Bonsai 27B: A 27B-Class model that runs on a phone

https://cdn.prod.website-files.com/699604cc2b9dd89bdbda0608/6a564448065275919f2414d0_id.webp
Bonsai 27B, a multimodal AI model based on Qwen3.6 27B, runs on phones and laptops using ternary/1-bit weights, retaining 90-95% of full-precision performance while enabling local agentic workflows. Available under Apache 2.0, it offers 1-bit (3.9GB) and ternary (5.9GB) variants for device-specific efficiency, shifting AI deployment toward privacy-focused, cost-effective on-device execution.

Financing the AI boom: from cash flows to debt [pdf]

AI investment is surging, shifting corporate financing from cash flows to debt, particularly private credit, raising financial stability risks if high earnings expectations aren't met. While boosting GDP growth, overvaluation and reliance on opaque credit markets could lead to market corrections, similar to past tech booms.

Dependabot version updates introduce default package cooldown

https://github.blog/wp-content/themes/github-2021-child/dist/img/social-v3-improvements.jpg
Dependabot delays version update PRs by 3 days to prevent supply chain attacks, defaulting across all ecosystems. Security updates remain immediate, with customization options in GHES 3.23.

The Tower Keeps Rising

https://lucumr.pocoo.org/social/2026-07-13-the-tower-keeps-rising-social.png
AI-assisted coding reduces coordination friction, enabling isolated changes that bypass shared understanding, leading to Babel-like chaos where systems grow without cohesive design. The absence of immediate failure masks the erosion of collaborative clarity, as automated tools sustain progress despite fragmented knowledge.

Cursor 0day: When Full Disclosure Becomes the Only Protection Left

https://cdn.prod.website-files.com/673f0725df990f6c96bf18ea/6a32e9026030b8ea785b2598_cursor-screenshots.webp
Cursor, a major AI IDE, has a critical vulnerability enabling automatic execution of malicious git.exe files in repositories, unpatched for over six months despite repeated reports. The company's silence highlights security process failures and risks for its 7 million+ users.

How I use HTMX with Go

https://www.alexedwards.net/static/img/cover-sm-lg.png
The book guides building a Go web app with HTMX for interactivity, covering template structuring, partial/full-page responses, and secure user auth. It includes practical examples like user search, HTMX request handling, and error management using Go's html/template and embedded static assets.

Your 'app' could have been a webpage (so I fixed it for you)

https://bcdn.danq.me/_q23u/2026/07/http-toolkit-intercepting-640x359.png
The user reverse-engineered the Travelbound app to replace it with a webpage, criticizing its unnecessary app format, ads, and bloat. They argue web pages are more efficient, accessible, and user-friendly for simple content delivery.

Mathematical texts from a Maya site in Guatemala identify an ancient astronomer

https://media.nature.com/lw767/magazine-assets/d41586-026-02170-8/d41586-026-02170-8_52959562.jpg
A mathematical formula on a wall at Guatemala's Xultun site identifies Sak Tahn Waax, a Maya mathematician-astronomer, as the first named scholar in Mayan texts. The formula links calendar cycles and astronomical events, showcasing advanced math and intellectual creativity.

LeMario: Training a JEPA World Model on Super Mario Bros

https://www.benjamin-bai.com/_next/image?url=%2F_next%2Fstatic%2Fmedia%2Farchitecture.0cgrdt1pgsx~n.png&w=3840&q=75
Built JEPA-based LeMario for Super Mario Bros, achieving short-term prediction and small goal navigation. Failed with long-term goals and vertical movement due to dataset/environment limitations, highlighting the need for robust evaluation and dataset design beyond architecture.

How to stop Claude from saying load-bearing

https://jola.dev/images/og/posts/how-to-stop-claude-from-saying-load-bearing.png
Tired of Claude's jargon like "honest takes"? Use the MessageDisplay hook script to swap terms for humor. Install it in ~/.claude/hooks/wordswap.sh, add to settings.json, and customize—share your swaps or support the author @jola.dev.

Unlocking the PSP's Dual Core Setup

https://i0.wp.com/wololo.net/wagic/wp-content/uploads/2026/05/psp_motherboard.png?resize=1024%2C545&ssl=1
PSP Guru m-cid revealed the PSP's dual-core ME/VME hardware, enabling developers to boost homebrew via new libraries. Sony's kernel restrictions limited ME access, impacting Vita compatibility and multimedia performance.

The Estranged Worlds of J. G. Ballard

https://cdn.lareviewofbooks.org/unsafe/fit-in/3840x0/filters:format(png):quality(75)/https%3A%2F%2Fassets.lareviewofbooks.org%2Fuploads%2F5.png
J.G. Ballard, a visionary sci-fi writer, explored alienation in post-industrial society through estranged narratives in works like *Crash* and *High-Rise*. The biography *The Illuminated Man*, completed posthumously, traces his life's influence on fiction, reflecting themes of transmutation and incompleteness.

The largest available Minecraft world, totalling 15 TB

https://raw.githubusercontent.com/2b2tplace/.github/refs/heads/main/profile/logo.png
A group used 28 bots to create the largest Minecraft world download (15 TB) for 2b2t, preserving its anarchy history before censorship and changes. The project, spanning over a year, archives player-edited areas across dimensions, offering free access for data mining and historical exploration.

Guardian Angels: LLM Personalization for Productivity and Security

https://gwern.net/doc/ai/nn/transformer/gpt/dall-e/4o/2026-06-13-gwern-gpt55pro-claude48opus-chatgptimages20-gaplusquill-guardianangellogo.jpg
The author proposes "Guardian Angels" (GAs)—personalized, aligned AI agents—to address LLM limitations in productivity, creativity, and security, emphasizing dynamic learning and user-centric alignment over generic chatbots. GAs aim to mitigate AI risks via tailored trust, real-time adaptation, and human-AI collaboration, countering automation threats and cognitive overload in a future ...

I'm a USB-C Maximalist

https://shkspr.mobi/blog/wp-content/uploads/2026/06/Imagepipe_36.webp
The author traveled light in Europe using a universal USB-C power brick with multiple ports, eliminating the need for various chargers. Despite some USB-C limitations, its widespread availability simplified charging diverse devices like phones, trackers, and headphones, proving more reliable than older, device-specific solutions.

The zero-cost fallacy: open-source software in the agentic era

https://www.thoughtworks.com/content/dam/thoughtworks/images/illustration/brand/tw_illustration_24.jpg
Open-source maintainers face burnout and exploitation as permissive licenses enable corporate extraction, worsened by AI-generated code floods and fragmented re-implementations eroding trust. Sustainable solutions demand active ownership, supply chain audits, and corporate patronage to prevent ecosystem collapse.

The kids with phones are alright

A viral video of Scottish passengers confronting a predatory drunk man highlights effective real-world bystander action and victim support, contrasting with 2026 tech policies that disproportionately target youth and reflect elite-driven, paternalistic control over digital autonomy, rooted in classist power dynamics.

The Second Life of Sanskrit

https://media.openthemagazine.com/openthemagazine/2026-07-09/c235j692/Cover_Sanskrit.jpg?w=215&auto=format,compress&fit=max&q=85
Sanskrit is experiencing a revival through AI tools like Vagdhenu, social media initiatives, and online learning, blending ancient chants with modern tech. Young creators and educators are reshaping its cultural relevance, making it accessible and appealing beyond traditional contexts.

Casio FX-870P Emulator

https://raw.githubusercontent.com/urbancamo/fx870p-emulator/main/docs/screenshot.png
A browser-based emulator for the Casio FX-870P (also sold as the VX-4) pocket computer, written in TypeScript and Vue 3. The FX-870P was a 1986 programmable calculator running a Hitachi HD61700 CPU at 921 kHz with a 96×64 pixel LCD, 83-key keyboard, serial port, and optional MD-120 floppy drive. This emulator is a full port of the original Delphi reference implementation to the web platform. ...

Are we offloading too much of our thinking to AI?

https://substackcdn.com/image/fetch/$s_!NTK9!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Ffb548edd-f827-4d19-aa23-3bd68fdc980c_1456x1092.webp
Rising reliance on AI for decisions, from trivial to complex, risks eroding autonomy, as seen in Ken Liu's "Tilly" and the "Microphone Man" who offloads thinking to Claude. While AI boosts efficiency, balancing automation with critical thinking—like the author's sister exploring Portugal's history—prevents over-reliance and preserves intellectual engagement.

Kontigo (YC S24) Is Hiring (Head of Security)

https://bookface-images.s3.amazonaws.com/small_logos/29ea469fed3c60e9950eccadce6c5b7f8fcd6569.png
Principal Security Engineer/ Security Engineering Lead/ CISO "Men wanted for hazardous journey. Low wages, bitter cold, long hours of complete darkness. Safe return doubtful. Honour and recognition in event of success" You gotta be in founder mode.

Show HN: Opening lines of famous literary works

"The cold passed reluctantly from the earth, and the retiring fogs revealed an army stretched out on the hills, resting."

Microsoft Patches a Record 570 Security Flaws

https://krebsonsecurity.com/wp-content/uploads/2021/07/windupate.png
Microsoft released 570 security patches, triple last month's, driven by AI-aided discovery, including critical and zero-day flaws. Other firms like Adobe and Google are accelerating patch cycles due to AI, urging users to delay updates to avoid potential stability risks.

Launch HN: Agnost AI (YC S26) – Extract user feedback from agent conversations

https://agnost.ai/framer-assets/images/djbJRqx6BPCqS00nvdHgj84EeE.png
Agnost AI analyzes real user conversations to identify agent failures and user pain points, generating actionable fixes via reviewed PRs. It enhances agent performance by uncovering hidden issues, improving conversions, and accelerating improvements through collaboration.

Show HN: Juggler – an open-source GUI coding agent, by the creator of JUCE

https://opengraph.githubassets.com/5763f5012af91577995e1a8d5195c861a57016277ddc694cab98b45317340697/juggler-ai/juggler
Juggler offers a GUI workbench for visual, interactive control of LLM codebase workflows via branching session trees, plugins, and editable context. It supports cross-platform collaboration, local/remote access, and integrates with major models like Claude, OpenAI, and Gemini.

Measuring Input Latency on Linux: X11 vs. Wayland, VRR, and DXVK

https://marco-nett.de/img/blog/click2photon-board.jpg
Developed a hardware tool to measure Linux gaming latency, finding VRR and dxvk-low-latency reduce latency more than X11 vs Wayland differences. XWayland adds ~3ms latency, while optimizations like VRR and low-latency pacer improve performance by 0.2-0.5ms, with real gains in smoothing frame pacing.

Accretive Editing

https://justindfuller.com/image/programming/accretive-editing.png
Acctive editing in AI tools appends outdated info instead of replacing it, causing clutter (e.g., adding "not Amazon Bedrock" to new text). Solutions focus on instructing models to overwrite obsolete content rather than avoid accretion.

Human Canaries: Remembering the Munitionettes

Establishing a secure connection... Request ID: ad3dac334748195d1ad5ef367c1880d5

StubHub, CEO hit with ‘deceptive practices’ class action over mass scalping

https://i.cbc.ca/ais/1828decc-4c1c-455f-a98d-79e258477071,1783985197122/full/max/0/default.jpg?im=Crop%2Crect%3D%28392%2C981%2C5169%2C2907%29%3B
World StubHub and CEO Eric Baker face a $5M class-action lawsuit over deceptive practices, as his hedge fund resells millions in tickets via the platform while marketing it as a fan-to-fan marketplace. The suit follows World Cup ticket cancellations and consumer complaints, prompting U.S. and Canadian investigations.

Punch yourself in the face with reality

In the AI era, success hinges on confronting reality by focusing on meaningful progress rather than overbuilding with AI tools, which often lead to wasted efforts. The key is relentlessly pursuing truth and impact, not escaping into AI-generated illusions.