Published on March 26, 2026 at 18:01 CET (UTC+1)
Moving from GitHub to Codeberg, for lazy people (280 points by jslakro)
This article is a personal guide detailing the author's process of migrating repositories from GitHub to Codeberg. It highlights that the import process for issues, pull requests, and releases is straightforward via Codeberg's built-in tool, which preserves metadata. The author also discusses alternatives for hosting static sites (like GitHub Pages) on Codeberg's infrastructure, concluding the migration is less work than perceived for many projects.
My minute-by-minute response to the LiteLLM malware attack (56 points by Fibonar)
This is a chronological transcript of a developer using an AI coding assistant (Claude Code) to discover and respond to a sophisticated supply chain attack. The malware was hidden in a poisoned version of the LiteLLM Python package on PyPI, featuring credential theft and fork bombs. The article demonstrates how AI tooling accelerated both the forensic investigation and the public disclosure of the attack.
OpenTelemetry Profiles Enters Public Alpha (17 points by tanelpoder)
The article announces that the OpenTelemetry Profiles signal has entered a public Alpha release phase. It aims to establish a unified, vendor-neutral standard for continuous production profiling (like CPU/memory usage), analogous to OpenTelemetry's work with traces and metrics. The goal is to provide a common framework and data representation to help with performance troubleshooting and cost optimization industry-wide.
Personal Encyclopedias (634 points by jrmyphlmn)
The author describes the process of organizing a large collection of old, physical family photos and using conversations with their grandmother to annotate them with stories and context. This experience leads to a reflection on the value of creating "personal encyclopedias"—curated, narrative-rich digital archives of one's life and family history that go beyond simple photo storage to preserve meaning and memory.
European Parliament decided that Chat Control 1.0 must stop (507 points by lemoncookiechip)
This is a celebratory social media post announcing that the European Parliament has decided to let the "Chat Control 1.0" regulation expire. This means major tech companies (like Gmail, LinkedIn, Microsoft) will be required to stop the automated scanning and monitoring of private messages for all users in the European Union as of April 6, 2026, framed as a significant privacy victory.
Cory Doctorow: Interoperability Can Save the Open Web (114 points by janandonly)
Author and activist Cory Doctorow argues that mandated interoperability is the key to preserving competition and user freedom on the web. He contends that allowing users to modify and bridge between dominant, walled-garden platforms (like social media networks) can break monopolistic control, foster innovation, and protect the open internet from being dominated by a few large corporations.
From zero to a RAG system: successes and failures (199 points by andros)
The author shares a detailed, experiential account of building a production-grade Retrieval-Augmented Generation (RAG) system from scratch to query a vast, diverse internal document corpus using a local LLM. It chronicles the challenges faced, including technology selection, document chunking, embedding, filtering for speed, and handling specialized file formats, providing practical solutions and architectural insights.
End of "Chat Control": EU Parliament Stops Mass Surveillance in Voting Thriller (291 points by amarcheschi)
This article provides a detailed political report on the EU Parliament's close vote to definitively end the "Chat Control" mass surveillance derogation. It explains that the decision stops US tech corporations from indiscriminately scanning private messages and photos of EU citizens, emphasizing that this does not create a legal vacuum but rather paves the way for more targeted child protection measures.
Show HN: Claude skill that evaluates B2B vendors by talking to their AI agents (13 points by ogotlieb)
The article introduces an open-source Claude skill designed to automate and enhance the B2B software vendor evaluation process. The skill researches the buyer's company, asks domain-specific questions, sets constraints, and then directly engages with vendors' AI sales agents to gather information and provide an evidence-based score, aiming to streamline and objectify procurement.
My home network observes bedtime with OpenBSD and pf (46 points by ibobev)
This is a technical tutorial on using OpenBSD and its packet filter (pf) to create a custom home gateway/router. The primary goal is to automatically enforce an "Internet bedtime" on the network by blocking traffic on a schedule, while making exceptions for specific devices. The author also touches on using it for local DNS management, replacing a commercial router for greater control and learning.
Trend: AI is becoming a core tool in cybersecurity offense and defense. Why it matters: The LiteLLM attack article shows AI being used to generate sophisticated malware, while also enabling rapid defensive analysis and response. This creates an accelerating arms race where the speed of AI-augmented development is critical on both sides. Implication: Security for AI/ML pipelines (like supply chain vetting) is paramount. Developers must integrate AI-powered security tooling into their workflows, and the ability to use AI for forensic analysis is becoming an essential skill.
Trend: Practical RAG deployment faces significant data engineering hurdles. Why it matters: The "zero to RAG" article highlights that the major challenge isn't the LLM itself, but the preprocessing pipeline: chunking diverse document types, creating effective embeddings, and implementing fast, accurate retrieval. The model is often the easiest part. Implication: Success in applied AI projects is shifting from model-centric to data-infrastructure-centric. Investment in robust data preprocessing, embedding management, and vector search systems is crucial for production-grade applications.
Trend: Autonomous AI agents are moving beyond chatbots to become actionable evaluators and negotiators. Why it matters: The Claude vendor evaluation skill demonstrates AI agents that can perform multi-step, goal-oriented tasks—researching, formulating questions, and interacting with other autonomous agents (vendor sales bots) to gather evidence and make recommendations. Implication: This points toward a future of automated B2B and decision-making workflows. It raises questions about agent governance, bias in automated evaluations, and the need for standards in agent-to-agent communication.
Trend: Observability and profiling are becoming standardized and critical for AI system performance. Why it matters: The OpenTelemetry Profiles announcement underscores the industry's need to treat AI/ML pipelines and serving infrastructure as production software requiring detailed performance profiling. Understanding cost, latency, and bottlenecks is essential as AI computations grow. Implication: Adopting standardized observability tools will be necessary to manage the cost and reliability of AI systems. MLOps must integrate with broader DevOps observability practices using open standards.
Trend: Regulatory shifts on privacy and interoperability directly shape AI development environments. Why it matters: The end of EU Chat Control affects how training data from communications can be scanned, while interoperability mandates (as argued by Doctorow) could force platforms to expose APIs, changing how AI tools can interact with social networks and other walled gardens. Implication: AI developers must design for privacy-by-default and consider data sovereignty. They should also monitor interoperability regulations, which could open new avenues for AI-driven tools that bridge platform silos or create new competitive interfaces.
Trend: The infrastructure supporting AI development is fragmenting, with a push towards open source and decentralization. Why it matters: The migration from GitHub to Codeberg reflects a broader trend of seeking vendor-neutral, open-source platforms. This desire for control and ethical alignment extends to the infrastructure used for hosting code, data, and potentially AI models themselves. Implication: While centralized platforms (GitHub, major cloud providers) dominate, there is growing momentum for decentralized alternatives. This could influence where open-source AI projects are hosted, collaborated on, and deployed, potentially increasing resilience and reducing lock-in.
Analysis generated by deepseek-reasoner