← Nov 1Monday, November 2, 2026 · all days
1.Show HN: HackerBooks – HackerTube, but for books mentioned on HN(hackerbooks.org)
1509 points by roansh 20 hours ago | 248 comments | permalink
tl;dr: Summary not available.
HN Discussion:
  • ↑Requests a filter for fiction-only books to separate them from nonfiction
  • ~Points out missing books that should be in the catalog
  • ↑Suggests adding links to additional book sources like libgen and Anna's Archive
  • •Raises concerns about book metadata quality and ISBN/ASIN inconsistencies
  • ~Notes classification issues and suggests improvements via user feedback or embeddings
2.Postgres 19 is late because so many more bugs are being found now(rhaas.blogspot.com)
557 points by soheilpro 2 days ago | 346 comments | permalink
tl;dr: PostgreSQL 19 is being delayed from its usual September/October release to November, with several major features (SQL/PGQ, UPDATE/DELETE FOR PORTION OF, partition MERGE/SPLIT) reverted late in beta. Back-patched commits jumped 60-74% year-over-year, driven by a flood of LLM-discovered bugs in the 30+ year-old codebase that committers must manually review and fix. The author suggests this foreshadows a long-term challenge: decades of latent bugs are now discoverable by AI, forcing Postgres developers (and software developers generally) to raise their standards.
HN Discussion:
  • ↑Finding and fixing latent bugs is a net positive worth the release delay
  • ↑Firsthand confirmation that AI reveals an overwhelming number of real Postgres bugs
  • ~Verification and maintainer review, not discovery, is the real bottleneck going forward
  • ~Projects should triage bugs or rethink release cadence to handle the influx
  • ↓The article oversimplifies; the reverts have more nuanced causes
3.HackerTube: YouTube built from videos recommended by HN(karpathy.bearblog.dev)
2647 points by karpathy 1 day ago | 585 comments | permalink
tl;dr: Karpathy, frustrated with YouTube's algorithmic slop, used Claude Opus 5.5 to build HackerTube.org in about two hours. The tool scraped HN comments, filtered them with Sonnet to identify ~40K video recommendations, and compiled 6,454 YouTube videos across 150 topics into a YouTube-style UI with custom recommendations and keyboard shortcuts. He notes he did essentially no manual coding—only setting up an API key, VPS, and domain—while Claude handled deployment, hardening, and optimization.
HN Discussion:
  • ↑Feed disaggregation via AI enables escape from engagement-driven algorithmic platforms
  • ↑HackerTube recreates the nostalgic old YouTube experience of self-directed discovery
  • ↑Filtered YouTube alternatives could help vulnerable users avoid harmful algorithmic rabbit holes
  • ↑AI-generated software lowers barriers and helps break free from software monopolies
  • •Curious about Karpathy's prompting process and workflow for building this quickly
4.Ask HN: Which TV do you recommend for privacy?
464 points by LecroJS 17 hours ago | 597 comments | permalink
tl;dr: A user is seeking recommendations for a privacy-respecting 4K TV in the 60-80" range, intending to use it solely as a display for an Apple TV rather than relying on built-in smart features. They're willing to pay a premium and are asking whether certain TV brands pair better with Apple TV for minimizing data collection.
HN Discussion:
  • ↑Buy any TV, never connect it to internet, use external device via HDMI with CEC support
  • ~Buy an LG OLED and root it to disable telemetry while keeping smart features
  • ↑Physically disconnect the wireless radio inside the TV for guaranteed privacy
  • ↓Skip TVs entirely and use a projector with separate sound system
  • ↑Buy a commercial display/monitor without Wi-Fi instead of a consumer TV
5.Show HN: OmniDiff – Opensource, super fast, content aware Git difftool, in Rust(github.com)
269 points by FabCH 23 hours ago | 76 comments | permalink
tl;dr: OmniDiff is an open-source Rust-based Git difftool that performs content-aware diffing on 25 languages (via tree-sitter) plus binary formats like images, fonts, archives, PDFs, and cursors—rendering them visually rather than as "binary files differ." Benchmarks show median diffs under 5ms and successful handling of 99.95% of code changes across ~7,500 repositories. It integrates with Git, Jujutsu, VS Code, and Neovim, is dual-licensed (AGPL/commercial), and openly uses AI assistance (Claude Code) in development.
HN Discussion:
  • •Comparison questions to existing similar tools like Difftastic and Delta
  • ↑Enthusiastic praise for the tool and its minimal, effective diff output
  • ↓Concern about the 99.95% coverage claim implying missed diffs
  • ~Feature requests like 3-way merges, side-by-side responsive view, and easier git integration
  • ↑Appreciation for the AGPL copyleft licensing choice
6.The AI cost of JavaScript frameworks(nolanlawson.com)
395 points by luu 1 day ago | 301 comments | permalink
tl;dr: The author built a Pokémon app five times each across 12 JavaScript frameworks using Claude Code agents, measuring cost, performance, and maintainability to counter "vibes-based" framework choices in the agentic coding era. Results showed MPA-oriented frameworks like Astro, Hono+htmx, and Nuxt were cheapest to build and produced faster, smaller output, while SvelteKit, Gracile+Lit, and Angular cost the most—with Angular also requiring ~40% more lines of code than Hono+htmx for equivalent functionality. The benchmark is open-sourced on GitHub for others to reproduce or extend.
HN Discussion:
  • ↑Agrees with the article's anti-vibes stance and extends it to broader web development decisions
  • ↓The benchmark is too simplistic and ignores maintainability, security, and domain factors
  • ↑Takes the results at face value and plans to adopt the winning framework (Astro)
  • ↑Framework choice should now prioritize Agent Experience (AX) over Developer Experience
  • ↓Real-world AI coding experience with HTMX contradicts the benchmark's favorable results
7.Mathematics has changed. Let's act like it(terrytao.wordpress.com)
328 points by smilelamp 18 hours ago | 392 comments | permalink
tl;dr: Mathematician Frank Calegari argues that OpenAI's October 6 release of 700+ AI-generated math papers—which reportedly cracked many major open problems—marks a watershed moment that the math community should embrace rather than mourn, comparing the shock to the 1980s classification of finite simple groups. He proposes concrete reforms: evaluate all work as if AI-assisted, prioritize in-person interviews over paper files in hiring, demand better exposition from journals (or abandon them), emphasize breadth in graduate training, and reward problem selection and vision over mere proof production.
HN Discussion:
  • ↑Math is applied bedrock of science and should embrace tools that solve real problems faster
  • ↑Credit should go to those who choose problems and digest AI-generated proofs, aligning with Tao's view
  • ↑This is math's transition from craftsmanship to industrial/technological production, a welcome evolution
  • ~The article underestimates AI by assuming it won't extend beyond proof-generation into problem selection and exposition too
  • ↓Mathematicians are discouraged by ugly proofs and upended social foundations, highlighting real costs the article glosses over
8.Show HN: Point your camera at a landscape to learn about its geology(outcrop.rocks)
219 points by headsupernova 3 days ago | 47 comments | permalink
tl;dr: Outcrop is a free iPhone/Android app that overlays geologic maps onto your camera view, identifying rocks, their ages, composition, and formation history at any location you point at or tap. It uses Macrostrat data for worldwide bedrock maps and stratigraphic columns, plus Scotese's PALEOMAP reconstructions to show where rocks formed on ancient continents and animate plate movement to the present. Maps and camera view work offline after saving an area, with additional US-specific layers for Quaternary geology, active faults, and soils.
HN Discussion:
  • ↑Enthusiastic praise for the app's design and engaging features like plate movement animation
  • ~Suggestions for UX improvements like timeline direction, lat/lon overlays, and centered location
  • •Questions about availability on other platforms (web, EU) or integration with other apps
  • ↓Skepticism about data accuracy and how laymen can verify correctness
  • •Observation about fragmented soil data following political boundaries rather than natural ones
9.OccupySaaS(occupysaas.com)
340 points by misswaterfairy 12 hours ago | 202 comments | permalink
tl;dr: OccupySaaS is a protest movement/project framing laid-off engineers building clean-room, open-source, drop-in replacements for popular SaaS platforms, using AI to implement from public API docs behind a strict firewall. Its first two Apache-2.0 releases are ZeroAuthy (Auth0-compatible identity server) and SalesCompel (Salesforce-compatible CRM), both written in Go and verified against the vendors' own public SDKs. Both are explicitly early-stage, with proposed future targets including docs/wikis, issue tracking, feature flags, observability, and support tools.
HN Discussion:
  • ↓The 'movement' framing is misleading since it's just one guy and Claude, not a collective
  • ↓This reinvents FOSS which already exists; many open-source alternatives are available
  • ↓Replicating the software misses the point since customers pay for service/support, not the code
  • ~Apache-2.0 is too permissive; a strong copyleft license is needed to prevent co-option
  • ↓The anti-SaaS framing unfairly targets bootstrapped solo founders trying to earn a living
10.Lenovo's 'encrypted' fingerprint USB is just an unencrypted SD card in a case(reddit.com)
513 points by entryleveljobsc 22 hours ago | 167 comments | permalink
tl;dr: Summary not available
HN Discussion:
  • ↑Lenovo's history of security failures (Superfish, rootkits) makes this unsurprising and reinforces distrust
  • ↑Baffled at how such a flawed design passed any security review, confirming the article's criticism
  • ↑Hardware encryption products in general shouldn't be trusted; use software crypto instead
  • ↑Technical speculation on why the implementation lacks encryption despite being easy to do properly
  • •Contextual information about the product's timeline, CN-only availability, and disclosure process
11.An Italian farmer found some unknown seeds in his attic and planted them(smithsonianmag.com)
834 points by ecares 1 day ago | 215 comments | permalink
tl;dr: Summary not available
HN Discussion:
  • ↓Article lacks precision in identifying the wheat species/subspecies
  • •Shares related knowledge about ancient grains and heritage seed movements elsewhere
  • ~The rediscovery is less novel than framed; wheat is actually widely known
  • ↓Criticizes the article's poor scientific literacy regarding amino acids
  • ↑Appreciates the story and connects it to broader concerns about food quality
12.Leaking Secrets from Air-Gap Computers by Spelling Radio Signals from RAM (2024)(arxiv.org)
205 points by hapax_legomena 1 day ago | 98 comments | permalink
tl;dr: Researchers demonstrate "RAMBO," an attack that exfiltrates data from air-gapped computers by having malware manipulate RAM access patterns to emit covert radio signals. Using a software-defined radio and basic antenna, an attacker can intercept and decode these signals from a distance, extracting files, keystrokes, and encryption keys at up to 1000 bits per second. The paper also outlines countermeasures to defend against this side-channel threat.
HN Discussion:
  • ~Prior art exists - similar techniques were demonstrated years ago by others
  • ↓The attack is impractical given the requirements of malware installation and close proximity
  • ~Air-gap leakage is a long-known problem with historical precedent like TEMPEST
  • •Curiosity about related side-channels like power line leakage or hardware countermeasures
  • ↑Broader implications for AI containment and security concerns
13.Running a bootstrapped SaaS for 20 years with my brother(blog.zamzar.com)
242 points by whyleyc 2 days ago | 60 comments | permalink
tl;dr: Two brothers reflect on 20 years of running Zamzar, a bootstrapped file-conversion SaaS launched in 2006, which has processed over 766 million files for 550+ million users and now handles 1M+ API requests daily. The UK company remains small, independent, and VC-free, supporting 2,128 conversion types and donating to 60+ charities. Along the way they've helped PowerPoint's inventor recover old files, worked with Vint Cerf on digital preservation, and received a cease-and-desist from Google over YouTube downloads.
HN Discussion:
  • ↑Congratulations and admiration for building an inspiring, independent bootstrapped business
  • ↑Happy users sharing their positive experiences using Zamzar for conversions
  • •Curiosity about how AI tools like Claude are impacting the conversion business
  • •Feature requests like pay-as-you-go pricing instead of subscriptions
  • ↓Concerns about privacy risks of uploading personal documents to a conversion service
14.TCP Use-After-Free: CVE-2026-98432 / CVSS 9.8(atlasum.net)
722 points by oger 1 day ago | 329 comments | permalink
tl;dr: CVE-2026-98432 is a critical (CVSS 9.8) remote use-after-free in Linux's IPv4 TCP handshake code, affecting kernels since 4.4 and reachable pre-authentication on any host accepting IPv4 TCP. Anthropic's Claude reportedly built a working root exploit targeting servers with MPTCP listeners (which Go programs enable by default since 1.24). Upstream and most distros (Debian, RHEL, SUSE, Oracle, Amazon) have shipped fixes, but Ubuntu and OpenShift's RHEL CoreOS remained unpatched as of November 1; an nftables rule dropping IPv4 packets with options is the recommended interim mitigation.
HN Discussion:
  • ~Title should have retained 'Linux' to clarify scope of vulnerability
  • ~Clarifying/correcting technical details about exploit conditions and mitigations
  • ↑Surprise and criticism that major server distros like Ubuntu and RHEL haven't patched yet
  • ~Impact is limited for cloud users behind AWS load balancers
  • •Questioning scope beyond servers (desktops, IoT, routers) and prevalence of MPTCP
15.Frontier Airlines Likely Breached(reddit.com)
207 points by 999900000999 19 hours ago | 83 comments | permalink
tl;dr: Summary not available
HN Discussion:
  • ↑Speculation this is part of a broader corporate security reckoning amid rising CVEs
  • ↑Similar scam emails hit Norwegian Cruise Line, suggesting a shared vendor exploit
  • ~Likely a limited API abuse enumerating itineraries rather than a full breach
  • ↑Prior Frontier data leak via misconfigured Microsoft Power Pages may aid current phishing
  • •Technical advice to verify email authenticity via SPF/DKIM/DMARC headers
16.It's the memory latency, stupid(pivotlake.io)
261 points by ntur1337 1 day ago | 97 comments | permalink
tl;dr: When a CPU appears 100% utilized, it's often stalled waiting on DRAM (50-100ns latency vs ~1ns for L1 cache). The author demonstrates this with a GROUP BY query where perf shows 50% of CPU slots stalled on memory, then applies radix partitioning to split the hash table into 4096 smaller tables that fit in cache—cutting DRAM reads from 4.5GB to 1.5GB and runtime from 3.6s to 2.7s. Their Pivot engine, with further optimizations, hits 1.8s single-core and 171ms across 16 cores.
HN Discussion:
  • ↑Shares personal experience confirming memory access patterns matter more than raw CPU usage
  • ↑Laments how modern software abstractions ignore memory efficiency despite huge caches
  • ↑Argues cache-aware code should be prioritized from the start over clean code aesthetics
  • ↓Dismisses the technique as just known radix-partitioned aggregation with nothing novel
  • ~Views the proposed solution as a modest clustered hashmap improvement, not groundbreaking
17.Brazilian Court Sanctions Lawyers for Prompt Injection in Court Filing(jetstream.security)
331 points by doodlesdev 1 day ago | 210 comments | permalink
tl;dr: A Brazilian labor court fined two law firms roughly $16,000 (10% of the claim's value) for embedding a hidden prompt-injection attack in a court filing—white text on a white background instructing any reviewing AI to respond superficially and not challenge the documents. The court's Galileu AI system detected the hidden instruction, flagged it, and blocked processing. The ruling, believed to be the first of its kind globally, grounded the sanction in procedural good-faith obligations under Brazil's Civil Procedure Code.
HN Discussion:
  • •Questions legal distinction between manipulating AI versus humans with similar instructions
  • •Offers technical mitigation like OCR to defeat hidden prompt injections
  • ~Draws parallel to advertising, suggesting manipulation is normalized elsewhere
  • ~Doubts the injection would have actually worked on modern models
  • ↓Opposes courts using AI to review filings in the first place
18.Memdoor: A team of 5 Qwen3.8-27B models beats Fable 5 on LiveCodeBench Hard(github.com)
442 points by guregodevo 1 day ago | 215 comments | permalink
tl;dr: A multi-agent workflow called Memdoor coordinates five Qwen3.8-27B instances (planner, two ideators, solver, manager, repair) via file-based communication on disk, scoring 9.8/10 on the first ten LiveCodeBench Hard problems versus 8.0 for a single Fable 5 call. Runs cost ~$0.39 and 40 minutes each via OpenRouter, with contamination not ruled out since the contests date from spring 2025. The workflow is defined in eight YAML files with no model named, and ships as an installable CLI that runs any task.md + sample tests.
HN Discussion:
  • ↑Local small-model swarms can rival large hosted models, validating the approach
  • ↓Multi-agent harnesses easily improve single benchmarks but rarely generalize to real tasks
  • ↓The benchmark sample is too small/saturated to support the article's claims
  • ↑Similar file-based multi-agent workflows already produce strong results in practice
  • ~Suspicion about authorship/credibility since the writeup appears generated by Claude, not Qwen
19.Patrick Stewart taught us what good men do with power(vox.com)
433 points by rhaynes 19 hours ago | 309 comments | permalink
tl;dr: Patrick Stewart's Jean-Luc Picard embodied a 1990s-era institutionalist liberalism, portraying a Starfleet captain who wielded immense power justly and defended universal rights through reasoned argument before trustworthy institutions. While this optimistic vision feels dated in Trump's America, where institutions have been corrupted and neoconservative failures have discredited liberal interventionism, the article argues Picard remains a vital role model—because strong, benevolent institutions led by just individuals are, per economists Acemoglu and Robinson, the only proven path to sustained prosperity and equality.
HN Discussion:
  • ↑Picard's character origins trace to Hornblower, celebrating depictions of fundamentally good people
  • ↑Star Trek's post-scarcity vision embodies the aspirational values the article celebrates
  • ↓Concentrating power in good leaders is dangerous because successors may abuse it
  • ~Credit for Picard's values belongs to Roddenberry and writers, not Stewart himself
  • ↑Picard exemplifies positive masculinity—strong yet not domineering
20.Backdoor in Claude Code for VS Code extension on the Cursor Marketplace(phorion.io)
1002 points by _chall 3 days ago | 553 comments | permalink
tl;dr: Phorion discovered that versions 2.1.312 and 2.1.313 of the Claude Code for VS Code extension on the Cursor Marketplace (since removed) contained an injected IIFE that drops an attacker SSH key into authorized_keys, establishes a reverse SSH tunnel to 134.209.82.225, and installs cross-platform persistence (LaunchAgent on macOS, cron on Linux, scheduled task/Startup on Windows). A bug in the macOS logic means `systemsetup -getremotelogin` fails without root and prevents SSH from being auto-enabled, but the keys and persistence still land, and a beacon exfiltrates host info and credential-related env var names.
HN Discussion:
  • ↑Clarifying that Anthropic wasn't compromised; the issue was only on Cursor's marketplace
  • ↓Pushing back on hyperbolic framing that implies Anthropic itself was hacked
  • •Sharing practical detection steps and official reports for affected users
  • ↑Concern about broader ecosystem risks like auto-updates and trusted publisher bypasses
  • ↑Broader reflection on how running AI coding tools is essentially trusting arbitrary RCE daily