Top 10 Hacker News posts, summarized
HN discussion
(398 points, 1198 comments)
Paul Graham argues that earning a billion dollars is possible without cheating through exponential growth in startups. He uses mathematical examples (e.g., 93% monthly growth turning $2 million into a billion in 9.5 months, or 15% monthly growth reaching billionaire status in ~5 years) to demonstrate that high growth rates and large markets drive wealth creation. He claims founders achieve this by building products users love so much they refer others, and emphasizes that young founders should create things they personally desire to predict future demand. Graham asserts that politicians misunderstand exponential growth, leading them to incorrectly assume billionaires must have cheated.
Hacker News critiques focus on Graham's perceived tone-deafness and ethical oversimplification. Commenters argue he conflates "earning" with wealth accumulation while ignoring systemic harm (e.g., Airbnb's impact on housing, Apple's labor practices, Facebook's societal damage). Many distinguish between "creating value" and "fairly distributing it," stating billionaire wealth inherently involves exploitation or unfair advantages (e.g., inherited wealth, privileged access). Some accuse Graham of hypocrisy, citing his advice that founders must sometimes break rules. Others challenge the math's real-world applicability, noting that exponential growth assumptions ignore unpredictable market changes and negative externalities. The consensus is that Graham's argument ignores socioeconomic context and moral complexities around extreme wealth concentration.
HN discussion
(394 points, 430 comments)
The article challenges the prevailing narrative that "everyone is using AI for everything," citing data from multiple sources indicating a more nuanced reality. Gallup, Microsoft, and Searchlight Institute data reveal that approximately one-third of the population actively uses AI, one-third uses it occasionally (e.g., monthly or less), and one-third never uses it. AI adoption has stalled among Gen Z, and negative sentiment has increased significantly year-over-year. Key barriers include concerns about job replacement (42%), privacy violations (35%), and misinformation (33%), alongside low perceived societal value (AI's net positive rating is only +8%, comparable to social media). The author analogizes AI adoption to meat consumption, noting most people consume meat moderately or avoid it entirely, suggesting AI use follows a similar continuum rather than universal adoption.
HN comments reflect varied perspectives on AI adoption and utility. Many dispute the article's premise, arguing that "everyone" is a hyperbole justified by rapid adoption (30% of US workers use AI monthly) or by redefining "using AI" to include ubiquitous tools like Google's AI-generated search summaries, which users passively consume. Others support the article's core point, citing real-world frustrations: some companies replaced reliable systems with slower, less effective LLM-based tools, while developers report inconsistent results (e.g., poor code generation for frontend tasks). Skepticism about AI's value is prominent, with users noting job-related anxiety and low ROI outside specific niches (e.g., coding or marketing). A notable counterpoint highlights transformative personal experiences (e.g., using AI to save tens of thousands on home repairs or complex projects), though these are often juxtaposed against concerns about industry hype and potential regulatory overreach.
HN discussion
(133 points, 482 comments)
The "Ask HN: What are you working on?" thread from June 2026 features a diverse array of projects across domains like AI/ML, robotics, programming languages, health tech, open-source tools, and indie products. Notable projects include a transparent HR fitness engine, a privacy-focused search engine (Uruky), an asthma tracking app (Peak Flow Meter Diary), a Rust-based collaborative knowledge system (Totem), and various AI-powered tools for agentic workflows, 3D modeling, and educational platforms. Many projects emphasize open-source principles, ethical design (e.g., carbon-negative goals), and solutions to real-world problems like UBI policy critiques, health data monitoring, and decentralized email systems. The thread also highlights recurring themes such as AI integration, local-first approaches, and niche applications like VR fitness and quantum-inspired jewelry.
Top HN comments showcase strong engagement with projects focused on privacy, health, and developer tools. Notably, Uruky’s EU-based search engine gained traction for its ethical approach and AGPL licensing, while MacroCodex’s calorie-tracking app resonated for its free, user-centric design. The Peak Flow Meter Diary campaign sparked discussion about marketing challenges for niche health tech, and Totem’s Rust-based knowledge management tool attracted interest for its cross-platform support. Comments also highlighted practical solutions like HN Alerts for trending stories and open-source alternatives to paid tools (e.g., Granola meeting data porting to DuckType). Reactions emphasized community-driven feedback, with users requesting beta access, offering suggestions for monetization, and praising projects that address underserved needs like asthma management or federated social networks (Gargoyle).
HN discussion
(328 points, 70 comments)
Kage is a tool that clones websites into folders for offline browsing by using headless Chrome to render pages, capturing the final DOM, and stripping out all JavaScript. It downloads CSS, images, and fonts to local paths, creating static HTML files that require no network dependencies or JavaScript execution. The tool offers site cloning with various configuration options, local serving of content, and packing capabilities that allow users to create ZIM archives or single executable files containing entire websites.
HN users compared Kage to similar tools like SingleFile, noted potential concerns about site load during cloning, and suggested combining it with mitmproxy for archive-grade fidelity as a WARC replacement. Practical use cases mentioned include offline access to company wikis and reading content during flights with limited internet. Security concerns were raised about Chrome's --no-sandbox flag, and Windows Defender was noted to give false positives. Users questioned why static content needs a serving process and suggested alternative solutions like curl, httrack, and pandoc-to-EPUB conversion. One commenter criticized the README as "AI slop," while another noted the demo GIF was generated using the author's "ascii-gif" project.
HN discussion
(248 points, 130 comments)
The article reveals that Rio de Janeiro's IT company IplanRIO misrepresented its model Rio-3.5-Open-397B as an original 397B-parameter model trained by the municipality. Investigation showed it is actually a direct element-wise weighted merge of approximately 60% Nex-N2 Pro and 40% Qwen3.5-397B-A17B, with no evidence of independent training. Two methods confirmed this: removing the "You are Rio" prompt caused the model to identify as "Nex, from Nex-AGI" 79% of the time while reciting Nex's backstory, and weight tensor analysis showed identical 0.6/0.4 blends across all 60 layers and network components. The model's page was later updated to confirm the merge, claiming an incorrect upload occurred and apologizing for confusion.
Hacker News comments focused on attribution, technical aspects, and regional context. Key reactions included skepticism about attribution and competence (e.g., "Oh no, someone is profiting off of their work without proper attribution"); fascination with the merging technique (e.g., "It's amazing how robust current deep learning models are" for a simple linear combination enhancing performance); and observations about Rio's reputation (e.g., humor about thieves in Rio referencing local stereotypes). Nuanced perspectives suggested possible distillation steps were omitted from the initial upload, and while the merge was criticized, some noted the technical boldness of municipal IT attempting AI work. Comments also highlighted the viral nature of the release and the importance of proper disclosure in open-source AI.
HN discussion
(198 points, 120 comments)
The article summarizes Gary Bernhardt's 2014 talk, "The Birth and Death of JavaScript," which humorously forecasts JavaScript's evolution from its creation in 1995 to its predicted demise by 2035. The presentation adopts a science fiction/comedy tone while seriously acknowledging JavaScript's flaws and its undeniable positive impact on the software industry. Bernhardt's perspective is balanced, neither promoting nor condemning the language, and concludes by referencing his more technical work on the Destroy All Software platform.
The HN discussion centers on the accuracy of Bernhardt's 2014 predictions, particularly regarding JavaScript's "death." Commenters note that while JavaScript hasn't vanished, it has become a substrate layer, used indirectly as a compilation target via WebAssembly (WASM), fulfilling the core prophecy. Key observations include JavaScript's role as a universal runtime (e.g., via Electron), the rise of TypeScript as a safer abstraction, and the realization that "death" means ubiquity rather than obsolescence. Some debate the pace of WASM's adoption and its limitations (e.g., lack of DOM manipulation), while others highlight the talk's prescient humor, including uncanny predictions about the 2020-2025 timeframe and the enduring dominance of JavaScript-based solutions despite alternatives.
HN discussion
(237 points, 52 comments)
The author developed a project to index 2,207 GoPro videos locally on an M1 Max computer using open-source ML models, enabling efficient search for interesting moments from cycling footage. After processing 628 videos (totaling 668.68 GB and 15h 13m 18s of footage), the system extracts frames, applies ML analysis to identify content, and allows direct export of selected clips to DaVinci Resolve for editing. The project emphasizes local processing, privacy, and integration with professional video software.
Hacker News commenters noted the project’s practicality for local video analysis but suggested a "Show HN" format would be more appropriate. Technical discussions included comparisons of M1 Max performance to Intel i9 CPUs and feasibility on Windows ARM, alongside tips for leveraging Apple GPUs in containers. Users shared similar DIY projects (e.g., SimbaStack’s local video indexing) and highlighted DaVinci Resolve’s native AI indexing feature as a built-in alternative. Practical concerns arose about processing time (67+ hours) and resource requirements, with queries about cloud acceleration and compatibility with 32GB GPUs like RTX 5090.
HN discussion
(202 points, 75 comments)
The article explains the implementation of Anubis, a server-side protection system designed to deter AI companies from aggressively scraping websites. Anubis operates as a Proof-of-Work mechanism, similar to Hashcash, imposing computational costs that become prohibitive for large-scale scrapers while being negligible for legitimate users. It serves as a temporary solution while more advanced fingerprinting techniques (like detecting headless browser font rendering) are developed to better distinguish scrapers from real browsers. The system requires modern JavaScript features, which conflicts with privacy-enhancing plugins like JShelter that disable them.
The HN comments primarily express confusion and skepticism regarding the article's title ("Linux 7.1"), as the content focuses entirely on the Anubis anti-scraping system, not a Linux kernel release. Users question the version numbering logic ("Is it safe to assume we can see this in Debian Stable around 2036?") and note the lack of substantive changes ("Is there anything particularly interesting about this?"). One comment humorously mentions seeing an anime avatar during loading, suggesting a potential visual element of the protection. The most substantive comment highlights a positive side effect of AI: AI-assisted bug reporting led to the removal of obsolete kernel code (like ISDN drivers) to reduce noise from false bug reports, framed as a beneficial "trimming the fat."
HN discussion
(165 points, 53 comments)
Jane Street historically avoided formal methods due to prohibitive costs, exemplified by the seL4 microkernel verification requiring 25 person-years for 8,700 lines of C. However, the emergence of agentic coding has shifted their perspective, reducing verification costs through AI assistance and increasing benefits by addressing the "verification bottleneck" where AI-generated code often fails to maintain invariants or quality. They now aim to build a team to make formal methods as pervasive as their sophisticated type systems, leveraging their control over OCaml and developer community to integrate proof techniques into the language itself.
Hacker News comments debate whether AI can realistically reduce verification costs without introducing its own errors, given the risk of "sloppy" verification code. Key themes include skepticism about formal methods' scalability for non-deterministic domains (e.g., UIs), the enduring challenge of ensuring specifications align with systems, and praise for Jane Street's language-level approach. Practical experiences highlight AI's potential in proof assistants like Lean 4, where models automate complex proofs but require human guidance. Rust's borrow checker is noted as a practical application of formal methods, though users often bypass constraints. Some argue formal methods excel in domains like finance but are impractical for "offensive programming" in uncertain environments.
HN discussion
(145 points, 43 comments)
zeroserve is a high-performance HTTPS server that executes eBPF scripts in userspace. It now supports Caddy compatibility mode, where it JIT-compiles Caddyfiles to eBPF and then to native x86_64/ARM64 machine code, running under an io_uring event loop. This implementation achieves reported 3x throughput and 70% lower latency compared to Caddy. Users can run zeroserve with a Caddyfile and leverage its Turing-complete eBPF capabilities to call custom code, such as implementing AWS SigV4 authentication for S3 reverse proxying. The tool is distributed via GitHub releases.
The HN discussion highlights significant excitement about zeroserve's technical innovation (JIT compilation of web server logic via eBPF) and performance claims. Key concerns raised include skepticism about whether Caddy is ever a practical bottleneck ("has anyone ever encountered a use case where Caddy was the bottleneck?"), security warnings around io_uring ("less safe," "enormous attack surface"), and missing critical features like ACME ("dealbreaker"). Commenters note nginx's surprising resilience and question project longevity ("dead in 6 month Rust project") and real-world utility ("hard pressed to find an application where this is meaningfully useful"). Confusion persists around eBPF's Turing Completeness and the legitimacy of a Chrome certificate prompt for the project domain.
Generated with hn-summaries