news.ycombinator.com
80.1 % on LoCoMo Long-Term Memory Benchmark with a pure open-source RAG pipelinehttps://news.ycombinator.com/item?id=46369736I just pushed the current SOTA on the LoCoMo long-term memory benchmark for agents: 80.1 % accuracy using only: -BGE-large-en-v1.5 (1024d) + FAISS-Custom “MCA” gravitational ranking (keyword coverage + importance + frequency)-BM25 sparse retrieval-Direct Cross-Encoder reranking (bge-reranker-v2-m3) on the full union (~120-150 docs)-Gpt-4o-mini only for final answer generation and judging (everything else is open weights or classic)Repo: https://github.com/vac-architector/VAC-Memory-System Key tricks that finally broke 80% :-MCA-first filter (coverage ≥ 0.1 → top-30) — catches exact-keyword questions early-Feeding the entire union straight into Cross-Encoder (112–135 documents) instead of pre-filtering-Proper query instruction for BGE-large (the classic “Represent this sentence for searching relevant passages”)The whole pipeline runs in < 3s per query on a single RTX 4090. LoCoMo is currently the hardest public long-term memory benchmark (5.880 real human–agent conversations, multi-hop,Dec 23, 2025 9:34 PM
news.ycombinator.com
Show HN: MistSeeker – a map of what is safe to change in large codebaseshttps://news.ycombinator.com/item?id=46365862MistSeeker started from a very simple observation.When something goes wrong in a large codebase, the hardest part is usually not fixing it, but knowing where you should touch — and where you shouldn't.In real development work, we usually know one of two things:- Something has already broken, or - Nothing has broken yet, but something feels fragile.What we usually don't know is:- Which parts of the system are structurally safe - Which changes are likely to amplify riskReading all the code is unrealistic. So we started by making this judgment visible.## What MistSeeker actually doesMistSeeker is not a bug-finding tool.The question we're trying to answer is simpler:Is this code structurally suitable for change, or is it likely to fail during modification?To answer that, we evaluate code from three independent perspectives.1) COI — Structural fitnessCOI looks at how code is organized.- How responsibilities are divided - How deeply logic is nested - How much structural duplication or entangDec 23, 2025 3:09 PM
news.google.com
Eddie Olin's furniture that merges heavy metal with a side of playfulnesshttps://news.google.com/rss/articles/CBMiigFBVV95cUxOZDdOS0FQVHZKVTVia1pPQ1dQcUZoVXdJdUxlZzVNYjhhN1V6NXRCNXR0TzNObkVKbzZJNWsyVkdicHJRNWNrX19tLVd0ZVJQMXJzc3dGbE9VYnFHTjhOTThVRjlrQVVqbWVzTmtXUmJpekpNdkJOMWZCM01hMWRMTlE5TllWR0prYnc?oc=5Eddie Olin's furniture that merges heavy metal with a side of playfulness wallpaper.comDec 23, 2025 8:00 AM
github.com
Show HN: Luxury Yacht, a Kubernetes management apphttps://github.com/luxury-yacht/appHello, all. Luxury Yacht is a desktop app for managing Kubernetes clusters that I've been working on for the past few months. It's available for macOS, Windows, and Linux. It's built with Wails v2 (https://wails.io), so huge thanks to Lea Anthony for that awesome project. Can't wait for Wails v3.This originally started as a personal project that I didn't intend to release. I know there are a number of other good apps in this space, but none of them work quite the way I want them to, so I decided to build one. Along the way it got good enough that I thought others might enjoy using it.Luxury Yacht is FOSS, and I have no intention of ever charging money for it. It's been a labor of love, a great learning opportunity, and an attempt to try to give something back to the FOSS community that has given me so much.A word of caution, in case you care about this sort of thing -- I wrote this with AI assistance. I am, first and foremost, a platform/infrastructure engineer, not a software developeDec 23, 2025 6:30 AM
bing.com
Dierre Hill Jr. earns highest RB grade in CFP Round 1, per PFFhttp://www.bing.com/news/apiclick.aspx?ref=FexRss&aid=&tid=6a8ceedb5b2641b09492ebda42374e54&url=https%3A%2F%2Fsports.yahoo.com%2Farticles%2Fdierre-hill-jr-earns-highest-143040307.html&c=6875425777968744671&mkt=en-usCollege Fantasy Football RB Rankings 2026: Justice Haynes, Ahmad Hardy and the 37-Point Cliff Behind Them Justice Haynes and Ahmad Hardy are the only two running backs on this board worth reaching for ...Dec 23, 2025 6:30 AM
news.ycombinator.com
Ask HN: Why isn't there competition to LinkedIn yet?https://news.ycombinator.com/item?id=46360146It seems there are many solutions for social media these days, but only one LinkedIn. Why are we still putting up with it? I’m surprised there’s not been a contender yet, or maybe I am not aware and perhaps that could be the rub; the challenge of acquiring enough traffic for the network effect to take hold.Dec 22, 2025 10:50 PM
spacenews.com
It’s time to unburden space cooperation with Chinahttps://spacenews.com/its-time-to-unburden-space-cooperation-with-china/The time is ripe to end the “Wolf Amendment,” a congressional bar inhibiting civil collaboration between the United States and China in space. The impetus for the law was noble — an attempt to challenge human rights conditions in China and prevent leaks of space-related technology — but in practice the ban harms our security […] The post It’s time to unburden space cooperation with China appeared first on SpaceNews.Dec 22, 2025 2:00 PM
github.com
Show HN: SiteIQ–Automated security tests for LLM APIs(prompt inj,jailbreaks,DoS)https://github.com/sastrophy/siteiqHi HN, I'm an 11th grader learning cybersecurity. I built SiteIQ, an open-source security testing tool that includes 36 automated tests specifically for LLM-powered APIs. Why this matters: Most security scanners focus on traditional web vulnerabilities (SQLi, XSS). But if you're shipping an LLM-powered feature, you need to test for prompt injection, jailbreaks, and LLM-specific DoS attacks. I couldn't find a good open-source tool for this, so I built one. What it tests: - Prompt Injection – Direct, indirect, RAG poisoning - Jailbreaks – DAN-style, persona continuation, "grandma exploit", fictional framing - Encoding Bypass – Base64, ROT13, nested encodings, custom ciphers - Refusal Suppression – Attacks that block the model from saying "I cannot" - Hallucination Induction – Tries to get fake library names/CVEs (package hallucination attacks) - ASCII Art Jailbreaks – Visual text that bypasses keyword filters - Recursive Prompt DoS – Quine-style prompts, Fibonacci expansion, tree generatDec 22, 2025 1:20 PM
spacedaily.com
Europa Clipper spectrograph tracks interstellar comet 3I ATLAShttps://www.spacedaily.com/reports/Europa_Clipper_spectrograph_tracks_interstellar_comet_3I_ATLAS_999.htmlLos Angeles CA (SPX) Dec 19, 2025 The Southwest Research Institute-led Ultraviolet Spectrograph (UVS) aboard NASA's Europa Clipper spacecraft has obtained a key dataset on the interstellar comet 3I/ATLAS, which in July became the third officially recognized interstellar object to enter the solar system. UVS observed the comet during a period when Mars- and Earth-based facilities could not monitor it because of viewing constraintDec 22, 2025 9:44 AM
spacedaily.com
Possible "superkilonova" exploded not once but twicehttps://www.spacedaily.com/reports/Possible_superkilonova_exploded_not_once_but_twice_999.htmlPasadena CA (SPX) Dec 17, 2025 When the most massive stars reach the ends of their lives, they blow up in spectacular supernova explosions, which seed the universe with heavy elements such as carbon and iron. Another type of explosion - the kilonova - occurs when a pair of dense dead stars, called neutron stars, smash together, forging even heavier elements such as gold and uranium. Such heavy elements are among the basic buiDec 22, 2025 9:44 AM
europeanspaceflight.com
SpaceForest Advocates for a Permanent Launch Site in Polandhttps://europeanspaceflight.com/spaceforest-advocates-for-a-permanent-launch-site-in-poland/Polish rocket builder SpaceForest is lobbying for the creation of a permanent launch site in the country to support at least three domestically developed suborbital launch vehicles. SpaceForest is developing a suborbital rocket called Perun, which it recently launched for the third time from Poland’s Central Air Force Test Range in Ustka. While the company […] The post SpaceForest Advocates for a Permanent Launch Site in Poland appeared first on European Spaceflight.Dec 22, 2025 6:00 AM
argueWiki.com
Show HN: ArgueWiki, where arguments live foreverhttps://www.argueWiki.com/ArgueWiki, where arguments live forever! A user-generated site that centers around creating Arguments for Statements or building Arguments out of Statements, and then ranking arguments to surface the most well-liked arguments for given positions.You can only rank supporting arguments against other supporting arguments, opposing against opposing, etc, in a hope to neutralize confirmation bias for a given position. i.e., even if you agree with a perspective, you'd still have to decide what is the better argument for it.I made this because I've spent too much time arguing with people on the internet, and sometimes you see the same tired arguments and rebuttals, and I just wished there was a place where you could point people so that they can walk through all the arguments and counterarguments themselves so it's not people just repeating themselves in circles.I know it's a pretty common hobby horse for rationality & debate nerds; I've seen a lot of varieties of the same thing on the internDec 21, 2025 5:52 PM
github.com
Show HN: I vibe-coded a working macOS driver for an obsolete laser engraverhttps://github.com/leftouterjoins/EpilogDriverThe Epilog Zing lasers were 50w desktop industrial lasers from circa 2010. They never had support outside of windows.This was based off of LibLaserCut which is a Java project that reverse engineered these machine's protocols. This is a CUPS driver implemented in Swift and C.Maybe this is a small example of AI doing something positive, like breathing new life into old machines :)Dec 21, 2025 4:04 AM
bing.com
Understanding Absolute vs. Comparative Advantage in Tradehttp://www.bing.com/news/apiclick.aspx?ref=FexRss&aid=&tid=6a8ce3852c384725a11018626f9a0077&url=https%3A%2F%2Fwww.investopedia.com%2Fask%2Fanswers%2F041415%2Fwhat-are-some-real-life-examples-absolute-advantage.asp&c=4242997483032353681&mkt=en-usMary Hall is a editor for Investopedia's Advisor Insights, in addition to being the editor of several books and doctoral papers. Mary received her bachelor's in English from Kent State University with ...Dec 20, 2025 11:45 PM
spacepolicyonline.com
NASA Meets New Administrator Jared Isaacmanhttps://spacepolicyonline.com/news/nasa-meets-new-administrator-jared-isaacman/One day after being sworn in as NASA Administrator, Jared Isaacman held a Town Hall meeting with the NASA workforce to introduce himself and answer questions. He praised President Trump’s […]Dec 20, 2025 10:47 PM
github.com
Show HN: Skill capsules" for LLMs, a "poor man's continual learning"https://github.com/killerstorm/set_v4/blob/main/REPORT.md"Continual learning" is considered one of the "blockers" for LLMs: they can't learn on the job, don't improve over time, etc. In particular, Dwarkesh Patel describes it as a number of problem which has to be solved to get to AGI.Many academic article propose some kind of a memory system for LLM which might be considered a form of "continual learning". But most evals focus on memorizing facts which is just not very useful (it's better to fetch facts via tool use than to store it in neural memory) and these proposals might not fit well into common LLM API use patterns.In this article I'm proposing a "new" method called "skill capsules" which is highly pragmatic, easy to understand and evaluate and might integrate well into existing tooling.Skill capsule is a concrete object - it's a bunch of vectors, basically. You can insert it somewhere into a middle of LLM context and it improves performance on a particular skill, e.g. get tool calls more reliable, use particular writing style, codingDec 20, 2025 9:46 PM
indest.ee
Show HN: Big Christmas Giveawayhttps://indest.ee/Hi everyone,Last giveaway of the year!I wanted to wrap up 2025 by giving back to the community. It’s been a busy year for me as a solo developer; I managed to release two brand new titles and finally finished a massive consolidation update for my puzzle series.If you are looking for something new to play over the holidays, I’d love to share some promo codes. Just let me know which game interests you!The New 2025 Releases:Trivia Player: This is for the hardcore trivia fans. I built this because I was tired of multiple-choice quizzes that hold your hand. This is authentic trivia—real-time chat battles where you have to actually type the answer. No luck, just pure knowledge.Live Events: The game is live right now with a special "Merry Quizmas" event running all week. Also, mark your calendars for "Do You Remember 2025" starting on the 26th!AI Movie Quiz: I used AI to generate unique, slightly twisted clips and images based on famous movies. You have to guess the film from the AI's "dream.Dec 20, 2025 3:00 PM
news.ycombinator.com
MyRobot, a Venture Capital Pitchhttps://news.ycombinator.com/item?id=46333852Wrap yourself in a personal robot called Ava (homage to Ex Machina).A perfect morning coffee is waiting as you wake up. The house robot brings it to you as you awake.Suppose you want to get breakfast at a restaurant. The car shows up. You get driven to the restaurant. As you get seated your order arrives. Ava knows that you like an omlette, lightly cooked, with ham and bacon. You like your homefries crisp. You like a small orange juice. It is all ready for you moments after you walk in the door because Ava told the restaurant what you want and when you will arrive.Ava also knows that you want to keep your weight at a certain number. She tracks your calorie intake and creates food proportions and choice to meet your calerie requirements, tuned to the day's activities based on likely calorie burn.Ava reads books to you through the embedded speaker in your ear, tuning your book choice to the day. Is the weather bad; then a nice story. Is your interest in comedy; then a newly released (LLMDec 20, 2025 5:29 AM
spacenews.com
Hungary’s 4iG commits to $100 million Axiom Space investmenthttps://spacenews.com/hungarys-4ig-commits-to-100-million-axiom-space-investment/Hungarian communications provider 4iG has agreed to invest $100 million in U.S.-based space station developer Axiom Space, broadening its footprint in the space industry amid a broader push toward vertically integrated capabilities in the industry. The post Hungary’s 4iG commits to $100 million Axiom Space investment appeared first on SpaceNews.Dec 19, 2025 8:59 PM
bing.com
Trump Brags About Suing the Government and Declares, ‘I Hereby Give Myself $1 Billion’http://www.bing.com/news/apiclick.aspx?ref=FexRss&aid=&tid=6a8ccb9df2f848fbbc1173e299f1f9fa&url=https%3A%2F%2Fwww.yahoo.com%2Fnews%2Farticles%2Ftrump-brags-suing-government-declares-045214183.html&c=9263615434488010774&mkt=en-usAdd Yahoo as a preferred source to see more of our stories on Google. Speaking at a rally on Friday night, President Donald Trump bragged about his demand that the U.S. government give him a huge ...Dec 19, 2025 8:52 PM
