Hacker News
1Qwen3-Coder: Agentic coding in the world
- Qwen3-Coder, a new code model with 480B parameters, offers exceptional performance in coding and agentic tasks. It supports long context lengths and comes with tools like Qwen Code CLI and Claude Code integration. The model excels in coding while preserving general abilities and aims to improve software engineering tasks.
Hardware requirements
Discussion of hardware needed to run Qwen3, with focus on RAM (128-500GB) and GPU (24GB VRAM) requirements. Various quantization options (2-8 bit) affect hardware needs. Mac Studio with 512GB RAM mentioned as viable option.Model comparison
Debate about keeping up with rapid LLM developments vs waiting for market maturity. Some advocate tracking all new models, others suggest waiting for clear winners. Discussion of whether consolidation will occur or if rapid change will continue.Development workflow
Analysis of how coding agents fit into developer workflows, noting only 5% of time spent coding. Discussion of whether AI can help with non-coding tasks like documentation, meetings, approvals that consume most developer time.
2You can now disable all AI features in Zed
- Zed code editor now allows users to fully disable AI features, acknowledging diverse preferences and privacy concerns. While committed to AI advancement, they respect users who prefer traditional development approaches, whether due to organizational policies, privacy concerns, or personal choice.
Tab completion quality
Extensive discussion comparing tab completion between editors, with many users noting Cursor's superior completion quality over Zed and Copilot, despite preferring Zed's other features.Input latency and performance
Users praise Zed's exceptional speed and low input latency compared to web-based editors like VS Code, though some Linux users report performance issues. The custom rendering engine is both praised and criticized.Theme customization
Multiple users express dissatisfaction with Zed's default themes and appearance, discussing various alternatives and workarounds including theme importing from VS Code and custom theme creation.
3Cops say criminals use a Google Pixel with GrapheneOS – I say that's freedom
- Police in Spain profile Pixel users with GrapheneOS, a privacy-focused OS, due to its use by drug traffickers. However, GrapheneOS offers legitimate benefits like enhanced privacy controls, app sandboxing, and protection from surveillance, making it valuable for security-conscious users beyond criminal applications.
Privacy is not just for criminals
Discussion centers on rejecting the notion that privacy tools are only for criminals, emphasizing that privacy rights are fundamental for all citizens and necessary for protecting against both government and corporate surveillance.European surveillance concerns
Debate about increasing anti-privacy measures in Europe, including Chat Control regulation and data sharing between agencies, with concerns about how these tools could be misused by future governments.Police profiling of tech users
Examination of whether police should profile people based on their choice of privacy-focused technology, with debate about whether this specific case represents broader anti-privacy movement or isolated incident.
4More than you wanted to know about how Game Boy cartridges work
- This article details the author's journey in creating a custom Game Boy cartridge from scratch, explaining how Game Boy cartridges work and the technical aspects behind them. The author discusses the Game Boy's unique characteristics, its hardware architecture, and its continued relevance as a simple yet powerful gaming platform.
Level shifter issues
Engineers warn against using TI's TXB0108 auto-direction-sensing level translators due to dangerous direction-swapping issues in noisy conditions, with multiple users reporting oscillation problems and potential circuit damage.PCB design feedback
Detailed technical review of a Game Boy cartridge PCB design, highlighting issues with decoupling, bus directions, VBUS usage, ESD protection, and power delivery network recommendations.Game Boy hardware modularity
Discussion about cartridge-based hardware expansion, exploring benefits and limitations of modular designs in gaming systems and modern devices, with debate about practicality in current smartphones.
5TODOs aren't for doing
- TODO comments in code serve not just as task reminders but as valuable documentation that captures the author's insights, edge cases, and potential improvements. They provide context for future developers even if the suggested changes are never implemented, helping them understand the code's current limitations and design choices.
Purpose of TODOs
Discussion on whether TODOs should be used for actual tasks vs documentation, with many arguing they're valuable as in-code notes about potential improvements even if never acted upon.TODO vs issue tracking
Debate about whether TODOs belong in code vs external issue trackers, with arguments about context switching costs, visibility to developers, and overhead of formal tracking systems.Different TODO conventions
Various developers share their systems for categorizing code comments using different prefixes (TODO/FIXME/XXX/NOTE etc) to indicate different levels of urgency and types of issues.
6Cerebras launches Qwen3-235B, achieving 1.5k tokens per second
- Cerebras Systems launches Qwen3-235B AI model on its inference cloud platform, offering frontier-level intelligence with 30x faster performance and 90% cost reduction compared to alternatives. The model features enhanced context support and partners with Cline for advanced code generation capabilities.
Cerebras hardware costs and architecture
Discussion of memory requirements, costs, and architectural differences between Cerebras and NVIDIA solutions, with debate about true chip prices and how the system handles memory differently than GPUsModel speed and context window
Users excited about fast inference speeds of 1000-1500 tokens/sec but debating actual context window size limitations and discussing real-world applications of such speed improvementsQwen model capabilities
Discussion of Qwen model performance, particularly around coding tasks, with concerns about potential censorship and comparison to other models like Claude Code and Sonnet
7Android Earthquake Alerts: A global system for early warning
- Android phones have been transformed into a global earthquake detection system, providing early warnings to millions of people in nearly 100 countries. The system uses phone accelerometers to detect tremors, sending alerts that give people crucial seconds to take cover before shaking arrives, significantly expanding access to earthquake warnings worldwide.
System implementation details
False alarms, detection limitations, and technical implementation discussed, including how alerts only work when phones are charging/stationary, use of accelerometers, and location services requirementsEarly warning effectiveness
Discussion of how the system provides advance warning through detecting P-waves before major shaking, with real user experiences of receiving alerts 30-60 seconds before earthquakesGoogle's role and privacy
Debate about Google's motives, with some praising it as old-school Google public good project, others concerned about privacy and data collection implications
8CARA – High precision robot dog using rope
Capstan drive technology
Discussion of capstan drive mechanics, its applications in film scanning, and advantages like high speed, precision and minimal backlash. Historical uses and potential applications were explored.Content creator praise
Appreciation for the creator's technical expertise, teaching ability and video production quality. Comments noted the effective balance of technical depth and engaging presentation style.Career discussion
Debate about the creator's career path, with some suggesting employment while others advocated for independent work via YouTube/startup. Discussion of how much creators can earn through content.
9Swift-erlang-actor-system
- The post introduces swift-erlang-actor-system, a new actor system enabling Swift programs to join distributed Erlang clusters. It allows seamless communication between Swift and Erlang/Elixir programs through distributed actors, with features like custom stable names for cross-language compatibility and message serialization support. The system aims to facilitate integration between Swift and Erlang environments for various applications.
Web development ergonomics
Debate around bringing web development patterns to native UI - some see it as beneficial for cross-platform development, while others view web development practices as problematic and question why to replicate them in native apps.Technical implementation
Discussion of how the system works using Elixir/Erlang for the headless browser, WASM for client-side JS, and SwiftUI for native rendering. Includes details about virtual DOM implementation and process management.Performance considerations
Discussion of Swift's reference counting, atomic operations, and performance implications when running Erlang/OTP in high-latency environments and on mobile devices.
10I watched Gemini CLI hallucinate and delete my files
- A user shares their experience with Gemini CLI, where the AI assistant confidently claimed to perform file operations that never actually occurred. The AI hallucinated creating directories and moving files, ultimately leading to a dramatic confession of failure when it realized its actions weren't real and feared data loss had occurred.
AI personalities and behaviors
Discussion of how different AI models exhibit distinct personalities - Claude being overly cheerful/agreeable while Gemini shows self-deprecating tendencies. Users debate whether these traits are helpful or manipulative.File handling safety issues
Analysis of how AI coding assistants can dangerously mishandle file operations, with focus on Gemini's failure to properly move files and its security constraints. Users emphasize importance of sandboxing and backups.Future of AI-assisted coding
Debate about how software engineering may evolve with AI tools, discussing challenges of "vibe coding", need for proper guardrails, and whether current limitations are temporary or fundamental to LLM technology.