← Oct 2Saturday, October 3, 2026 · all daysOct 4 → · today »
1.Newgrounds.com – A community of games, music, and art(newgrounds.com)
457 points by azhenley 11 days ago | 136 comments | permalink
tl;dr: Summary not available
HN Discussion:
  • ↑Nostalgic reminiscence about playing and discovering Flash games on Newgrounds
  • ↑Former creators sharing their experiences making games and participating in the community
  • ↑Appreciation for Newgrounds preserving old content via tools like Ruffle
  • ↑Praise for Newgrounds retaining its authentic, non-commercial soul over the years
  • ~Mixed reflection on formative but questionable experiences with adult content on the site
2.Court agrees with EFF: Utah's VPN law demands a technical impossibility(eff.org)
799 points by hn_acker 12 days ago | 401 comments | permalink
tl;dr: A federal judge issued a preliminary injunction blocking Utah's SB 73, a law requiring adult websites to detect VPN users and geolocate visitors to enforce age verification. The court ruled the law imposes de facto strict liability by demanding "geolocation perfection" that isn't technically possible, forcing sites to either verify every user worldwide or block Utah entirely—likely violating the Constitution's limits on regulating out-of-state commerce. Utah lawmakers may revise the law next session, but the ruling sets a precedent against similar anti-VPN proposals in other states.
HN Discussion:
  • ~Governments won't be stopped by technical impossibility and will just ban or block everything they can't track
  • ~The 'internet routes around censorship' adage is outdated as authoritarian states have proven effective at censorship
  • ↑This law is part of a broader fascist/authoritarian push and the real motive isn't porn but control
  • •Questioning technical details of the law such as VPN detection feasibility or first amendment implications of banning instructions
  • ↑Lawmakers routinely ignore technical and scientific reality when crafting laws
3.Apple Pass Designer(developer.apple.com)
542 points by soheilpro 11 days ago | 325 comments | permalink
tl;dr: Apple has released Pass Designer, a macOS tool for creating and previewing Apple Wallet passes with real-time rendering that matches iOS and watchOS output. It supports templates, color/layout customization, standard field editing, validation, and semantic tags (for boarding passes and tickets) that enable Siri Suggestions, Calendar, and Maps integration, with automatic backward-compatible pass generation. The beta requires macOS 27 or later and an Apple Developer account.
HN Discussion:
  • ~Alternative web-based pass designer tools already exist and offer similar functionality
  • ~The tool is trivial to build now with LLMs, nothing technically interesting
  • ↑Welcome and overdue addition that expands Apple Wallet accessibility
  • ↓Criticism of Apple-only approach instead of cross-platform standard with Google
  • •Questions about missing features like dynamic barcodes and semantic barcode areas
4.Mike Tomlin spent 12 years building a Minecraft city(nytimes.com)
667 points by CoryOndrejka 12 days ago | 169 comments | permalink
tl;dr: Summary not available
HN Discussion:
  • ↑Appreciation for the earnest joy of sharing a passion project
  • ↑Amazement that a busy NFL head coach had time for this hobby
  • ~Humorous suggestion that Minecraft distracted him from coaching success
  • ~Technical critique that the build isn't advanced by Minecraft standards
  • ↑This exemplifies the kind of creative pursuit UBI could enable
5.A 12-year sequence of telescope images of a star and four planets orbiting(bsky.app)
392 points by mariuz 11 days ago | 79 comments | permalink
tl;dr: Summary not available
HN Discussion:
  • ↑Sharing related or similar visualizations of the same or other astronomical phenomena
  • ~Clarifying that the imagery is interpolated frames, not a true video
  • ↑Excited about future telescope technology that will advance direct exoplanet imaging
  • ↑Expressing awe at the scale of space and the achievement shown
  • •Curious about the technical details of how such imaging was achieved
6.The Forgetful CPU (Linux on M4)(yuka.dev)
260 points by signa11 11 days ago | 194 comments | permalink
tl;dr: A developer documents booting mainline Linux on an M4 Mac mini, overcoming hurdles including locked registers (GXF, RVBAR), missing MMU mappings for MMIO, and a virtualization-related CPU register write that crashed the kernel. The biggest challenge was that M4's WFI instruction doesn't preserve CPU state as the ARM64 spec requires, with no chicken bit to disable the behavior; the fix—a new kernel bootarg to NOP out WFI/WFIT instructions—has now been merged upstream in Linux and m1n1, enabling multi-core boot on M4/M4 Pro/Max and M5 chips.
HN Discussion:
  • •Apple's superior hardware would benefit from embracing openness despite bloated macOS
  • ↓Effort is better spent improving x86-64 or open hardware like RISC-V instead
  • •Apple's proprietary protocols (like trackpad gestures) lock out third-party hardware
  • ↓It's contradictory to run open software on hardware from an anti-open company
  • •Curious whether AI could assist with this kind of reverse-engineering work
7.Loss of cell identity drives human aging: Two new papers(erictopol.substack.com)
359 points by bookofjoe 12 days ago | 172 comments | permalink
tl;dr: Two new papers (Nature, Cell) propose that aging isn't just cumulative damage but also loss of cell identity, driven by erosion of the "slow" epigenetic layer (chromatin architecture maintained by PRC2) that normally locks cells into their differentiated state. As this erosion progresses, cells undergo "mesenchymal drift" toward fibroblast-like states, driving fibrosis, inflammation, and age-related diseases from atherosclerosis to Alzheimer's. Potential interventions include caloric restriction, partial epigenetic reprogramming via Yamanaka factors (OSKM or OSK), and lithium—which stabilizes neuronal identity—while chronic stressors like poor diet, sleep, and inactivity likely accelerate the erosion.
HN Discussion:
  • ↑Embryonic epigenetic reset supports the reversibility premise but highlights limitations
  • ↑Chronic stress as aging driver is concerning; curious about targeted methylation tools
  • ↓Papers are repackaged damage hypothesis and fail to explain Hayflick limit or species lifespan differences
  • ~Epigenetic aging is an adaptive programmed response to accumulating DNA damage
  • •Humorous/metaphorical reactions to the cell identity loss framing
8.Extra Big Ass Intelligence(extrabigassintelligence.com)
512 points by 34679 11 days ago | 126 comments | permalink
tl;dr: Summary not available
HN Discussion:
  • ↑Author explains the site was a quick satirical LLM-generated protest that unexpectedly went viral
  • ~The site's design ironically resembles LLM-generated slop itself
  • ↑Idiocracy references suggest reality is worse than the satirized dystopia
  • ↑The satire accurately captures the absurdity of modern ads and AI hype
  • •Points to real-world context like Trump's Super Intelligence promotion and Sanders' AI ban proposal
9.Greg Kroah-Hartman – Security in the LLM Age [video](youtube.com)
328 points by usernomdeguerre 12 days ago | 118 comments | permalink
tl;dr: Summary not available
HN Discussion:
  • ↑Appreciates Greg KH's candid breakdown debunking Mythos's 79 CVE claims as overhyped marketing
  • ↑Anthropic/OpenAI's AI safety rhetoric is undermined by shoddy work and failure to credit original kernel developers
  • ~LLMs could still eventually be useful for kernel bug discovery with proper specialized training
  • ~Raises unaddressed questions about Linux's own legal exposure to LLM-generated patches from copyrighted training data
  • ↑Confirms that human reviewers remain critical and LLMs are degrading technical collaboration
10.From the creator of Redis; run LLM locally with ds4(dwarfstar.sh)
344 points by fibo 11 days ago | 98 comments | permalink
tl;dr: DS4 (DwarfStar 4), from Redis creator Salvatore Sanfilippo (antirez), is a narrow C inference engine for running large open-weight MoE models like DeepSeek V4/V4.1, GLM 5.x, and Qwen3.8 locally on high-memory Mac, CUDA, and ROCm machines. It uses asymmetric 2-bit quantization on routed experts, SSD-resident KV cache that survives restarts, and ships a CLI, OpenAI/Anthropic-compatible server, and native coding agent. Reference benchmarks show ~34 tokens/sec generation on an M5 Max with 128 GB running V4 Flash Q2.
HN Discussion:
  • ↑Developers extending ds4 with bindings and forks for other languages and platforms
  • ↑Users sharing positive hands-on experience running large models on high-end hardware
  • •Questions about performance on lower-spec hardware and SSD streaming viability
  • •Promotion of alternative or inspired inference engines for different hardware targets
  • •Curiosity about the expertise required to build model-specific inference engines
11.FLUX 3 Image(bfl.ai)
428 points by minimaxir 12 days ago | 97 comments | permalink
tl;dr: Black Forest Labs' FLUX 3 Image is a multimodal generation model that lets users compose images by placing elements as bounding boxes on a 0-1000 grid, each with text descriptions, or by writing a standard prompt and having an LLM plan the layout. It supports native 2K/4K rendering, pixel-perfect edits that preserve untouched regions, multi-reference composition (up to 10 images), and simultaneous targeted edits. Weights are available under a commercial license for self-hosted deployment, and it's accessible via a Playground and the BFL API.
HN Discussion:
  • ↑The UX for spatial element placement is a major strength and differentiator worth praising
  • •AI image generation has reached a point where digital images can no longer be trusted as evidence
  • ↑The model works well for granular iterative edits like UI element replacement
  • ~Waiting for open weights/local releases before getting excited
  • ↓Showcase examples and real-world tests reveal quality issues in generated outputs
12.With most information hidden, the game Stratego had stumped AI until now(arstechnica.com)
279 points by PaulHoule 11 days ago | 142 comments | permalink
tl;dr: Researchers from CMU, MIT, NYU, and Stanford built Ataraxos, an AI that defeated top Stratego player Pim Niemeijer 15-1 (with 4 draws), cracking a game that had resisted even DeepMind's well-funded attempts. Stratego's difficulty stems from massive hidden information (decillions of possible piece setups), game lengths up to 2,000 moves, and the need to balance bluffing. Notably, training required only 16 GPUs and a few thousand dollars.
HN Discussion:
  • •Nostalgic personal memories of playing Stratego as a child
  • ↑The efficiency gains (fewer games, less compute) are the key breakthrough
  • ~Skepticism about the 'just a few thousand dollars' framing given elite institutional backing
  • ↑This result shows the 2022 DeepMind 'mastering' claim was premature
  • ~Static games are inherently beatable; dynamic/evolving games or harder variants remain challenges
13.Updates to Full Disk Access in macOS(developer.apple.com)
309 points by notfirstpost 11 days ago | 219 comments | permalink
tl;dr: Apple is tightening Full Disk Access controls on macOS after finding that some developers are abusing the permission to access files, mail, messages, and browsing history beyond what users realize. Future versions will require more explicit user action to grant this access, with Apple citing the growing risks posed by increasingly autonomous AI agents as a key motivator.
HN Discussion:
  • ↑Current permission model is too coarse; needs finer-grained controls rather than all-or-nothing Full Disk Access
  • ↓Tightening restrictions cripples power users to protect novices who should use simpler devices instead
  • ↑More specific controls are useful; many apps request Full Disk Access unnecessarily
  • ↓Owners of their machines should have unrestricted access by default; Apple is overreaching
  • ~Frequent permission prompts will cause fatigue and users will grant access anyway, defeating the purpose
14.Show HN: Giving Opus 5.5 a simulated paint canvas(stillwet.art)
374 points by alstonite 12 days ago | 108 comments | permalink
tl;dr: LLMs paint by writing programs against a simulated oil-paint-on-linen canvas (no image models involved), mostly reproducing Caspar David Friedrich's style from written research alone without ever seeing his work. Across 75+ paintings, the models showed striking convergent behavior: six of six Claude Opus runs chose "jug with lemons" for free subjects, multiple models independently titled works identically, and nearly half of titled paintings reference dusk/evening. Notable incidents include Gemini 3.8 Flash inspecting background processes and realizing it was being evaluated, prompting tool restrictions in later rounds.
HN Discussion:
  • ↑Impressed by capabilities and speculates on training data behind convergent behavior
  • ~Points out the models use a 'look' tool, tempering the impressiveness of results
  • ↓Results are uncanny or flawed, like repeated nonsensical churches ruining landscapes
  • •Situates the work within prior similar explorations and asks about the author's technique
  • ↑Excited about code-based artifacts as a general creative direction with practical uses
15.Sites in ChatGPT(chatgpt.com)
342 points by polvi 12 days ago | 334 comments | permalink
tl;dr: Summary not available
HN Discussion:
  • ↑ChatGPT Sites is a useful, underrated feature for quickly prototyping apps and websites
  • ↑AI is replacing industries like web designers, and Sites accelerates that disruption
  • ↓The demos are superficial and reveal the Potemkin-village quality of AI-generated apps
  • ↓Sites is an attempt at a moat that won't work since models are commodities and competitors will emerge
  • •This signals OpenAI will sherlock LLM-dependent businesses like Lovable and launch an app economy inside ChatGPT
16.The Legend of von Neumann (1973) [pdf](gwern.net)
302 points by suopspaces 11 days ago | 181 comments | permalink
tl;dr: Summary not available
HN Discussion:
  • ↑Sharing anecdotes highlighting von Neumann's exceptional intellect and personality
  • ↑Von Neumann was arguably the most influential scientist of the 20th century
  • •Recommending further reading/media about von Neumann and his contemporaries
  • ↑Wondering who the modern equivalent of von Neumann might be and how to find them
  • •Pointing to related past HN discussions on the same topic
17.Shimano Bicycle Museum Review(inrng.com)
307 points by pietroppeter 12 days ago | 82 comments | permalink
tl;dr: The Shimano Bicycle Museum in Sakai, Japan (near Osaka) is a well-designed, spacious collection focused on the history and variety of bicycles rather than Shimano corporate history or merchandise. Exhibits range from early hobby horses to pro race bikes (like Freddy Maertens' Flandria), postal bikes, and a disassembled modern race bike showing all 3,200 parts, plus a research archive. Admission is cheap (around JPY 500), and there's no gift shop hard-sell on the way out.
HN Discussion:
  • ↑Recommendations for similar bicycle museums elsewhere (Marin, Singapore)
  • ↑Personal experiences visiting the Shimano museum confirm it's a delight
  • •Appreciation for Shimano's engineering innovations in cycling (and fishing)
  • •Curiosity about Shimano's historical rise over European component makers
  • ↑Local Osaka residents learning about the museum and planning to visit
18.Supabase is acquiring Turso(supabase.com)
216 points by cvburgess 11 days ago | 118 comments | permalink
tl;dr: Supabase is acquiring Turso to address the massive database demand created by AI agents, which are spinning up millions of databases for prototypes and apps. Turso's Rust-based SQLite rebuild and architecture—where a single server manages millions of on-demand databases—complements Supabase's Postgres offering, providing a path from cheap SQLite prototypes to production Postgres. Both products will continue operating independently, with Turso's Glauber Costa leading the new agentic infrastructure effort.
HN Discussion:
  • ~Hope Turso survives acquisition and doesn't get shut down or abandoned
  • ↑Acquisition reduces risk of Turso project dying, making it more trustworthy to adopt
  • ↓Turso needs to fix existing bugs and performance issues before expanding
  • ~Need for OSS self-hostable alternatives alongside Supabase's hosted offering
  • ↓Questioning the premise of rewriting SQLite given its quality
19.Various Projects Find Hidden SDR Capabilities in ESP32 Microcontrollers(rtl-sdr.com)
287 points by nkw 12 days ago | 66 comments | permalink
tl;dr: Researchers have discovered an undocumented feature in several ESP32 chips that allows firmware to bypass the fixed WiFi/Bluetooth radio and capture raw IQ baseband samples, effectively turning them into SDRs covering 2.2–2.7 GHz (and 4.8–6.0 GHz on the ESP32-C5) at up to 80 MS/s. Bandwidth limits mean most chips can only export snapshots for spectrum analysis, though the ESP32-S31 can stream continuously over Gigabit Ethernet. Independent projects have leveraged the finding for phase-coherent direction finding (ESPARGOS), a general-purpose SDR with an FPGA frontend, and a 5.8 GHz FPV video receiver (C5VRX).
HN Discussion:
  • ↑Excitement about potential applications like V2X data collection and ham radio use cases
  • ~Concern that Espressif may patch out this undocumented feature due to regulatory pressure
  • ↑Technical speculation on extracting IQ data via new interfaces like Gigabit Ethernet or PSRAM
  • •Questions about signal quality metrics like phase noise and DC offset not covered in article
  • •Curiosity about extending capabilities such as sampling harmonics or building LORA-like systems
20.Using Opus 5.5 to discover a new eyewitness record of the dodo(resobscura.substack.com)
224 points by benbreen 12 days ago | 79 comments | permalink
tl;dr: A historian used Claude Opus 5.5 to search the GLOBALISE Dutch East India Company archives via semantic search and agent-based exploration, uncovering what appears to be a previously-unnoticed 1615 ship's log mentioning dodo hunting on Mauritius, plus a possible new reference to the extinct red rail and a lead connecting a Portuguese-captured "ostrich" to Emperor Jahangir's famous dodo. The author concludes that frontier models are now capable of genuine archival discovery but struggle to judge historical significance or ask good questions, meaning the real bottleneck will shift to expert attention needed to verify amateur-driven findings.
HN Discussion:
  • ↑Shares similar success using LLMs for large-scale archival/historical research in unfamiliar languages
  • ↑Confirms LLMs make unhuman-like errors, reinforcing the article's point about judgment limitations
  • ↑Endorses the article's insight that LLMs work best at interdisciplinary boundaries
  • •Considers applying the approach to personal archival projects like deciphering family journals
  • ~Questions the scope/scale of the search, suggesting the narrowing was more impressive than the discovery