Claude creates a playable shooting game with a short prompt.
Decrypt
07-29 02:33
Ai Focus
Claude Opus 5 uses three prompts to generate a playable shooting game, which was subsequently replicated by several developers, shifting the focus to multi-agent collaboration and prompting engineering methods.
Helpful
No.Help

Following the release of Claude Opus 5 by Anthropic, a short three-word prompt has garnered attention in the developer community. AI investor and former HyperWrite CEO Matt Shumer showcased a first-person shooter game that runs in a browser, claiming it was built entirely by Claude in a single session and that no external assets were used in the project.

Three prompts drive collaboration

This case has garnered discussion not only because the finished product is playable, but also because the prompts themselves are very short. Unlike past prompt engineering approaches that emphasized detailed requirements, this method only sets the goal: to create a shooting game approaching the quality of recent Call of Duty titles, then breaks down the task into multiple sub-agents and allows independent reviewers to repeatedly identify and correct errors.

Shumer later released the complete prompts and code repository, calling the method the "Gauntlet Loop." The core approach is to first give the model a clear, observable, and comparable goal, rather than a long list of construction details; then let the model break down the task itself, assigning different parts to independent sub-agents; finally, a review module that is not involved in the construction continuously scores and provides suggestions for improvement.

The finished product runs in the browser

According to publicly available information, this game runs on Three.js and WebGL2, with approximately 55,000 lines of code distributed across 11 subsystems. Textures, models, animations, and sound effects are all generated when the browser loads; there are no additional downloads of model, image, or audio files.

Shumer's review logs show that the system underwent continuous improvement through multiple iterations, but its score consistently lagged behind real Call of Duty footage. This means that the so-called "one-time generation" does not mean that the model completes all the work directly in a few minutes, but rather that the results are gradually improved through repeated corrections within a single session.

Reproduction and doubts arise simultaneously

After the hint was made public, several developers began to reproduce the game. Former hedge fund manager James Altucher used the same hint and claimed to have spent about 10 hours and about 1.3 million tokens to create a browser-based shooter game called "Operation Blackout." Other developers submitted the same request to models within the OpenAI ecosystem, demonstrating similar results.

However, the point of contention lies in whether the "generating from scratch" approach holds true. First-person shooter games are among the most resource-rich genres in terms of programming, with the relevant perspective control, movement, and shooting logic already abundantly available in open-source codebases. The article mentions that researchers often refer to this type of problem as "data pollution," meaning that a model's excellent performance may partly stem from a large number of approximate samples already present in the training data, rather than entirely from new solutions derived through on-the-spot reasoning.

Tip
$0
Like
0
Save
0
Views 517
HQYC reminds readers to view blockchain rationally, stay aware of risks, and beware of virtual token issuance and speculation. All content on this site represents market information or related viewpoints only and does not constitute any form of investment advice. If you find sensitive content, please click“Report”,and we will handle it promptly。
Submit
Comment 0
Hot
Latest
No comments yet. Be the first!
Related
Claude's shared chat was once searchable on Google.
Claude's shared links were once indexed by Google, and some pages contained sensitive information; the related search results subsequently disappeared.
TechCrunch
·2026-07-28 04:22:05
652
web3: MoonPay launches PayBox, integrating crypto wallets with Claude and ChatGPT
MoonPay launches PayBox, which embeds encrypted wallets into Claude and ChatGPT, enabling AI agents to complete payments and on-chain operations after user confirmation.
Decrypt
·2026-07-30 03:43:35
846
Claude's shared conversation was previously indexed by Google; Anthropic has fixed this.
Claude's share link was previously indexed by Google; Anthropic has fixed the relevant configuration.
Decrypt
·2026-07-28 03:13:31
157
web3: Anthropic releases Claude Opus 5, priced lower than Fable 5
Anthropic released Claude Opus 5, claiming it outperformed Fable 5 in multiple tests and has a lower API price.
Decrypt
·2026-07-25 03:38:15
1040
web3: Anthropic claims Claude discovered a new attack on post-quantum signatures.
Anthropic stated that Claude Mythos Preview discovered new attack methods using HAWK and 7 rounds of AES, indicating that AI has begun to enter the field of high-strength cryptanalysis research.
Decrypt
·2026-07-29 06:12:43
458