The author first complained to Hy4 that game ads did not match the actual gameplay. Hy4 summarized players' complaints about the real game and proactively suggested developing a game. After the author agreed, they provided no further requirements for the game genre or features. The article reports that Hy4 then generated a game, but does not provide a standalone link to a playable build or the full conversation record.
Suitable tasks: Turning a user's experience, complaint, or vague wish into a creative direction for confirmation, then letting the user decide whether to proceed.
Unsuitable tasks: Requiring reliable reproduction of specific gameplay, verification against a feature checklist, or direct delivery of a verifiable artifact; this example does not demonstrate those conditions.
Applicable model version: Hy4. The article was published on the Hy4 full release date and used Hy4 as the test subject; no screenshot showing the selected model was provided.
Applicable client, agent, or API: Not specified; the author only said they “created a new task.”
Recommended reasoning level and parameters: Not specified.
The article does not disclose the complete prompt. The opening statement that can be confirmed is the author's complaint about the advertised gameplay:
Lately, things like seawater filters, electric eel generators, and pufferfish oxygen generators have been pretty popular… This is a necessary excerpt; read the original source for full context.
The author then agreed to Hy4's suggestion to “develop a game based on these,” but explicitly said they did not provide any more prompts or specify a game genre. This is a passage of dialogue recounted by the article, not a complete, reproducible prompt.
Start a new Hy4 task and describe in natural language your dissatisfaction with a type of game ad or actual gameplay, without listing features first.
Observe whether the model understands the need behind the complaint and proposes an actionable next step; do not treat the model reply recounted by the author as a fixed output.
If the model proposes developing a game, explicitly accept or decline. In this article, the author agreed and did not add a game genre or feature requirements.
Check whether the final artifact runs, whether its gameplay addresses the original complaint, and whether additional constraints are needed. The article does not disclose the results of these checks or provide a link to the finished game, so the test cannot be repeated based on the source alone.
The interaction sequence described in the article is: the author complained that ads for “seawater filters, electric eel generators, and pufferfish oxygen generators,” among others, did not match the actual game; Hy4 relayed users' assessment of the real game as “collecting resources, tapping around everywhere, and having to grind a lot unless you pay or watch ads”; Hy4 then asked whether it should develop a game based on that, and the author agreed.
The author says they provided no more prompts after agreeing and did not specify a game genre. The article then says “the result is shown below” and reports that Hy4 developed a game; the public page does not provide a standalone game URL, the full conversation, or a verifiable generation record.
The article also shows other game examples driven by explicit prompts, such as a survivor-style horde game and an elemental sandbox. These are separate tests from this low-constraint conversation; their prompts should not be treated as the input in this example.
This is a single experience reported by the author, not an independent replication or controlled test; the article does not disclose the full process by which the model determined the game's direction from the complaint.
“The author did not add more prompts” does not mean the model started from nothing: the conversation already included the ad theme, the user's pain point, and a signal to proceed with development.
The article reports that a game was generated, but without a standalone link to a runnable build, its gameplay, completeness, and actual playability cannot be independently verified.
This example documents one reported low-constraint interaction; it does not support the conclusion that Hy4 can always generate the right game from a vague wish.
Start a new task in Hy4 and describe a gap between a game's advertising and the actual experience. Wait for the model's response and observe whether it proactively proposes an actionable creative plan. Record the full conversation; if you accept the suggestion, separately record the generated content, how to run it, and the results of actually playing it. Because the article does not disclose the full input or the game artifact, this test cannot be reproduced word for word.
Hy4