Description
More from Show
Show HN: Pyversity – Fast Result Diversification for Retrieval and RAG
Show HN: Pyversity – Fast Result Diversification for Retrieval and RAG Hey HN! I’ve recently open-sourced Pyversity, a lightweight library for diversifying retrieval results. Most retrieval systems optimize only for relevance, which can lead to top-k results that look almost identical. Pyversity efficiently re-ranks results to balance relevance and diversity, surfacing items that remain relevant but are less redundant. This helps with improving retrieval, recommendation, and RAG pipelines without adding latency or complexity.<p>Main features:<p>- Unified API: one function (diversify) supporting several well-known strategies: MMR, MSD, DPP, and COVER (with more to come)<p>- Lightweight: the only dependency is NumPy, keeping the package small and easy to install<p>- Fast: efficient implementations for all supported strategies; diversify results in milliseconds<p>Re-ranking with cross-encoders is very popular right now, but also very expensive. From my experience, you can usually improve retrieval results with simpler and faster methods, such as the ones implemented in this package. This helps retrieval, recommendation, and RAG systems present richer, more informative results by ensuring each new item adds new information.<p>Code and docs: github.com/pringled/pyversity<p>Let me know if you have any feedback, or suggestions for other diversification strategies to support!
Show HN: FLE v0.3 – Claude Code Plays Factorio
Show HN: FLE v0.3 – Claude Code Plays Factorio We're excited to release v0.3.0 of the Factorio Learning Environment (FLE), an open-source environment for evaluating AI agents on long-horizon planning, spatial reasoning, and automation tasks.<p>== What is FLE? ==<p>FLE uses the game Factorio to test whether AI can handle complex, open-ended engineering challenges. Agents write Python code to build automated factories, progressing from simple resource extraction (~30 units/min) to sophisticated production chains (millions of units/sec).<p>== What's new in 0.3.0 ==<p>- Headless scaling: No longer needs the game client, enabling massive parallelization!<p>- OpenAI Gym compatibility: Standard interface for RL research<p>- Claude Code integration: We're livestreaming Claude playing Factorio [on Twitch](<a href="http://twitch.tv/playsfactorio" rel="nofollow">http://twitch.tv/playsfactorio</a>)<p>- Better tooling and SDK: 1-line CLI commands to run evaluations (with W&B logging)<p>== Key findings ==<p>We evaluated frontier models (Claude Opus 4.1, GPT-5, Gemini 2.5 Pro, Grok 4) on 24 production automation tasks of increasing complexity.<p>Even the best models struggle:<p>- Most models still rely on semi-manual strategies rather than true automation<p>- Agents rarely define helper functions or abstractions, limiting their ability to scale<p>- Error recovery remains difficult – agents often get stuck in repetitive failure loops<p>The performance gap between models on FLE correlates more closely with real-world task benchmarks (like GDPVal) than with traditional coding/reasoning evals.<p>== Why this matters ==<p>Unlike benchmarks based on exams that saturate quickly, Factorio's exponential complexity scaling means there's effectively no performance ceiling. The skills needed - system debugging, constraint satisfaction, logistics optimization - transfer directly to real challenges.<p>== Try it yourself ==<p>>>> uv add factorio-learning-environment<p>>>> uv add "factorio-learning-environment[eval]"<p>>>> fle cluster start<p>>>> fle eval --config configs/gym_run_config.json<p>We're looking for researchers, engineers, and modders interested in pushing the boundaries of agent capabilities. Join our Discord if you want to contribute. We look forward to meeting you and seeing what you can build!<p>-- FLE Team
Show HN: SecretMemoryLocker – File Encryption Without Static Passwords
Show HN: SecretMemoryLocker – File Encryption Without Static Passwords I built SecretMemoryLocker (<a href="https://secretmemorylocker.com" rel="nofollow">https://secretmemorylocker.com</a>), a file encryption tool that generates keys dynamically from your answers to personal questions instead of using a static master password. This makes offline brute-force attacks much more difficult. Think of it as a password manager that meets mnemonic seed recovery, but without storing any sensitive keys on disk.<p>Why? I kept losing master passwords and wanted a solution that wasn't tied to a single point of failure. I also wanted to create a "digital legacy" that my family could access only under specific conditions. The core principle is knowledge-based encryption: the key only exists in memory when you provide the correct answers.<p>Status: * MVP is ready for Windows (.exe). * Linux and macOS support is planned. * UI is available in English, Spanish, and Ukrainian.<p>Key Features:<p>* No Static Secrets: No master password or seed phrase is ever stored. The key is reconstructed on the fly.<p>* Knowledge-Based Key Generation: The final encryption key is derived from a combination of your personal answers and file metadata.<p>* Offline Brute-Force Resistance: Uses MirageLoop, a decoy system that activates when incorrect answers are entered. Instead of decrypting real data, it generates an endless sequence of AI-created questions from a secure local database, creating an illusion of progress while keeping your real data untouched.<p>* Offline AI Generation Mode: Optional offline Q&A generator (prototype).<p>How It Works (Simplified):<p>1) Files are packed into an AES-256 encrypted ZIP archive.<p>2) A JSON key file stores the questions in an encrypted chain. Each subsequent question is encrypted with a key derived from the previous correct answer and the file's hash. This forces you to answer them sequentially.<p>3) The final encryption key for the ZIP file is derived by combining the hashes of all your correct answers. The key derivation formula looks like this:<p><pre><code> K_final = SHA256(H(answer1+file_hash) + H(answer2+file_hash) + ...) </code></pre> (Note: We are aware that a fast hash like SHA256 is not ideal for a KDF. We plan to migrate to Argon2 in a future release to further strengthen resistance against brute-force attacks.)<p>To encrypt, you provide a file. This creates two outputs: your_file.txt → your_file_SMLkey.json + your_file_SecretML.zip<p>To decrypt, you need both files and the correct answers.<p>Install & Quick Start: Download the EXE from GitHub Releases (no dependencies needed):<p><a href="https://github.com/SecretML/SecretMemoryLocker/releases" rel="nofollow">https://github.com/SecretML/SecretMemoryLocker/releases</a><p>Encrypt:<p><pre><code> SecretMemoryLocker.exe --encrypt "C:\docs\important.pdf" </code></pre> Decrypt:<p><pre><code> SecretMemoryLocker.exe --decrypt "C:\docs\important_SMLkey.json" </code></pre> I would love to get your feedback on the concept, the user experience, and any security assumptions I've made. Thanks!
No other tools from this source yet.