Do Thought Streams Matter? A Benchmark of VLM Reasoning in Gemini 2.5
Names compression-step hallucination, but it's a paper not a tool you can use.

The project is a one-note delight: prompt an LLM to speak like the 25th Hour mirror scene and feed it the HN front page, and you get punchy, topical rants that land because the copy is front-and-center. The site wisely keeps everything minimalist so the text reads like a performance piece, but there’s no account features, provenance for the model/prompts, or ways to subscribe — it’s a clever content experiment rather than a product with staying power.
Hacker News readers, developers, AI enthusiasts, people who enjoy tech satire
So here it is, 25th Hour mirror rant scene with Edward Norton meeting HN front page. It was some random idea, I was curious what AI will output and it turns out it's really good at it.
Hope you like it as much as I do.
Names compression-step hallucination, but it's a paper not a tool you can use.
Entire 3D scenes stored in URLs—clever constraint, but limited by URL length and browser storage.
Nice sunshine heatmap UX but Apple Weather and Carrot already solve this.
Single image to explorable 3D scene is technically impressive but mostly a novelty demo.
Pixel 3D editor running entirely in-browser — nothing leaves your machine.
The visual concept is strong: generous negative space, a micro-typographic centerline and restrained UI that actually sells the idea of 'ma'. Unfortunately the page reads like a demo/theme — repeated copy blocks and zeroed stats suggest placeholder content rather than a finished system or component library. Useful as a quiet starting point for designers who like minimalism, but it needs real docs, components and examples to move beyond a pretty landing.