examples on Game Boy Advance, Zelda, mobile app stores, and existing GitHub/itch.io projects.
- Some argue describing it as “an idea I’ve had for years” feels naive or misleading, given how common the concept is.
- Others counter that most games are recombinations of existing mechanics; originality of ingredients matters less than quality of synthesis.
AI Capability and Technical Discussion
- Several commenters are impressed that an LLM can one‑shot a smooth, reasonably balanced browser game, especially relative to earlier attempts.
- Others say this is more like “git pull” from training data than true creativity.
- Comparisons are made with cheaper or open models (e.g., DeepSeek, Qwen), which can also generate playable versions but with rougher behavior.
- Debate over “one‑shot” vs. incremental, architect‑driven development: many argue maintainability and serious projects require human‑designed structure, with AI as an accelerator.
Creativity, Learning, and “Taste”
- Long subthread on whether “all human work is derivative” vs. the role of research, iteration, and taste in making something good.
- Some worry genAI shortcuts bypass the traditional struggle with originality and exposure to prior art, weakening creative development.
- Others treat AI‑generated prototypes as a way to cheaply test ideas and focus human effort on design, playtesting, and refinement.
Ethics, Training Data, and Ownership
- Concerns that such games are effectively regurgitations of copyrighted or non‑open code and assets.
- One side frames this as “theft of all human knowledge”; others reply that derivative use of prior work is how humans create too.
Economic and Community Reactions
- Mixed views on paying ~€20 in tokens for a simple game: some see it as wasteful versus hiring or learning; others note people routinely pay premiums for custom work and for the “I helped build it” feeling.
- Several see the “world’s most dangerous AI” title as clickbait driving front‑page attention rather than reflecting the game’s actual stakes.