Published on June 21, 2026 at 18:01 CEST (UTC+2)
(How to Write a (Lisp) Interpreter (In Python)) (15 points by tosh)
How to Write a (Lisp) Interpreter (In Python)
This article by Peter Norvig presents a concise tutorial on building an interpreter for the Scheme dialect of Lisp using Python 3. The goal is to demonstrate the fundamental principles of computer language interpretation, which Norvig likens to "Maxwell’s Equations of Software." The piece covers both the syntax and semantics of Scheme programs, showing how to evaluate expressions step by step. It emphasizes that understanding compilers and interpreters is essential to understanding how computers work.
Beyond All Reason (Free Total Annihilation Inspired RTS) (226 points by mosiuerbarso)
Beyond All Reason (Free Total Annihilation Inspired RTS)
This is a real-time strategy game that redefines the genre with fully simulated ballistics, explosion physics, and terrain deformation. Units, projectiles, and explosions are simulated in real-time, allowing for battles involving thousands of units. The game emphasizes strategic use of terrain, radar, and nuclear warfare, with over ten different unit classes. It is free to play and currently in alpha, with servers occasionally unstable.
Who Owns Your ATProto Identity? Hint: It's Probably Not You (66 points by kevinak)
Who Owns Your ATProto Identity? Hint: It's Probably Not You
The article exposes a critical flaw in the ATProto identity system (used by Bluesky): the Personal Data Server (PDS) holds both the signing key and rotation key for a user’s decentralized identifier. This means the PDS operator can impersonate the user, posting and interacting as them, and the activity is cryptographically indistinguishable from the user’s own. Unlike traditional platforms, ATProto’s design gives the PDS full control over all data across multiple apps built on the protocol, not just Bluesky posts.
Commodore Made a Digital Detox Phone That Isn't Dumb (39 points by Audiophilip)
Commodore Made a Digital Detox Phone That Isn't Dumb
Commodore is launching the Callback 8020, a flip phone designed for digital detox. It runs Sailfish OS (Linux-based) and can run essential Android apps like Uber, WhatsApp, and Spotify, but deliberately blocks distracting apps such as social media, browsers, and email. The front screen shows only the time, date, and battery, with no notifications. The phone aims to provide a modern smartphone experience without the attention-hogging features.
The case against geometric algebra (2024) (93 points by Hbruz0)
The case against geometric algebra (2024)
The author argues that while Geometric Algebra (GA) has merit, it is often oversold as a replacement for linear algebra and multivariable calculus. GA’s concepts of multivectors and the wedge product are useful, but the subject has notable flaws and the community tends to ignore critical feedback. The article suggests that GA needs serious refinement and that advocates should address its shortcomings rather than promoting it uncritically.
David Ahl's Basic Computer Games Ported to C (34 points by theanonymousone)
David Ahl's Basic Computer Games Ported to C
This GitHub repository contains a port of the classic GW-BASIC games from David Ahl’s “BASIC Computer Games” and “More BASIC Computer Games” to the C programming language. The conversion was done using an automated tool (bas2c.py) and includes games from the original books. The code is provided as-is, with a warning that it has not been fully tested or debugged, making it a potential learning tool for debugging and retro computing enthusiasts.
Google Hits 50% IPv6 (301 points by barqawiz)
Google Hits 50% IPv6
Google’s measurements show that IPv6 has reached 50% global adoption among users accessing Google services, a significant milestone for Internet infrastructure. However, the article from APNIC notes that adoption is uneven across economies—countries like India, Vietnam, and Saudi Arabia have very different curves. APNIC’s own data shows a lower global figure (42%), highlighting discrepancies in measurement methodologies.
A 3D voxel game engine written in APL (105 points by sph)
A 3D voxel game engine written in APL
This project is a highly experimental voxel game engine written in Dyalog APL and SDL3, created as a bet to see if APL notation could simplify game development. It uses Vulkan, DirectX12, or Metal for graphics and supports basic movement, jumping, and block placement. The repository is buggy and intended as a proof-of-concept, requiring Dyalog APL 20.0 and a compatible C compiler.
Loupe – A iOS app that raises awareness about what native apps can see (422 points by Cider9986)
Loupe – An iOS app that raises awareness about what native apps can see
Loupe is an iOS/iPadOS privacy app that reads real values from public iOS APIs and displays them to the user. It demonstrates how device fingerprinting works by showing raw data that any third-party app can access, such as system settings, hardware identifiers, and other metadata. The goal is to educate users about how trackers can recognize them without needing personal information like name or email.
Two Qwen3 models on one DGX Spark: the residency math (38 points by devashish86)
Two Qwen3 models on one DGX Spark: the residency math
The author describes experiments running multiple large language models (Qwen3 variants) simultaneously on a single NVIDIA DGX Spark GPU. They compare Ollama (which monopolizes GPU memory) with vLLM, which offers better memory management through PagedAttention and per‑process memory budgets. The article details configuration pitfalls, such as the gpu_memory_utilization trap, and demonstrates how to coexist a heavy reasoning model and a fast response model on one GPU for a multi‑agent system.
Rise of local and multi‑model deployments for AI agents
The DGX Spark experiment (article 10) illustrates a growing trend: running multiple LLMs on a single local GPU to support agentic workflows. With frameworks like vLLM supporting PagedAttention and fine‑grained memory control, it becomes feasible to co‑host a heavy reasoning model and a lightweight response model. This matters because it reduces latency, costs, and reliance on cloud APIs, enabling more autonomous and privacy‑preserving AI systems. Actionable takeaway: developers should evaluate vLLM over Ollama for multi‑model scenarios and pay attention to gpu_memory_utilization settings to avoid out‑of‑memory errors.
Privacy and transparency tooling as a counter to AI surveillance
The Loupe iOS app (article 9) and the ATProto identity critique (article 3) both highlight growing concerns about data exposure in distributed systems. Loupe reveals how device fingerprinting works, while the ATProto article warns that centralized key management can be abused. For AI/ML, this is critical because models often rely on large‑scale data collection—and decentralized identity (like ATProto) is being promoted as a solution. The insight: without proper cryptographic safeguards, “decentralized” systems can centralize power over user identity, which undermines trust in AI‑powered applications. Developers building on such protocols must ensure users retain control of their keys.
Educational resources for foundational AI/ML concepts remain valuable
The Lisp interpreter tutorial (article 1) and the BASIC games C port (article 6) represent hands‑on learning tools relevant to AI/ML. Understanding interpreters and compilers is foundational for building domain‑specific languages (DSLs) used in ML frameworks (e.g., TensorFlow graphs, PyTorch JIT). Similarly, debugging legacy code is a transferable skill. The trend: despite advanced AI tools, the demand for low‑level programming understanding persists, especially for optimizing inference engines and custom model execution.
IPv6 adoption milestone impacts AI/ML infrastructure
IPv6 reaching 50% (article 7) has implications for distributed AI/ML training and inference. Larger address spaces enable more direct peer‑to‑peer connections for data‑parallel training or model serving across edge devices. The uneven global adoption (e.g., high in India, low elsewhere) means developers must plan for dual‑stack support in cloud deployments and data pipelines. As AI workloads become more geographically distributed, IPv6 readiness becomes a practical bottleneck.
Experimental programming languages and environments drive innovation in game AI
The voxel game engine in APL (article 8) and the Geometric Algebra critique (article 5) both point to a hunger for mathematical and notational clarity in simulation and graphics. APL’s array‑oriented nature could inspire new ways to represent neural networks or physics simulations. Meanwhile, geometric algebra debates show that the AI/ML community often adopts mathematical tools (like matrix algebra) without fully exploring alternatives—yet novel notations can lead to more efficient algorithms for tasks like 3D scene understanding or robotic control.
Hardware‑software co‑optimization for local LLMs is accelerating
The DGX Spark example (article 10) is part of a broader trend: specialized hardware (like NVIDIA’s DGX lineup) combined with open‑source software (vLLM, llama.cpp) to democratize local LLM serving. The “residency math” (memory budgeting, context window trade‑offs) is becoming a core skill for AI engineers. This matters because it allows smaller teams and individuals to run competitive models offline, reducing API costs and improving latency for interactive applications. The actionable insight is to invest in understanding GPU memory management (KV cache, paged attention, quantization) to maximize throughput on limited hardware.
Digital detox and AI‑free interfaces as a counter‑trend
The Commodore phone (article 4) represents a growing backlash against the attention‑economy design of modern smartphones—including AI‑powered recommendation engines and notification systems. For AI/ML, this signals a market for “constrained AI” that assists without distracting. Developers should consider building interfaces that respect user attention (e.g., no‑notification modes, on‑demand inference) rather than always‑on, engagement‑maximizing AI. This aligns with privacy trends and regulatory pressures (e.g., EU Digital Services Act).
Analysis generated by deepseek-reasoner