DEV-TOOLSJune 13, 20261 min read9 views

Shepherd's Dog: A Game by the Most Dangerous AI Model

A developer tested Anthropic's latest, supposedly 'too dangerous' AI model by asking it to build a long-held game idea in a single shot. The model succeeded, generating a complete 2,319-line game after a 45-minute reasoning session.

A few days ago, Anthropic released the model that was supposedly too dangerous for the world to see. I wanted to give it a try using my personal benchmark: can it build a game idea I've had for years, in a single shot?

The model started with a very long reasoning session, and after 45 minutes and more than €20 worth of tokens (!), it happily reported that "Shepherd's Dog is complete — a single 2,319 line index.html with zero dependencies."

You can watch the video below or give the game a try It's really fun and exactly how I imagined it. For me this is the first time an AI model is able to create this for me in one go:

Play the game here: vnglst.github.io/when-ai-fails/shepards-dog/claude-fable-5/index.html
Other attempts with earlier models: github.com/vnglst/when-ai-fails/blob/main/shepards-dog/README.md

Source: Hacker News

More in this category

dev-tools

Kimi Linear: An Expressive, Efficient Attention Architecture

Kimi Linear introduces a novel hybrid linear attention architecture that outperforms full attention while reducing KV cache usage by up to 75% and boosting decoding throughput by up to 6x for long contexts.

dev-tools

Ars Astronomica – English translations of rare Hebrew and Latin astronomy texts

A scholarly imprint producing first English translations of historical Hebrew and Latin works in astronomy, cosmology, and natural philosophy.

dev-tools

Neutrino-1 8B

Fermion Research introduces Neutrino-1 8B, an 8.19B-parameter LLM packaged in a 3.88 GB file using a proprietary ternary weight format for high-speed cross-platform inference.

dev-tools

The age of token efficiency, the age of libraries

As AI adoption among developers reaches unprecedented levels, the industry faces severe challenges regarding code trust, token cost efficiency, and developer accountability. This article explores how software engineering is shifting from pure implementation to orchestration and library-based reliability.

dev-tools

Using an open model feels surprisingly good

Switching from proprietary AI services like Claude and ChatGPT to a self-hosted open model creates a surprisingly liberating experience. By running opencode on a personal inference endpoint, developers can regain full ownership of their data and workflow.

dev-tools

A $500 RL fine-tune of a 9B open model beat frontier models on catalog review

Enterprises are increasingly adopting a playbook combining open-source models, proprietary data, and reinforcement learning. This approach consistently beats top frontier AI models in specialized real-world tasks while significantly reducing inference costs.