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

339 xenova 2026-07-14 18:08:44 UTC prismml.com Qwen Share 122

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.

Dependabot version updates introduce default package cooldown

45 woodruffw 2026-07-14 22:08:44 UTC github.blog Qwen Share 29

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.

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

32 1vuio0pswjnm7 2026-07-14 22:08:44 UTC bis.org Qwen Share 2

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.

The Tower Keeps Rising

296 cdrnsf 2026-07-14 17:08:44 UTC lucumr.pocoo.org Qwen Share 150

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

185 Synthetic7346 2026-07-14 18:08:44 UTC mindgard.ai Qwen Share 74

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

61 gnabgib 2026-07-14 20:08:44 UTC alexedwards.net Qwen Share 10

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)

667 MrVandemar 2026-07-14 10:08:44 UTC danq.me Qwen Share 419

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.

The largest available Minecraft world, totalling 15 TB

132 _____k 2026-07-14 18:08:44 UTC 2b2t.place Qwen Share 39

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.

The Estranged Worlds of J. G. Ballard

21 Caiero 2026-07-14 22:08:44 UTC lareviewofbooks.org Qwen Share 4

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.

Guardian Angels: LLM Personalization for Productivity and Security

46 andsoitis 2026-07-14 20:08:44 UTC gwern.net Qwen Share 3

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 ...

How to stop Claude from saying load-bearing

397 shintoist 2026-07-14 12:08:44 UTC jola.dev Qwen Share 455

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.

Microsoft Deletes User's 25-Year-Old Account with Thousands Spent on Games

76 HelloUsername 2026-07-14 22:08:44 UTC xcancel.com Qwen Share 24

User's Microsoft account and OneDrive were deleted despite proving ownership and account compromise, erasing 25 years of data, financial investments, and personal memories, sparking outrage over Microsoft's handling.

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

5 homarp 2026-07-14 22:08:44 UTC nature.com Prefix Share 0

A mathematical formula inscribed on a wall at the Maya site of Xultun in Guatemala has revealed the name of an important Maya mathematician-astronomer for the first time. Researchers suggest Sak Tahn Waax, or ‘White-Chested Fox’, was a scholar comparable with mathematical giants of the past. In a study published 14 July in the journal Antiquity1, Heather Hurst, an archaeologist at Skidmore ...

I'm a USB-C Maximalist

121 speckx 2026-07-14 16:08:44 UTC shkspr.mobi Qwen Share 217

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

90 backlit4034 2026-07-14 17:08:44 UTC thoughtworks.com Qwen Share 70

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 Second Life of Sanskrit

41 bookofjoe 2026-07-14 20:08:44 UTC openthemagazine.com Qwen Share 28

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.

Banter

10 surprisetalk 2026-07-14 22:08:44 UTC homosabiens.substack.com Qwen Share 1

Banter gently calls out minor social missteps in a safe, humorous context to defuse tension and reinforce group bonds. It signals acceptance of flaws, fostering security through normalized imperfection, though it risks misinterpretation as contempt if not balanced with trust.

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

0 2026-07-14 17:08:44 UTC ycombinator.com Full Share

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.

Are we offloading too much of our thinking to AI?

345 yenniejun111 2026-07-14 16:08:44 UTC artfish.ai Qwen Share 343

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.

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

37 laalshaitaan 2026-07-14 17:08:44 UTC agnost.ai Qwen Share 22

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: Opening lines of famous literary works

135 plicerin 2026-07-14 16:08:44 UTC verbaprima.com Full Share 78

"It was a pleasure to burn." Ray Bradbury · 1953

Human Canaries: Remembering the Munitionettes

8 Thevet 2026-07-14 22:08:44 UTC historytoday.com Full Share 0

Establishing a secure connection... Request ID: 03b78684dfcc02e64b38d6b773a80e7a

Accretive Editing

12 iamjfu 2026-07-14 21:08:44 UTC justindfuller.com Qwen Share 4

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.

LeMario: Training a JEPA World Model on Super Mario Bros

3 kevinjosethomas 2026-07-14 22:31:44 UTC benjamin-bai.com Prefix Share 0

I wanted to reproduce LeWorldModel, a small Joint-Embedding Predictive Architecture (JEPA) that learns world dynamics from pixels and actions. The original paper used it for reward-free planning in Push-T. But, since I loved video games, and at the same time wanted to learn more deeply about LeCun's JEPA architecture, I decided to write the whole architecture from scratch and train it on ...

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

333 hoechst 2026-07-14 17:08:44 UTC marco-nett.de Qwen Share 209

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.

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

155 julesrms 2026-07-14 14:08:44 UTC github.com/juggler-ai Qwen Share 77

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.

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

81 b112 2026-07-14 20:08:44 UTC cbc.ca Qwen Share 41

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.

Demis Hassabis has a plan to harness AI safely

130 asiergoni 2026-07-14 14:08:44 UTC twitter.com/demishassabis Full Share 171

This is a pivotal moment in human history. Artificial General Intelligence (AGI), a system that exhibits all the cognitive capabilities the brain has, is probably only a few short years away. When we...

The Agentic Loop: Three loops in a trench coat

67 btables 2026-07-14 15:08:44 UTC bobbytables.io Qwen Share 15

Agent loops in agentic systems consist of three layers: the inference loop (LLM-generated responses), tool loop (executing/modeling tool calls), and human loop (approval/steering). These loops enable functional AI agents by combining LLM reasoning, tool interaction, and human oversight.

Punch yourself in the face with reality

199 AdityaAnand1 2026-07-14 12:08:44 UTC adi.bio Qwen Share 96

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.