Published on June 07, 2026 at 06:00 CEST (UTC+2)
Valve P2P networking broken for more than 2 months (24 points by babuskov)
Valve P2P networking broken for more than 2 months
This GitHub issue reports a persistent problem with Valve's Steam Networking P2P system, primarily affecting players in Israel and possibly other Middle Eastern countries. Since mid-March, players in Israel experience unusually high ping (around 120ms) when playing PC-to-PC, while cross-play with consoles works normally (5-10ms). The issue has been ongoing for over two months without a fix from Valve or game developers, leaving the community frustrated. It highlights how a single backend networking component can disrupt multiplayer gaming for an entire region.
Field of clones: How horse replicas came to dominate polo (27 points by gscott)
Field of clones: How horse replicas came to dominate polo
The article explores the growing use of cloned horses in professional polo, where genetically identical replicas of champion ponies are now common. It discusses the science and ethics behind cloning, the regulatory landscape, and how this technology has shifted the sport's economics and competition dynamics. Cloned horses offer consistency in performance but raise questions about genetic diversity, animal welfare, and the definition of a "natural" athlete.
Tokenomics: Quantifying Where Tokens Are Used in Agentic Software Engineering (40 points by Anon84)
Tokenomics: Quantifying Where Tokens Are Used in Agentic Software Engineering
This paper analyzes token consumption in LLM-based multi-agent (LLM-MA) systems performing software engineering tasks. Using the ChatDev framework with a GPT-5 reasoning model, the authors track token usage across design, coding, review, testing, and documentation phases. They find significant variation in token distribution, with reasoning tokens dominating certain stages. The work aims to help developers predict costs and environmental impact when deploying agentic coding systems.
Harness engineering: Leveraging Codex in an agent-first world (112 points by pramodbiligiri)
Harness engineering: Leveraging Codex in an agent-first world
OpenAI describes a new paradigm called "harness engineering," where developers design systems that treat AI models like Codex as autonomous agents rather than simple copilots. The approach focuses on building robust scaffolding, task decomposition, and error recovery loops around the model. This shift implies a need for new engineering practices—testing agent behavior, managing context windows, and ensuring reliable multi-step execution.
Ntsc-rs – open-source video emulation of analog TV and VHS artifacts (293 points by gregsadetsky)
Ntsc-rs – open-source video emulation of analog TV and VHS artifacts
This free, open-source tool accurately emulates analog NTSC video and VHS tape degradation using real-world signal processing algorithms, not just visual filters. Written in Rust, it runs in real time at high resolutions and is available as a standalone app, web tool, and plugin for video editing software. It appeals to retro enthusiasts and filmmakers seeking authentic analog aesthetics without physical hardware.
Public Domain Image Archive (55 points by davidbarker)
Public Domain Image Archive
A curated online collection of over 11,000 out-of-copyright images, free to browse, download, and reuse. The archive is searchable by artist, century, style, theme, or tag, and new images are added weekly. It serves as a valuable resource for educators, designers, and historians needing unrestricted visual material.
Introducing Boron Buckyballs: Theory that B80 cages can’t be made is disproved (50 points by crescit_eundo)
Introducing Boron Buckyballs: Theory that B80 cages can’t be made is disproved
Chemists have experimentally observed a boron buckminsterfullerene (B80) for the first time, refuting a long-standing theoretical prediction that such a cage structure was impossible. The soccer-ball-shaped molecule resembles carbon C60 but is made entirely of boron. This breakthrough opens new avenues in nanotechnology and materials science, as boron buckyballs may have unique electronic and chemical properties.
Show HN: Oproxy – inspect and modify network traffic from the browser (14 points by sauravrao637)
Show HN: Oproxy – inspect and modify network traffic from the browser
Oproxy is an open-source local proxy (HTTP, HTTPS, SOCKS5) that lets developers intercept, replay, and modify network traffic for testing. It supports viewing request/response details, timing, and headers, and can be used with browsers, CLI tools, mobile apps, and Docker containers. The project includes a custom CA for HTTPS inspection and is built in Rust with a focus on performance and simplicity.
An Ohio Valley 100k-Watt FM Signal Is Severed in Broad Daylight – Radio World (72 points by pkaeding)
An Ohio Valley 100k-Watt FM Signal Is Severed in Broad Daylight
A thief cut the main transmission line of 100,000-watt FM station WDGG ("The Dawg") in Kentucky, knocking the station off the air. The suspect, Paul Crisp, severed the cable on a bridge leading to the tower and was later caught. The incident highlights the vulnerability of broadcast infrastructure to copper theft and the risks taken by thieves tampering with high-power equipment.
Biohub releases a world model of protein biology (14 points by gmays)
Biohub releases a world model of protein biology
Biohub introduced an open "world model" for protein biology, comprising three models: ESMC (for sequence-structure-function mapping), ESMFold2 (for structure prediction), and ESM Atlas (a database of predicted protein structures). These models can design new protein binders with therapeutic-level affinity, tested in lab experiments. The release aims to democratize protein design for drug discovery and biological research.
Token consumption is becoming a critical metric for agentic AI systems
The "Tokenomics" paper reveals that LLM-based multi-agent systems consume tokens very differently across software engineering phases—reasoning tokens, not just input/output, dominate in design and testing. This matters because cost and latency are major barriers to deploying agentic coding assistants at scale. Developers should instrument their pipelines to track token usage per phase, and researchers need to optimize reasoning paths to reduce waste.
The shift from copilot to agent-first engineering demands new validation methods
OpenAI's "harness engineering" concept signals that building reliable AI agents requires more than better models—it requires robust scaffolding, error handling, and behavioral testing. This trend matters because agents act autonomously, making mistakes harder to catch. Implications: expect growth in agent observability tools (like Oproxy for network-level debugging) and formal verification methods for multi-step agent workflows.
Open-source protein foundation models are democratizing drug discovery
Biohub's release of ESMC, ESMFold2, and ESM Atlas as open models brings cutting-edge protein design to academic and small-company labs. This matters because proprietary models (e.g., AlphaFold) limit access; open models accelerate innovation in therapeutic design and synthetic biology. Implication: we will see a wave of open "world models" for other scientific domains (e.g., materials, genomics) that combine sequence, structure, and function.
Agentic software engineering is moving from research to practical cost analysis
The "Tokenomics" paper is a concrete step toward understanding the economic and environmental trade-offs of AI-driven development. As companies adopt agentic coding tools, they need transparent cost models—not just model accuracy. This trend will drive the creation of "agent efficiency benchmarks" and pricing models that account for reasoning depth, task complexity, and inference hardware.
AI's role in simulating and emulating physical phenomena is expanding
While ntsc-rs is not AI-based, its approach (using accurate algorithms to model analog video) parallels how AI is used to simulate complex physical systems (e.g., protein folding, weather). The broader trend is combining algorithmic domain knowledge with learned components. For AI/ML, this suggests hybrid models—where traditional simulation is augmented by neural networks—will become more common in fields like graphics, physics, and biology.
The need for network-level debugging tools for AI agents is growing
Oproxy, though general-purpose, fills a gap as AI agents increasingly interact with web APIs, databases, and external services. When an agent misbehaves, developers need to inspect exactly what requests it made and what responses it received. This trend points to a new category of "agent debugging proxies" that log and replay agent-server interactions, essential for reliability in production agent systems.
Cross-domain model deployment (e.g., from text to proteins) highlights the value of transferable representations
Biohub's ESM models leverage transformer architectures originally developed for natural language; the "world model" approach treats protein sequences as a language. This underscores that foundation models trained on massive biological data can discover latent "grammar" rules—paralleling how LLMs learn syntax. The implication is that future breakthroughs may come from applying language model techniques to other symbolic domains (chemistry, genomics, materials), accelerating scientific discovery.
Analysis generated by deepseek-reasoner