Published on June 20, 2026 at 18:00 CEST (UTC+2)
CSSQuake (267 points by msalsas)
CSSQuake
CSSQuake is a browser-based first-person shooter game built entirely with CSS (using the PolyCSS engine). Players can move, shoot, and interact in a 3D environment using only CSS rendering, with options for dynamic lighting, particle effects, and multiplayer via game rooms. The project demonstrates the surprising capabilities of modern CSS and web standards for creating interactive experiences without JavaScript.
The European Social Stack (33 points by doener)
The European Social Stack
This open declaration advocates for building a decentralized, European-controlled social media ecosystem using open protocols like the Fediverse (Mastodon, PeerTube), Atmosphere (Eurosky, Tangled), and private messaging systems (Matrix, XMPP). It aims to reduce reliance on US-based monopolistic platforms, protect digital autonomy, and keep jobs and revenue within Europe. The initiative calls for joint efforts to create a resilient, democratic information infrastructure.
VPN ban update for UK households as government looks at 'age-gate' (52 points by iamnothere)
UK VPN Ban Update
The UK government is considering age-gating VPN usage as part of a broader under-16 social media ban, with details expected in July 2026. Children’s minister Josh MacAlister suggested that restricting VPN access could prevent minors from bypassing age restrictions. Challenges include effectiveness, privacy concerns, and enforcement, with Ofcom tasked to evaluate methods.
Bootimus – A Self-Contained PXE and HTTP Boot Server (54 points by car)
Bootimus – A Self-Contained PXE and HTTP Boot Server
Bootimus is a simple, single-binary PXE boot server that eliminates complex network configuration. It includes built-in proxyDHCP, automatic detection of 50+ operating systems, MAC-based access control, and one-click tools (e.g., GParted, Clonezilla). Designed for ease of use, it runs in Docker or as a standalone binary and supports JWT/LDAP authentication.
From PGP to Mythos: a brief history of export controls that didn't stop anyone (63 points by Brajeshwar)
From PGP to Mythos: Export Controls That Didn’t Stop Anyone
The article recounts how the US government ordered Anthropic to restrict export of its AI models Fable and Mythos, citing national security. The author draws parallels to past failed export controls on encryption (PGP) and spyware, arguing such measures are ineffective because technology crosses borders easily. The outcome could set a precedent for regulating frontier AI, but history suggests controls will be bypassed.
Where to Find the Colors Your Screen Can't Show You (317 points by moultano)
Where to Find the Colors Your Screen Can’t Show You
The author explains that many vivid cyan and other colors exist in nature but cannot be reproduced on standard RGB screens due to color space limitations (e.g., sRGB). They describe how to experience these “out-of-gamut” colors in the real world—through fluorescent lights, certain flowers, or prismatic effects. The piece highlights the gap between digital color reproduction and human perception.
I Stored a Website in a Favicon (239 points by theanonymousone)
I Stored a Website in a Favicon
A developer encoded an entire HTML page into the RGB pixel data of a favicon image, using steganography. The favicon, when loaded as an icon, renders the hidden website’s content (e.g., headings and paragraphs) directly in the browser tab. The project is a playful exploration of unconventional data storage in everyday web assets.
Web Browsers on PDAS (11 points by robin_reala)
Web Browsers on PDAs
This article surveys the history of full web browsers on personal digital assistants (PDAs) from the late 1990s to early 2000s, focusing on those that could parse standard HTML (not just WAP). It covers platforms like Palm OS, Windows CE, and EPOC, noting the challenges of limited screen size, memory, and network speeds. The author acknowledges gaps in historical records and invites corrections.
Mencius (5 points by jruohonen)
Mencius
The link points to a PDF file (raw binary content shown) from Indiana University, likely an academic paper or translation related to the ancient Chinese philosopher Mencius. The title alone suggests a work on Confucian philosophy, but the content is not readable from the preview. Score is low (5 points), indicating limited community engagement.
DOS Game "F-15 Strike Eagle II" reversing project needs DOS test pilots (4 points by LowLevelMahn)
DOS Game “F-15 Strike Eagle II” Reversing Project
A hobbyist reverse-engineers the 1989 DOS game “F-15 Strike Eagle II” by rewriting assembly code into C, having nearly completed all executables. The project now needs “test pilots” to help find bugs introduced during reconstruction, particularly in data layout. Community involvement is solicited to ensure the recreated game runs faithfully to the original.
Export controls on AI models are reemerging, but history suggests limited efficacy
The Anthropic Fable/Mythos case mirrors earlier failed attempts to restrict encryption and spyware. For AI/ML, this means frontier models will likely circulate through open-source derivatives, gray markets, or foreign labs. Developers should anticipate geopolitical friction and invest in self-regulation (e.g., responsible release frameworks) rather than relying solely on government bans.
Decentralized social platforms create new data pipelines for AI training
The push for a European Social Stack (Fediverse, Matrix) could fragment training data sources. AI models trained on centralized, US-based platforms may lose access to European content, while alternative datasets from decentralized networks become valuable. ML practitioners should monitor data availability shifts and consider federated learning or privacy-preserving techniques for multi-jurisdictional compliance.
Client-side AI inference benefits from extreme compression techniques
The favicon-stored website and CSSQuake game illustrate how heavy computational tasks can be squeezed into limited client resources. This trend points toward embedding small AI models (e.g., on-device LLMs) into favicons, service workers, or CSS, enabling offline or low-bandwidth inference. Efficiency gains in model quantization and web assembly will drive this further.
Age-gating and VPN bans create new adversarial challenges for AI moderation
UK plans to age-gate VPNs for social media bans will force AI-based age verification systems to handle evasion tactics (proxy detection, synthetic identities). Conversely, generative AI can produce realistic fake age documents, raising the arms race. ML researchers should develop robust evasion-resistant classifiers and advocate for transparent, privacy-preserving verification methods.
Niche, retro-computing reverse engineering projects rely on community-driven AI assistance
The F-15 Strike Eagle II project showcases how community volunteers test reconstructed code. AI tools like Copilot or GPT-based decompilers are already accelerating such efforts. Expect a rise in AI-assisted retro-gaming preservation, where models translate assembly to high-level code and generate test cases, lowering the barrier for hobbyists.
Color science insights inform AI for visual content generation
The article on screen-limited colors highlights the gap between digital reproduction and human perception. For AI image/video generation models (e.g., diffusion models), training on sRGB-only data leads to a loss of vivid cyans and other out-of-gamut hues. Incorporating wider color spaces (e.g., BT.2020) or spectral rendering in training pipelines could produce more photorealistic outputs, especially for AR/VR and HDR displays.
Self-contained, minimal-dependency tools (like Bootimus) reflect a trend in AI infrastructure
Bootimus’ single-binary design for PXE booting mirrors the rise of lightweight, containerized AI deployment (e.g., ONNX Runtime, tinyML). As edge AI grows, tools that require zero configuration (proxyDHCP, automatic OS detection) become critical. This suggests that AI server management will increasingly favor self-contained binaries over complex stacks, reducing operational overhead for distributed inference nodes.
Analysis generated by deepseek-reasoner