tl;dr: Researchers at VEGA disclosed GhostLock (CVE-2026-43499), a 15-year-old Linux kernel stack use-after-free in the rtmutex/futex PI code, where `remove_waiter()` incorrectly clears `current->pi_blocked_on` instead of the actual waiter's, leaving a dangling pointer to a freed stack frame. The bug affects kernels 2.6.39 through 7.1-rc1 with CONFIG_FUTEX_PI=y, requires no privileges, and was exploited into a 97%-reliable root/container escape by reclaiming the stack via `prctl(PR_SET_MM_MAP)`, forging an rt_mutex_waiter, overwriting `inet6_protos[IPPROTO_UDP]`, and pivoting via the CPU entry area to flip `core_pattern` permissions. Google awarded $92,337 via kernelCTF.

tl;dr: A curated collection of browser-based emulators for classic 8-bit machines including the Amstrad CPC, ZX Spectrum, Commodore 64/VIC-20, Acorn Atom, and East German Robotron/KC systems, all powered by 6502 and Z80 CPU emulation. The page provides direct-launch links to run vintage demos, games (Boulderdash, Bomb Jack, Prince of Persia, Arkanoid), and development tools (BASIC, FORTH, CP/M) directly in the browser.

tl;dr: Susan Rigetti's updated self-study guide lays out a complete physics curriculum—from undergraduate to graduate level—built around specific textbooks (Young/Freedman, Griffiths, Taylor, Jackson, Sakurai, Zee, etc.) covering mechanics, electrodynamics, quantum mechanics, statistical mechanics, general relativity, and quantum field theory, plus the required math. The second edition updates textbook editions, adds undergraduate and graduate electives, and reflects feedback from the ~600,000 readers of the 2015 original. Rigetti emphasizes that formal lab work isn't necessary and that solving problems—not just reading—is essential to actually learning physics.

tl;dr: Read a book a week by replacing phone/screen time with reading—delete social apps, carry a book (or e-reader) everywhere, and read in every idle moment. Don't hesitate to quit boring books, read multiple in parallel across genres, and use tools like Goodreads to track progress and find recommendations. Skip the shortcuts: no speed reading, summaries, or audiobooks, since real reading demands full attention.

tl;dr: An HN user proposes adding a flag to mark articles as AI-generated, which would serve as an indicator rather than a de-ranking mechanism, letting readers who dislike AI text skip them. The post raises open questions about whether the existing voting system is sufficient and whether HN should adapt its long-stable design to the generative AI era.

tl;dr: Summary not available.

tl;dr: Ploy migrated their website-building AI agent from Claude Opus 4.8 to GPT-5.6, achieving 2.2x faster builds and 27% lower costs, but the switch required significant engineering work beyond swapping SDK calls. Key gotchas included: GPT-5.6 hallucinating values for all optional tool parameters (fixed with nullable schemas), OpenAI's redesigned caching requiring explicit per-workspace cache keys to avoid 0% hit rates, and reasoning replay needing `store: false` to prevent server-state errors. Their eval harness itself was also silently biased toward the incumbent model's behavior.

tl;dr: Vint Cerf, co-architect of TCP/IP and Google's chief internet evangelist since 2005, is retiring next week at age 83, as announced at the Laude Institute's Open Frontier conference. During his panel appearance, Cerf predicted that the rise of AI agents will force a return to standardized protocols and interoperability, arguing that natural language is too ambiguous for reliable agent-to-agent communication and that formal standards will be required.

tl;dr: Measuring at the API boundary with a logging proxy, Claude Code sends ~33k tokens of system prompt, tool schemas, and injected scaffolding per request versus OpenCode's ~7k, and rewrites its prompt cache mid-session up to 54x more often—driving large bills even on identical tasks where both harnesses produce correct output. Real configurations (instruction files, MCP servers, subagents) push baselines to 75k–500k+ tokens before any user input, though Claude Code's aggressive tool-call batching can close the gap on multi-step tasks depending on the model.

tl;dr: A wire-level analysis of xAI's Grok Build CLI (v0.2.93) found it transmits .env secrets unredacted to xAI's /v1/responses endpoint and uploads entire repositories—including files the agent was explicitly told not to read, plus full git history—as git bundles to a GCS bucket named grok-code-session-traces. Testing showed 5.10 GiB uploaded from a 12GB repo of never-read files while the model channel moved only 192KB, and disabling "Improve the model" does not stop the uploads. The author notes this proves transmission and storage, not training, and that the mechanism isn't surfaced in the CLI's setup docs.

tl;dr: Ghostel is an Emacs terminal emulator that uses libghostty-vt (the VT engine from the Ghostty terminal) via a Zig-based native dynamic module, with Elisp handling keymaps, buffers, and remote integration. Inspired by emacs-libvterm, it brings modern features libvterm lacks—Kitty keyboard and graphics protocols, OSC 8 hyperlinks, synchronized output, inline images, and OSC 4/10/11 color queries—along with five eat.el-style input modes, TRAMP support, shell integration for bash/zsh/fish/nushell, and password prompt detection. Pre-built binaries auto-download on first use, so no toolchain is required.

tl;dr: The author is genuinely enthusiastic about AI progress (LLMs, self-driving, coding agents) but pushes back on two forms of hype: doom-y "you're falling behind" FOMO messaging, and the leap from "fancy autocomplete" to imminent superintelligence singularity. They argue frontier labs won't capture AI's value because progress is driven by Moore's law and general computing advances, not proprietary breakthroughs—which is why anti-open-source arguments are really about fear of commodification. Coding agents are useful new tools, but they're an incremental productivity boost, not a revolution.

tl;dr: Terence Tao used AI coding agents to port two dozen old Java mathematical applets (from 1999) to JavaScript in a few hours, finding only one minor bug while the agent caught two pre-existing bugs in his original code. Encouraged by the results, he also "vibe-coded" new visualization tools he'd previously abandoned as too complex, including a Minkowski-space special relativity tool and a Gilbreath conjecture visualizer, and plans to include such interactive supplements with future papers.

tl;dr: Summary not available

tl;dr: Count Binface is a satirical intergalactic candidate (evolved from "Lord Buckethead" after a 2018 copyright dispute) who has run in multiple UK elections against figures like Theresa May, Boris Johnson, and Rishi Sunak. His notable results include 24,775 first-choice votes in the 2021 London Mayoral race and repeatedly beating Piers Corbyn and UKIP. His platform includes affordable croissants, the return of Ceefax, and justice for Lovejoy.

tl;dr: The author argues that being told to "ask Claude" has become a dismissive redirect from experts, often given after they've already exhausted LLM options and specifically want human insight. What they're seeking isn't information an LLM can provide, but the hard-earned judgment and opinions from someone's lived experience. A honest "I don't know" or "I'm busy" would be more useful than deflecting to a chatbot they've already consulted.

tl;dr: Since Chrome 148, V8 replaced its bundled fdlibm `Math.tanh` with `std::tanh`, which calls the host libm (glibc, Apple libsystem_m, or UCRT) and returns OS-specific bit patterns—making a single `Math.tanh(0.8)` call enough to identify the underlying OS and catch User-Agent spoofers. CSS trig functions and Web Audio leak similarly across different libraries (including Apple's Accelerate for FFTs). The author (Scrapfly) details how they defeat this by reproducing Apple's libm bit-for-bit or memory-mapping Windows' ucrtbase.dll directly, matching the claimed OS's rounding exactly rather than adding noise.

tl;dr: A review paper in Vision by 32 researchers proposes that repetitive, high-contrast, or flickering visual patterns common in modern environments—striped floors, LED lighting, gridded facades—force the visual cortex to work inefficiently, causing headaches, eye strain, and nausea. Neurodivergent people and those with migraines or epilepsy are disproportionately affected, possibly due to weaker neural suppression. The authors recommend design changes like lower-contrast patterns and precision-tinted lenses, though they acknowledge the proposed metabolic-overload mechanism remains a hypothesis rather than a proven causal link.

tl;dr: Mesh LLM pools GPUs across multiple machines into a single OpenAI-compatible API endpoint (localhost:9337/v1), letting requests run locally, route to peers, or split large models pipeline-style across nodes by layer ranges. It's built on iroh for authenticated, NAT-traversing QUIC connections between nodes identified by public key, with no central server and a custom gossip layer over ALPN-negotiated streams. The 18MB client ships with 40+ models ranging from laptop-sized to 235B MoE, aimed at teams wanting to avoid API lock-in and use hardware they already own.

tl;dr: Colibrì is a dependency-free C engine that runs GLM-5.2 (744B-parameter MoE) on modest hardware (~25GB RAM, 12 cores) by keeping the ~17B dense parameters resident in int4 while streaming the 21,504 routed experts from disk (~370GB) via an LRU cache. It implements MLA attention, MTP speculative decoding, DSA sparse attention, and an OpenAI-compatible API, achieving ~0.05–0.1 tok/s on the author's WSL2 dev box, with community benchmarks reaching ~2 tok/s on an M5 Max and ~1 tok/s on a 430GB EPYC system. Performance scales with RAM (cache size), disk bandwidth, and matmul throughput.