Video yükleniyor...
Video Yüklenemedi
Argon is quite different: "A game about Imminence. Something very big, very strange is happening. A suburb & the arrival of a vast & unknowable presence. Not horror, invoke the feeling of the end of all things coming, inevitably, but also not sad or scary"
32,366 görüntüleme • 1 ay önce •via X (Twitter)
22 Yorum

I'd say it's a breath of fresh air. The difference is certainly welcome. Even the slop is different.

the 7:14 pm timestamp bit is a nice touch. whoever wrote that clearly had a plan for the ending

Argon is the best Argon na tell everyone about it 😊

Argon’s version feels way more deliberate. The whole harmonic lock / residual impression system turns the same “something vast is happening in the suburbs” prompt into an actual interactive meditation instead of just vibes. Models are starting to have distinct creative instincts.

As someone who makes AI videos, I'd love a model that can hold a mood this specific across a whole sequence — imminence without horror is a hard brief. Curious how Argon does when you ask it to keep that tone shot to shot.

Professor drops an existential suburb game from one prompt. I ask for a to-do app and get three competing CSS frameworks 😵

This brief specifies a mood, not a mechanic. Models usually fake that with surface cues (fog, slow camera, ambient drone), so the real test is whether the game state itself builds dread, e.g. event pacing on a rising curve. Mood briefs are a better eval than spec clones: you can't unit-test dread.

Different personality is useful. The open question is whether Argon’s difference survives a second and third turn when the user pushes back on the first output. Most model gaps show up after the demo, not on the first generation.

Same one-line prompt, wildly different games. The variance between models is the real story here — frontier models are diverging like creative studios, not converging.

a vast unknowable presence arrives, the end of all things is coming, and somehow it's not sad. the model just reinvented advent.

You can also work with the emotion conveyed by an existing work of art. For example, I prompted this by adding that I want to subtly visualize the feeling tension in Phil Collins' "In the Air Tonight" in regard to Astra's imminent release.

末日感但不吓人,brief真难写

writing the mood brief that hard is the actual skill. most prompts i see just say make it atmospheric and then wonder why it turns into horror

way better than i expected from a one-shot. how many tries?

game design briefs are becoming the honest model eval

Haven't tried Argon yet. A one-prompt game is the kind of job I'd hand to my lead bot after a factory shift, then wait for my OK before anything goes out.

That "imminence without horror" brief is chef's kiss 🌫️ Did Argon invent the suburb loop, or lean on a known Twine/ink pattern under the hood?

Three of those instructions are negations: not horror, not sad, not scary. That's the hardest kind of brief to hand anyone, model or human designer. Curious if Argon actually handled the 'not' parts or just picked a calmer palette.

The fascinating shift here is that we're moving beyond testing whether AI can build something to testing whether it has a recognizable creative signature. Give two frontier models the same unusual emotional brief something inevitable and unsettling, but neither scary nor sad and the differences in interpretation may reveal more than another coding benchmark. Ethan, here's an experiment I'd love to see: What if you generated 5 games from each model using the exact same prompt, removed the model names, and asked people to identify which AI created each one? If people could consistently tell them apart, wouldn't that be compelling evidence that models are developing distinct creative tendencies rather than just producing random variations?

Sounds quite interesting

the prompt is doing more work than the model here. "not sad or scary" is such a specific constraint and it completely changes the vibe

That brief is brutal on purpose. "Inevitable ending, not sad, not scary" is the kind of constraint that usually breaks codegen. If Argon holds the mood for more than a minute, that's a better eval than another snake clone.


