← Sep 30Thursday, October 1, 2026 · all daysOct 2 → · today »
1.Gemini 4 Argon(blog.google)
1684 points by bradleyg223 13 days ago | 1173 comments | permalink
tl;dr: Google announced Gemini 4 Argon, a frontier model with a 1M output token limit, priced at $2/M input and $10/M output tokens, initially rolling out to trusted cyber defenders before wider release. It claims state-of-the-art results on benchmarks like DeepSWE v1.1 (77.9%), LVBench (91.7%), and CWE-bench v1 (68%), and is already being used internally at Google for tasks like C++ to Rust migrations and data center memory optimization. Google is also releasing it without cyber guardrails to trusted defenders and emphasizing new safeguards against misuse, prompt injection, and misalignment.
HN Discussion:
  • ↑Personal anecdote showcasing impressive Gemini capabilities reinforces the model's power claims
  • •The leapfrogging competition disproves winner-takes-all AI theories, suggesting a distributed market
  • ↓Frustration that the model isn't actually available to paying subscribers
  • ↑The C++ to Rust migration at scale is the truly significant announcement here
  • ~Google should open-source Gemini weights to undermine competitors
2.The top secret URSALA, RAQUEL, and FARRAH satellites (2025)(thespacereview.com)
299 points by Bluestein 13 days ago | 158 comments | permalink
tl;dr: Recently declassified documents reveal details about a 40-year NRO program of small "hitchhiker" ELINT satellites (Program 989/Mission 7300), deployed off larger HEXAGON reconnaissance satellites and later launched independently, with names like URSALA, RAQUEL, FARRAH, GLORIA, and CARRIE. These spin-stabilized satellites collected radar and signals intelligence from low Earth orbit, evolving from strategic collection for national-level customers to directly supplying tactical military forces via mobile ground vans—pioneering the TENCAP (Tactical Exploitation of National Capabilities) concept. Some FARRAH satellites launched in the late 1980s and early 1990s may still be operational in orbit today.
HN Discussion:
  • ↑Awe at the US military's advanced satellite capabilities compared to civilian science programs
  • •Notes the NRO declassified archive as the likely source material for the article
  • ↑Curiosity about what current satellite programs will be declassified decades from now
  • •Speculation that the military has similarly advanced secret AI capabilities
  • ↓Technical correction about TALENT and KEYHOLE being distinct programs in the 1960s
3.Why the Bronze Age Collapsed(worksinprogress.news)
394 points by AnodicElegy 14 days ago | 290 comments | permalink
tl;dr: The Bronze Age collapse around 1200 BC wasn't just caused by droughts, invasions, or famines—those alone don't explain why empires failed to reconsolidate for centuries afterward. The key factor was iron: unlike bronze (which required rare tin transported across thousands of miles, giving elites a monopoly on weapons), iron ore was abundant and could be smelted locally, democratizing access to weaponry and making large empires harder to sustain. The author draws parallels to modern shifts in military technology like cheap drones and AI, which similarly redistribute—or concentrate—the power to wage war.
HN Discussion:
  • ↓Iron adoption was gradual, not sudden, undermining the causal claim about the collapse
  • ↓The article's historical examples (like Spanish defeating Aztecs via iron) are factually wrong
  • ~Palace economy structure and multiple stressors better explain the collapse than iron alone
  • ↓Counterexamples like Harappan civilization contradict the bronze-equals-stratification thesis
  • ~If iron prevents big empires, how did later Athenian and Roman empires form?
4.A brief history of the Bloomberg terminal(spectrum.ieee.org)
378 points by rbanffy 13 days ago | 168 comments | permalink
tl;dr: Michael Bloomberg launched the Market Master terminal in 1982 through Innovative Market Systems, initially selling to Merrill Lynch, by integrating real-time financial data with quantitative analysis tools on a dedicated terminal with a custom color-coded "Chiclet" keyboard and private network. The product expanded over decades to include news, chat, multimedia, and biometric login, and in 2000 the company dropped dedicated hardware in favor of software running on customer PCs. Subscriptions now run upwards of $32,000/year, and retired keyboards—including one from "Bond King" Bill Gross with his password taped on it—now reside in the Smithsonian.
HN Discussion:
  • ↑Admiration for the terminal's information density and UI design
  • •Technical additions about the terminal's architecture and backwards compatibility
  • ↓Criticism of Bloomberg's restrictive hardware login requirements driving users away
  • ~Requests for more content (screenshots, fonts) missing from the article
  • •Sharing related resources about competitors and other trading terminals
5.The last time my family was replaced by technology(manuel.darcemont.fr)
323 points by megalomanu 13 days ago | 625 comments | permalink
tl;dr: The author's great-great-grandfather, a farrier in rural France, pivoted to become a mechanic when cars arrived, preserving his underlying purpose—helping people get around—even as the technology changed. Drawing a parallel to developers anxious about AI, the author argues that what truly motivates most engineers is building things and solving problems, not writing code itself. The lesson: let go of the "how" to hold on to the "why."
HN Discussion:
  • ↑Author clarifies the piece is a personal tribute, not a dismissive lesson about adapting
  • ↓The analogy ignores real human destruction and immediate suffering of displaced workers who can't easily retrain
  • ↓The horse-to-car analogy fails because there's no next frontier for displaced knowledge workers to move into
  • ↑Agrees that code is just a tool and problem-solving is the real purpose, so embracing AI makes sense
  • ↓Philosophical reframing is a luxury for the employed; it doesn't help those already replaced
6.EDG C++ front-end goes public(edgcpp.org)
241 points by iandinwoodie 13 days ago | 125 comments | permalink
tl;dr: EDG's C++ front end—a 30-year-old proprietary source-to-source compiler engine used across the industry—went open source on September 30, 2026, with The C++ Alliance (also fiscal sponsor of Boost) taking over as its nonprofit steward. Development will proceed on three tracks: community pull requests, Alliance-funded maintenance (published immediately), and collectively funded larger features, all landing in the same public repo at github.com/edgcpp/compiler with no early access. John Spicer continues as the point of contact, chairing the Fiscal Sponsorship Committee that oversees contributions and priorities.
HN Discussion:
  • •Context that EDG the company is winding down, explaining the open-sourcing motivation
  • ↑Celebrating the historical and industry significance of EDG's frontend for C++
  • ↑Appreciation for the technical strengths like compiler emulation and preserved commit history
  • ↑Speculation about new uses enabled by open sourcing, such as source-to-source transpilation
  • ~Mixed personal experience with EDG, preferring Clang due to tech debt and cost
7.Singapore govt dating app uses Gale-Shapley stable marriage algorithm(twitter.com)
446 points by rzk 13 days ago | 463 comments | permalink
tl;dr: Singapore's government has launched FirstDate, a dating app for public servants aged 21-35 that uses the Gale-Shapley stable marriage algorithm to produce mathematically stable matches where no two users would mutually prefer each other over their assigned pairs. The app enforces one match per cycle with a 72-hour decision window, Singpass identity verification, and reveals contact info only on mutual acceptance. Unlike commercial apps optimized for engagement, FirstDate is designed to get users off the platform.
HN Discussion:
  • ↓People don't actually know their true preferences, undermining the algorithm's premise
  • ↓Technically it can't be real Gale-Shapley since full ranking of thousands is impractical
  • ↑Government-run app has better incentives than commercial apps that profit from engagement
  • ↓Singapore government is overly paternalistic in orchestrating citizens' lives
  • •Technical curiosity about implementation details like gender initiation or related algorithms
8.You said no MCP(earendil.com)
670 points by yarapavan 13 days ago | 360 comments | permalink
tl;dr: Pi reversed its stance against MCP and brought it into the core, driven by MCP's maturation and the realization that supporting it well required infrastructure (like deferred tool loading and metadata) that also benefits other features. The integration pairs MCP with "Codemode," a JavaScript/WASM sandbox running on the harness side that lets the agent compose tool calls programmatically—treating MCP more like OpenAPI with structured data and discoverable tools, rather than dumping everything into context.
HN Discussion:
  • ↑MCP has broad practical value beyond coding, validating its adoption
  • ↑Praise for the team's willingness to publicly reverse a strongly-held position
  • ↑MCP's reversal was obvious; critics were wrong to dismiss it earlier
  • ↑MCP is imperfect but worth adopting due to ecosystem breadth, like USB-C
  • ↓Skepticism that MCP offers anything meaningful over OpenAPI or CLI/shell tools
9.Solving Factorio Quality(exyr.org)
274 points by laurenth 15 days ago | 107 comments | permalink
tl;dr: A Factorio player models the game's Quality upcycling loops using matrix math, representing item throughput as vectors and crafting/recycling/filtering steps as transition matrices, then solves the resulting equilibrium equations (Ax=b) via Gaussian elimination for exact rational results rather than iterative approximations. The approach handles "washing," crafting+recycling upcycles, and 15-dimensional asteroid reprocessing. The author implemented it as an interactive web calculator (factoqual.grebedoc.dev) using TypeScript, VanJS, and a rational-number library after initially prototyping in Rust.
HN Discussion:
  • ↑Fellow enthusiasts share their own similar solver/analysis projects for Factorio quality
  • ↑Confirms the article's quantitative findings about washing vs upcycling and module effects
  • ~Suggests alternative/better math approach using transfer matrix diagonalization instead of Gaussian elimination
  • ~Notes the article is outdated as Space Casino is removed in Factorio 2.1
  • •Offers minor presentation improvements like LaTeX \text{} formatting
10.Phyllotaxis: An audio-reactive LED display(jagi.studio)
349 points by evakhoury 15 days ago | 50 comments | permalink
tl;dr: A developer built a sunflower-inspired LED display by generating a phyllotaxis pattern in code, Voronoi-tessellating it, and 3D printing cell walls to hold 89 addressable LEDs behind translucent mulberry paper. An STM32 (later ESP32) drives shader-style animations and runs FFT-based audio analysis via an I2S mic for real-time reactivity. A second version replaced the 3D-printed backplate with modular PCBs in 5-fold symmetry, now installed at Recurse Center with a web interface for uploading custom sketches.
HN Discussion:
  • ↑Appreciation for the 5-fold PCB symmetry and hand-soldering techniques for SMT components
  • ↑Practical suggestion to outsource LED assembly to the PCB fab to save time and avoid damage
  • •Noting similar existing commercial or hobby products as convergent examples
  • ↑Interest in replicating the build and requests for licensing, hardware details, and diffuser attachment method
  • ↑Nostalgic appreciation for audio visualization, favoring hardware over software versions
11.Livenerf: Has Opus 5.5 been nerfed yet?(github.com)
911 points by bryan0 14 days ago | 388 comments | permalink
tl;dr: Livenerf is an append-only benchmark that tests whether Claude Opus 5.5 degrades after launch, using a frozen panel of 78 "sometimes right" questions from GPQA Diamond, MMLU-Pro, and competition math run daily through a pinned Claude Code CLI. It uses pre-registered statistics (paired deltas with clustered SEs), a control arm running Opus 5, and a decision rule requiring a 3+ point change clearing 99% intervals in two consecutive 10-day windows. Validation shows it can detect lower reasoning effort but can't distinguish a same-family model swap in one window's samples.
HN Discussion:
  • ~Similar benchmarks already exist and nerfs are often honeymoon-effect perception rather than reality
  • ↑Nerfs are definitely real due to constant infrastructure changes causing regressions
  • ↑Personal experience strongly confirms models get nerfed after launch hype dies down
  • ↑Appreciates the tracking effort to settle recurring debates with actual data
  • ↓Nerfing claims aren't real; users just hit model complexity ceilings over time
12.I could've accessed 17T Microsoft records(blog.faav.net)
314 points by luispa 15 days ago | 127 comments | permalink
tl;dr: A 16-year-old researcher found that Microsoft's internal "Titan" analytics API validated JWT claims (tenant, audience, app ID) but never checked token signatures, allowing an unsigned token with `upn: admin` to execute SQL as administrator. This exposed roughly 17.3 trillion rows across 17 connected databases, including employee records and Bing search analytics. Microsoft patched it within four days of the report, paid a $5,000 bounty, and exercised editorial control over the published writeup.
HN Discussion:
  • ↑Microsoft's editorial control over the disclosure is unethical and shameful
  • ↑The $5,000 bounty is insultingly low given the severity of the vulnerability
  • ↑Admiration for a 16-year-old finding such a critical bug and writing it up well
  • ↑JWT's 'alg:none' option is a recurring, well-known security flaw
  • •Internal context: Microsoft has tools like MISE that would prevent this, but teams defer compliance
13.Dots: Always-on agents(openai.com)
763 points by alvis 14 days ago | 639 comments | permalink
tl;dr: Summary not available
HN Discussion:
  • ↑Positive user experience with always-on agents enabling domain expertise and collaboration
  • ↓Confusion about what Dots actually is and how it differs from existing products
  • ↓Skepticism about the practical utility of overnight/autonomous agents given human-in-the-loop needs
  • ↓Trust and safety concerns about giving autonomous agents access to sensitive accounts
  • •Technical observation about MITM certificates and infrastructure concerns
14.America.gov(america.gov)
769 points by plesiv 14 days ago | 732 comments | permalink
tl;dr: Summary not available
HN Discussion:
  • ↑Centralizing access to government services is a genuinely useful idea that helps citizens
  • ↑Chatbots powered by LLMs like Gemini are well-suited for navigating complex government bureaucracy
  • ↓The site shows poor design and quality control, appearing hastily built with broken elements
  • ↓Adding another government portal just increases fragmentation rather than solving it
  • ~Cautiously optimistic based on the team's prior successful work on similar projects
15.Vermont replacing power plants with home batteries(bbc.com)
384 points by devonnull 14 days ago | 300 comments | permalink
tl;dr: Vermont's largest utility, Green Mountain Power, leases two home batteries to customers for $55/month, aggregating 5,500+ installations into a virtual power plant that's now the state's biggest power source and has replaced most of its fossil-fuel peaker plants. The system saved customers $11M last year and provides outage protection, with GMP aiming to offer storage to all 300,000 customers and eventually tap EV batteries via vehicle-to-grid. Nationally, virtual power plants could save $10B+ annually by 2030, though utility incentive structures remain a barrier to broader adoption.
HN Discussion:
  • ↑Similar virtual power plant programs exist elsewhere and work well for participants
  • ↓Utilities exploit customers by charging for batteries while using them as distributed infrastructure
  • ~Technical concerns about battery efficiency losses and whether the economics truly scale
  • ~Regulatory and building code barriers make adoption impractical for many homeowners
  • ↑Home and grid-scale battery storage is a promising idea worth expanding
16.NASA asked several former SR-71A staffers to help secret restart(aviationweek.com)
320 points by ilamont 14 days ago | 394 comments | permalink
tl;dr: NASA has quietly removed SR-71A Tail No. 844 from its display pole at Armstrong Flight Research Center and is attempting to return the Blackbird to flight after a 27-year hiatus, recruiting retired test engineers for the secret project. The effort faces enormous hurdles: the $600M spare parts inventory was destroyed in 2007, specialized JP-7 fuel and ground equipment are gone, and the engines have sat unlubricated for decades. NASA Administrator Jared Isaacman teased the project at a September summit as part of a broader push to rebuild the agency's X-plane fleet.
HN Discussion:
  • ↑Supports the restart for research value and preserving engineering expertise
  • •Questions whether destroyed parts and tools might actually be recoverable
  • ↑Skeptical due to practical hurdles like missing JP-7 fuel and tankers
  • ↓Believes classified successor aircraft already exist, making SR-71 revival pointless
  • ↓Views the project as a political vanity effort rather than genuine engineering need
17.Fuck Android Developer Verification Program(twitter.com)
312 points by 0xcrypto 13 days ago | 131 comments | permalink
tl;dr: A developer's Play Store account was closed for inactivity and cannot be restored, forcing them to create a new one to publish a hobby app. Attempting to sideload via F-Droid instead, they discovered Google's new Android Developer Verification Program requires a $25 fee, identity verification, and a payments profile—even for sharing dev builds with a small number of users. The author is frustrated that Google now effectively gatekeeps all Android app distribution, not just the Play Store.
HN Discussion:
  • ↑Google closes developer accounts for inactivity even with published apps, with no recourse
  • ↑Google knowingly terminates accounts without due process because they have monopoly power
  • ↑Verification program may warrant antitrust scrutiny as it restricts alternative app stores
  • ~Technical clarification that sideloading unsigned APKs is still possible with a 24-hour wait
  • •PWAs offer a viable alternative to avoid Google's developer system entirely
18.When did Google get so weird?(sancho.bearblog.dev)
2002 points by sancho-panza 16 days ago | 1110 comments | permalink
tl;dr: The author searched Google for an old basketball meme phrase ("hes never coming over dario") and received an AI Overview that misinterpreted the query as emotional distress over a romantic rejection, offering empathetic consolation instead of links. The actual relevant results were buried hundreds of pixels below the AI response. The author argues this illustrates how Google has drifted from its mission of organizing information into forcing parasocial chatbot interactions where they aren't wanted.
HN Discussion:
  • ↑AI Overviews produce misleading or fabricated information, confirming Google search quality has degraded
  • ↑The tech industry is deliberately manipulating users with hyped-up LLMs disguised as AI
  • ↑Users should stop using Google since it has abandoned its search mission
  • ↑Google exploits loneliness by pushing parasocial AI interactions instead of real human connection
  • ↓This chatbot-style search is actually what average users have always wanted and is a product win
19.Show HN: Real-time Solar System with 526k asteroids and all tracked satellites(space.bl2.net)
398 points by wanick 14 days ago | 107 comments | permalink
tl;dr: Summary not available.
HN Discussion:
  • ↑Praise for the project's technical execution and browser performance
  • ↑Enjoyment of exploring specific spacecraft and missions via the tool
  • •Author explanation of data sources and rendering architecture
  • ↓Comparison to prior art like Celestia suggests it's not novel
  • ~Concern about satellite proliferation and Kessler syndrome polluting the night sky
20.Tcl/Tk 9.1(tcl-lang.org)
302 points by dmux 14 days ago | 143 comments | permalink
tl;dr: Tcl/Tk 9.1.0 is in development with stable releases targeted for September 2026. Tcl additions include new commands (`unicode` normalization, monotonic `timer`, `lfilter`), expanded C99 math functions in `expr`, new C APIs, and improved large-list memory efficiency. Tk gains screen reader accessibility, initial bidirectional/RTL text support, a `ttk::toggleswitch` widget, rotated label text, and improved treeview/notebook states.
HN Discussion:
  • ↑Tcl is quirky and fun to play with but risky for professional use
  • ↑Tcl's string-based metaprogramming and design is elegantly powerful
  • ↑Tk remains the easiest GUI toolkit, deserving of modern updates
  • •Nostalgic recollection of past Tcl/Tk projects and ecosystem (AOL, SQLite, Perl/Tk)
  • ↑Appreciation for the core team's stability and continued legacy (e.g., VLSI CAD tools)