JetZero’s homepage is pitching the Z4, a 250-seat, international-range, all-wing commercial aircraft that it says could use up to 50% less fuel than today’s jets. The site also highlights recent momentum: airline interest or investment from JAL, Gulf Air, Alaska Airlines, and United, a roughly $175 million Series B, a Greensboro manufacturing campus, and a possible EXIM Bank financing package. The HN discussion treated the fuel-efficiency claim as compelling, but not enough by itself to answer the hard question: whether this can become a certifiable, passenger-friendly, airport-compatible airliner.
Discussion: Mixed — Commenters were intrigued by the promise of a radically more efficient airliner, but the dominant mood was skeptical and engineering-focused. The thread centered less on the company’s fundraising and more on whether a blended-wing commercial aircraft can clear practical barriers like pressurization, passenger windows, motion sickness, airport gates, evacuation rules, and certification. (Interest in lower-fuel aviation and emissions reductions, Skepticism about blended-wing commercial viability, Passenger-experience worries around no windows and motion sickness)
0:00 / 3:15hardwaredeep dive GM bets on sodium-ion batteries for the grid#
IEEE Spectrum reports that Colorado startup Peak Energy is partnering with General Motors to develop sodium-ion batteries for grid-scale storage, with a planned $71 million factory near Sacramento slated to come online in 2027. Peak says its lower-energy-density cells can still beat lithium-iron-phosphate systems on lifetime cost because they tolerate higher temperatures, avoid active cooling, and are designed for roughly 20 years and 20,000 cycles. The catch is that sodium-ion remains a tiny share of U.S. deployments, and Peak is currently buying commercial cells from Chinese suppliers while China’s CATL is already moving at much larger scale.
Discussion: Mixed — HN was interested in sodium-ion’s fit for stationary storage, especially lower cooling needs, long cycle life, and the possibility of cheaper grid batteries. But commenters were cautious about manufacturing reality, dependence on Chinese cells and processing, inverter and voltage-range issues for home use, and whether U.S. industrial capacity is being overstated. (Sodium-ion looks better suited to grid storage than EVs because energy density matters less, Passive cooling and reduced HVAC load were seen as potentially important cost advantages, Skepticism that Peak is currently more assembler than cell maker because it buys commercial cells from Chinese suppliers)
0:00 / 2:56policydeep dive Debian debates whether LLM-assisted work belongs in the distro#
Debian is considering a General Resolution on LLM usage, with competing proposals ranging from prohibiting direct Debian contributions written with generative AI assistance, to discouraging LLMs while requiring disclosure and human-only messages, to accepting AI-assisted Debian-specific work under responsibility, disclosure, DFSG, and data-sensitivity rules. The debate matters because Debian’s trust model depends on clear licensing, maintainability, reviewer capacity, and security practices, and the project is trying to decide how those values apply to AI-assisted development.
Discussion: Mixed — HN readers largely treated this as an unresolved Debian governance fight, not a final ban. The discussion split between people who see LLM use as a legal, quality, and community risk, and others who think broad bans are unenforceable or could slow useful security and maintenance work. (Clarifying that this is a set of proposals to be debated and voted on, not a final Debian decision, Arguments over whether LLM assistance should be distinguished from submitting LLM-generated output)
0:00 / 0:02ai A tiny LLM runs on an eight-dollar microcontroller#
A developer says they ran a 28.9-million-parameter language model directly on an ESP32-S3 microcontroller that costs about $8, generating text at roughly 9.5 tokens per second with no server connection. The trick is storing most of the model — a 25-million-parameter lookup table — in flash memory and only reading the small pieces needed per token, using a Per-Layer Embeddings idea from Google's Gemma models. The model is trained on TinyStories, so it writes simple short stories rather than answering questions or following instructions; the point is the memory architecture, not chatbot capability.
Discussion: Mixed — HN was mostly impressed by the engineering stunt and the learning value, with several commenters calling out the surprising speed and the appeal of offline, tiny-device AI. The skeptical thread argued that running LLMs on microcontrollers is too constrained to be useful, but replies pushed back with examples from embedded ML and emphasized experimentation over immediate practicality. (Admiration for resource-constrained engineering, Debate over practicality versus fun and learning, Interest in offline TTS)
0:00 / 1:33general A one-person map for 18,000-plus U.S. golf courses#
A Show HN project called Golf Course Browser presents a free, no-login directory and map for U.S. golf courses, using OpenStreetMap as a base layer and adding filters for access, holes, course type, amenities, nearby courses, scorecards, weather, and user corrections. The creator says the goal is a more complete and current course directory, with expansion beyond the U.S. planned after the pipeline is refined. The HN discussion doubled as live QA: users found missing courses, bad links, and classification issues while also debating the legal and environmental implications of cataloging golf courses.
Discussion: Mixed — HN was broadly appreciative and constructive, with many users suggesting features, reporting missing or stale course data, and asking about international expansion. The thread also veered into concerns about data licensing, liability, and whether golf courses are a poor use of land and water, so the mood was not purely celebratory. (Constructive bug reports and missing-course reports, Requests for polygons, mini golf)