frontpage.
newsnewestaskshowjobs

Made with ♥ by @iamnishanth

Open Source @Github

fp.

Open in hackernews

Show HN: Neural window manager, neural network moving windows from mouse actions

https://lusob.github.io/neural-os/
2•lusob•1h ago
I'd been mulling over this crazy idea for a while. Can programs be generated? Inspired by recent advances in world models, I wondered if we could do away with source code and generate pixels directly and interactively.

As an experiment to answer this, I set out to create a neural window manager, training a neural network to predict what the screen would look like next.

Basically, the idea was to generate the next frame based on the last two frames and the mouse position. That's it: moving windows without programming an event system, just a simple convolutional neural network guessing pixels.

To implement the experiment, I used Pygame to simulate a turquoise desktop background, a gray window with a navy blue title bar, a white cursor, and four colors in total. Then, a bot randomly dragged the window, and I recorded everything, processing the frames as color index matrices (not RGB, to avoid complications) and the mouse delta (dx, dy, click) that caused each transition. 8000 frames, a few minutes in Colab.

The model is a unitary neural network (UNET). The encoder compresses the stacked frames, the decoder reconstructs the next one, and the mouse vector coordinates are projected with a linear layer to fit the spatial size of the bottleneck. There, they are concatenated before decoding, so that motion information feeds each jump connection.

And it works! Which still surprises me a little. You can drag, and the window follows you; when you release, it stops. There's no internal state, no (x, y) coordinates anywhere. The model infers the position from what it sees, which works until it doesn't. But after a couple of seconds of strange movement, the window starts to distort.

This will probably improve with more computing power for training and more examples, but to narrow the scope of the experiment and test it within a web browser, I decided to abandon the rendering aspect and have the model predict primitives instead of pixels, simply converting the motion engine into a neural network.

Basically, I trained a small MLP to receive (distance to the title bar, distance to the resize point, click) and generate (dx, dy, dw, dh), with two separate heads: one for moving and one for resizing. The trick is that they share nothing except the click signal, so the model can't confuse dragging with resizing. I then exported it to ONNX as well, and now everything runs in the browser, without a server, just a canvas element and two small neural networks communicating with each other.

With this new approach, the renderer remains deterministic, with rectangles drawn in JavaScript, but the window's behavior (where it moves, how it resizes) is learned from examples. It feels like a peculiar middle ground between traditional and neural, so you can feel the space the network has learned by interacting with it: dragging near the title bar moves it, but approaching the corner resizes the window. There are no conditionals or hitbox code; the network simply learned where those areas are from examples.

Sometimes it gets confused near the edges, which, frankly, is more interesting than if it worked perfectly; you can perceive how the probability changes. This makes sense when you think about it, because no (x, y) coordinates are stored in these models; the position is implied in the activations. It works well for short sequences, but fails when asked to maintain state over time.

Update: A few weeks later, Meta published the Neural Computers article (2604.06425, it's worth reading). The premise is the same, but they go much further: cli and uis, real programs. Their failure modes are practically identical to those I found with the pure pixel version: "challenges persist with routine reuse, controlled updates, and symbolic stability." which is a fancy way of saying that the window blurs after a few seconds (that was the reason for choosing deterministic rendering).

C-suite exodus at Boston Dynamics

https://www.semafor.com/article/05/01/2026/c-suite-exodus-at-boston-dynamics
1•ilamont•1m ago•0 comments

How My Mother Made Magic Happen

https://lithub.com/how-my-mother-made-magic-happen/
1•speckx•1m ago•0 comments

China to Invest in DeepSeek at $50B Valuation

https://www.wsj.com/tech/ai/china-to-invest-in-deepseek-at-50-billion-valuation-045041d0
1•gmays•2m ago•0 comments

Using QUIC backscatter to infer hypergiant deployment configurations

https://blog.apnic.net/2026/04/21/using-quic-backscatter-to-infer-hypergiant-deployment-configura...
1•PaulHoule•6m ago•0 comments

18 year old critical vulnerability found in Nginx

https://twitter.com/Markak_/status/2054599711764750498
6•ejcho•6m ago•0 comments

MAME now emulates a Stasi numbers-station speech synthesizer

https://blog.adafruit.com/2022/12/13/mame-now-emulates-a-stasi-numbers-station-speech-synthesizer/
1•austinallegro•7m ago•1 comments

Esploro.app – Modern, sleek and fast SQL client for Mac (open-source / free)

https://esploro.app/
1•mmunj•7m ago•0 comments

Reading the Gerlich AI/critical thinking paper more closely

https://old.reddit.com/r/AICognitiveWatch/
1•GlyphWeaver_a•10m ago•0 comments

Mammalian-like steroidogenesis in plants gives rise to cardenolides

https://www.science.org/doi/10.1126/sciadv.aeb5460
1•PaulHoule•10m ago•0 comments

Altman forced to confront claims at OpenAI trial that he's a prolific liar

https://arstechnica.com/tech-policy/2026/05/altman-forced-to-confront-claims-at-openai-trial-that...
7•Bender•10m ago•0 comments

The Wide Paved Road versus the Narrow Footpath

https://peterspath.net/blog/essay-the-wide-paved-road-versus-the-narrow-footpath/
1•peterspath•11m ago•0 comments

Takeda, slimming down for 'new era,' plots 4,500 restructuring layoffs

https://www.fiercepharma.com/pharma/takeda-slimming-down-new-era-plots-4500-layoffs-latest-restru...
1•randycupertino•11m ago•1 comments

Nginx.org/En/Changes

https://nginx.org/en/CHANGES
2•qwertox•14m ago•0 comments

Scanread.ai – Free OCR for images, PDFs, and handwriting in 100 languages

https://scanread.ai
2•voqusa•15m ago•1 comments

Is Google or Exa better for RLing agents?

https://exa.ai/blog/rl-search-outcomes
1•willbryk•15m ago•1 comments

Figure humanoid robot package handling live stream [video]

https://www.youtube.com/watch?v=luU57hMhkak
4•Animats•16m ago•1 comments

Freedom to Drive Initiative

https://ops.fhwa.dot.gov/freedom-to-drive-initiative.htm
2•aziaziazi•16m ago•1 comments

Build a Basic AI Agent from Scratch

https://medium.com/@rogi23696/build-a-basic-ai-agent-from-scratch-0be3e3022b55
1•ruxudev•16m ago•0 comments

Notepad++ Mac Port Renamed Nextpad++ After Trademark Row

https://www.macrumors.com/2026/05/13/notepad-plus-plus-mac-nextpad-dispute-settlement/
2•01-_-•16m ago•0 comments

Show HN: Micromort Risk Visualizer

https://boxed.github.io/micromort/
3•boxed•18m ago•0 comments

Verisign to delete .name 3LDs and email addresses

https://domainincite.com/31699-verisign-to-delete-name-3lds-and-email-addresses
2•zacwest•19m ago•1 comments

An overview of NVMe and its support on Maestro

https://blog.lenot.re/a/nvme
1•PaulHoule•20m ago•0 comments

Neanderthals drilled cavities to treat a toothache 59,000 years ago

https://arstechnica.com/science/2026/05/neanderthals-drilled-cavities-to-treat-a-toothache-59000-...
4•Bender•26m ago•0 comments

MacBook Neo Deep Dive: Benchmarks, Wafer Economics, and the 8GB Gamble

https://www.jdhodges.com/blog/macbook-neo-benchmarks-analysis/
1•tosh•26m ago•0 comments

A new US Military wargame series began by simulating a nuclear weapon in orbit

https://arstechnica.com/space/2026/05/a-new-us-military-wargame-series-began-by-simulating-a-nucl...
2•Bender•26m ago•0 comments

Tinyhumansai/openhuman: Your Personal AI super intelligence

https://github.com/tinyhumansai/openhuman
1•ulrischa•28m ago•0 comments

Restricting automatic kernel-module loading (2017)

https://lwn.net/Articles/740455/
1•kerneloops•29m ago•0 comments

Apple Working on Plan to Allow AI Agent Apps on the App Store

https://www.macrumors.com/2026/05/13/apple-ai-agent-apps-app-store/
1•thm•31m ago•0 comments

A sentimental tour of late 1990s and early 2000s hacking tools

https://andreafortuna.org/2026/05/13/amarcord/
3•speckx•32m ago•0 comments

Use the Claude Agent SDK with Your Claude Plan

https://support.claude.com/en/articles/15036540-use-the-claude-agent-sdk-with-your-claude-plan
7•AlexErrant•32m ago•4 comments