The best Hacker News stories from Show from the past day
Latest posts:
Show HN: LastShelf – an emergency map of your family's documents bills& contacts
After my father was diagnosed with Stage 3 kidney cancer, my family was thrown into a tailspin. Getting second opinions, planning surgery, ensuring insurance coverage, coping with the fear. It was a lot to process.<p>In the middle of dealing with all the medical logistics, I realized none of our family could answer if he:
- Had a medical directive?
- How to trigger his life insurance policy?
- Where is his will and who is the executor?
- What bank accounts and credit cards existed?
- What bills are not on auto-pay?
- When these bills due and how are they paid?<p>That wasn’t solved by password managers or budgeting apps. So I built it.<p>LastShelf: automatically discovers, documents and distributes a map of critical life documents, expenses & contacts in the event of an emergency. Register here: <a href="https://www.lastshelf.ai/" rel="nofollow">https://www.lastshelf.ai/</a><p>If you’ve lived through a similar crisis, I really want to hear what would have made the process easier.<p>Anyone who shares their feedback with me will get the first year free. Send a note to support [at] lastshelf.ai
Show HN: I mapped 8.5M research papers into an interactive atlas
When I read papers, I have to jump between multiple tabs to find the dataset, code, videos, peer reviews, and so on. I tried to fix this with this project.<p>It started as a project just for papers on arXiv, but after its initial success on Twitter (got like 1.9k views: the most I have gotten for a post), I have now expanded it to include other openly available papers from PubMed Central, bioRxiv, medRxiv, and eLife. These papers have been linked with their genes, proteins, diseases, drugs, clinical trials, 3D protein structures, code, and cited and similar papers.<p>This project now has four parts:<p>First, a map. I embedded nearly 8.5M papers (with SPECTER2), ran UMAP for 2D representation, and rendered them as a scatterplot. The dots can be clicked to see brief information about the papers, like an LLM TLDR, key findings, peer reviews, linked entities, and more. The clusters are also labeled, though you might have to zoom in.<p>Second, I built a detailed paper page for each paper. They give you the paper's full text, images, videos, peer reviews (from OpenReview), GitHub links, Hugging Face dataset/model links, clinical trials, genes, diseases, 3D protein structures, cited papers, and similar papers. You can also copy the whole page, including the full paper text and image URLs, as markdown for your LLM.<p>Third, I have released an extension so you can read all this information in your sidebar by clicking "open in Tomesphere" that shows up in arXiv, PMC, bioRxiv, Google Scholar, or medRxiv. I have tried to provide as much information as possible in the extension, though for things like viewing all the images or a 3D protein structure, you might still have to go to the paper page using the link provided in the extension.<p>Fourth, all this data is available for your LLM via MCP. The MCP does have a 50-query free limit (this jumps 10x with signup).<p>Note: this project is still in beta, so papers might have some mismatched information. I am rolling out feedback forms soon to improve the data quality. Thank you so much for taking the time to read this.
Show HN: I mapped 8.5M research papers into an interactive atlas
When I read papers, I have to jump between multiple tabs to find the dataset, code, videos, peer reviews, and so on. I tried to fix this with this project.<p>It started as a project just for papers on arXiv, but after its initial success on Twitter (got like 1.9k views: the most I have gotten for a post), I have now expanded it to include other openly available papers from PubMed Central, bioRxiv, medRxiv, and eLife. These papers have been linked with their genes, proteins, diseases, drugs, clinical trials, 3D protein structures, code, and cited and similar papers.<p>This project now has four parts:<p>First, a map. I embedded nearly 8.5M papers (with SPECTER2), ran UMAP for 2D representation, and rendered them as a scatterplot. The dots can be clicked to see brief information about the papers, like an LLM TLDR, key findings, peer reviews, linked entities, and more. The clusters are also labeled, though you might have to zoom in.<p>Second, I built a detailed paper page for each paper. They give you the paper's full text, images, videos, peer reviews (from OpenReview), GitHub links, Hugging Face dataset/model links, clinical trials, genes, diseases, 3D protein structures, cited papers, and similar papers. You can also copy the whole page, including the full paper text and image URLs, as markdown for your LLM.<p>Third, I have released an extension so you can read all this information in your sidebar by clicking "open in Tomesphere" that shows up in arXiv, PMC, bioRxiv, Google Scholar, or medRxiv. I have tried to provide as much information as possible in the extension, though for things like viewing all the images or a 3D protein structure, you might still have to go to the paper page using the link provided in the extension.<p>Fourth, all this data is available for your LLM via MCP. The MCP does have a 50-query free limit (this jumps 10x with signup).<p>Note: this project is still in beta, so papers might have some mismatched information. I am rolling out feedback forms soon to improve the data quality. Thank you so much for taking the time to read this.
Show HN: I mapped 8.5M research papers into an interactive atlas
When I read papers, I have to jump between multiple tabs to find the dataset, code, videos, peer reviews, and so on. I tried to fix this with this project.<p>It started as a project just for papers on arXiv, but after its initial success on Twitter (got like 1.9k views: the most I have gotten for a post), I have now expanded it to include other openly available papers from PubMed Central, bioRxiv, medRxiv, and eLife. These papers have been linked with their genes, proteins, diseases, drugs, clinical trials, 3D protein structures, code, and cited and similar papers.<p>This project now has four parts:<p>First, a map. I embedded nearly 8.5M papers (with SPECTER2), ran UMAP for 2D representation, and rendered them as a scatterplot. The dots can be clicked to see brief information about the papers, like an LLM TLDR, key findings, peer reviews, linked entities, and more. The clusters are also labeled, though you might have to zoom in.<p>Second, I built a detailed paper page for each paper. They give you the paper's full text, images, videos, peer reviews (from OpenReview), GitHub links, Hugging Face dataset/model links, clinical trials, genes, diseases, 3D protein structures, cited papers, and similar papers. You can also copy the whole page, including the full paper text and image URLs, as markdown for your LLM.<p>Third, I have released an extension so you can read all this information in your sidebar by clicking "open in Tomesphere" that shows up in arXiv, PMC, bioRxiv, Google Scholar, or medRxiv. I have tried to provide as much information as possible in the extension, though for things like viewing all the images or a 3D protein structure, you might still have to go to the paper page using the link provided in the extension.<p>Fourth, all this data is available for your LLM via MCP. The MCP does have a 50-query free limit (this jumps 10x with signup).<p>Note: this project is still in beta, so papers might have some mismatched information. I am rolling out feedback forms soon to improve the data quality. Thank you so much for taking the time to read this.
Show HN: FableCut – A browser video editor AI agents can drive (zero deps)
Show HN: FableCut – A browser video editor AI agents can drive (zero deps)
Show HN: Analog Watch
Show HN: Analog Watch
Show HN: Yamanote.fun – A complete soundscape for Tokyo's Yamanote line
After visiting Japan for the first time a decade ago I became completely enamoured with Tokyo's Yamanote Line railway loop. Particularly the sonic experience of it. Like so many others I fell in love with the charming departure melodies and enjoyed discovering experiences like Yamanot.es (<a href="https://news.ycombinator.com/item?id=45045307">https://news.ycombinator.com/item?id=45045307</a>) here on Hacker News when I returned home.<p>But it wasn't until my second trip to Tokyo that I truly appreciated how much the door chimes, on-board announcements and train noise were contributing to the rich soundscape that I loved.<p>I returned home and found myself playing YouTube videos of Yamanote Line journeys as I worked. The combination of sonics, ambience and softly spoken Japanese was incredibly soothing to me.<p>But these recordings were often incomplete, poorly captured or out of date, and I wanted something far more comprehensive.<p>So I gathered up all of the constituent parts from Reddit threads, YouTube videos and Japanese fan sites, and set about recreating the experience of riding the Yamanote Line in Logic Pro X. Melody, door chimes and announcement, all stitched together under a bed of train noise and ambience.<p>I turned those soundscapes into an Alexa Skill (<a href="https://www.amazon.co.uk/Paul-Jackson-Yamanote-Line/dp/B07S18QRMV" rel="nofollow">https://www.amazon.co.uk/Paul-Jackson-Yamanote-Line/dp/B07S1...</a>) in 2019 and began to think about a companion website to share the soundscapes with a wider audience.<p>Seven years later and that website is Yamanote.fun: <a href="https://www.yamanote.fun/" rel="nofollow">https://www.yamanote.fun/</a>.<p>It's a small installable web app that plays the soundscapes like a playlist. All 30 stations and in both directions, since the inner and outer loops use different melodies. You can skip forward or back a station, and there's a scrub bar broken into melody / chime / ambience / announcement so you can jump straight to the bit you want. Each station has its own shareable link (yamanote.fun/jy13-ikebukuro-inner) that unfurls with the right station name and artwork when you share it.<p>It's a progressive web app too, so you can add it to your home screen and it behaves like a native app. There's an option to offline the audio too.<p>Under the hood it's relatively basic stuff: plain HTML, CSS & JS, audio served from Cloudflare R2 and the site hosted on Netlify. I was impressed to see how far I could get with the free tiers of these services. I designed the whole thing in Figma (I'm a Product Designer) and used Claude Code to architect and deliver the polished UI, PWA plumbing, offline caching and share-link infrastructure.<p>I would love feedback, particularly from anyone who's ridden the real thing.
Show HN: Getting GLM 5.2 running on my slow computer
A few days ago I found myself trying out GLM 5.2 and was really positively impressed. The capabilities and security I was getting from this LLM are similar to those I've gotten from models like Claude or GPT, and this really surprised me.<p>But then I thought, "I wonder how it would work on a normal computer like mine," and above all, "I wonder if it would work without going into OOM on a computer like mine." So I started working with the help of agents to test this possibility.<p>I started converting the model to int4, understanding MTP usage, and if possible implementing DSA for long context. How it responds in int4 and whether the quality is maintained or not. Until I got to the point, on my computer with 32GB of RAM, I was able to communicate with GLM 5.2 with times that, of course, aren't high in cold start, but even then, we're talking about 0.1 tok/s, but that wasn't important to me. The important thing was the journey to reach this goal. I just wanted it to work at all costs, even slowly.<p>So I created Colibrì, which was born from a very simple idea, to be honest, but tested in every way, where a 744B Mixture-of-Experts model activates only ~40B parameters per token—and only ~11 GB of those change from token to token (the routed experts). So:<p>The dense part (attention, shared experts, embeddings—~17B params) stays resident in RAM at int4 (~9.9 GB); The 21,504 routed experts (75 MoE layers × 256 experts + the MTP head, ~19 MB each at int4) live on disk (~370 GB) and are streamed on demand, with a per-layer LRU cache, an optional pinned hot-store, and the OS page cache as a free L2.<p>The engine is a single C file (c/glm.c, ~1,300 lines) plus small headers. No BLAS, no Python at runtime, no GPU.No GPU or serious hardware because I don't have that hardware so I can't test it on hardware that is more powerful than my computer.Colibrì is a one-person project, written and tested entirely on a 12-core laptop with 25 GB of RAM — the numbers above are the ceiling of what I can measure at home.<p>Any feedback is welcome! (and if anyone wanted to participate in the project I would be delighted)<p>Repo: <a href="https://github.com/JustVugg/colibri" rel="nofollow">https://github.com/JustVugg/colibri</a>
Show HN: Getting GLM 5.2 running on my slow computer
A few days ago I found myself trying out GLM 5.2 and was really positively impressed. The capabilities and security I was getting from this LLM are similar to those I've gotten from models like Claude or GPT, and this really surprised me.<p>But then I thought, "I wonder how it would work on a normal computer like mine," and above all, "I wonder if it would work without going into OOM on a computer like mine." So I started working with the help of agents to test this possibility.<p>I started converting the model to int4, understanding MTP usage, and if possible implementing DSA for long context. How it responds in int4 and whether the quality is maintained or not. Until I got to the point, on my computer with 32GB of RAM, I was able to communicate with GLM 5.2 with times that, of course, aren't high in cold start, but even then, we're talking about 0.1 tok/s, but that wasn't important to me. The important thing was the journey to reach this goal. I just wanted it to work at all costs, even slowly.<p>So I created Colibrì, which was born from a very simple idea, to be honest, but tested in every way, where a 744B Mixture-of-Experts model activates only ~40B parameters per token—and only ~11 GB of those change from token to token (the routed experts). So:<p>The dense part (attention, shared experts, embeddings—~17B params) stays resident in RAM at int4 (~9.9 GB); The 21,504 routed experts (75 MoE layers × 256 experts + the MTP head, ~19 MB each at int4) live on disk (~370 GB) and are streamed on demand, with a per-layer LRU cache, an optional pinned hot-store, and the OS page cache as a free L2.<p>The engine is a single C file (c/glm.c, ~1,300 lines) plus small headers. No BLAS, no Python at runtime, no GPU.No GPU or serious hardware because I don't have that hardware so I can't test it on hardware that is more powerful than my computer.Colibrì is a one-person project, written and tested entirely on a 12-core laptop with 25 GB of RAM — the numbers above are the ceiling of what I can measure at home.<p>Any feedback is welcome! (and if anyone wanted to participate in the project I would be delighted)<p>Repo: <a href="https://github.com/JustVugg/colibri" rel="nofollow">https://github.com/JustVugg/colibri</a>
Show HN: Getting GLM 5.2 running on my slow computer
A few days ago I found myself trying out GLM 5.2 and was really positively impressed. The capabilities and security I was getting from this LLM are similar to those I've gotten from models like Claude or GPT, and this really surprised me.<p>But then I thought, "I wonder how it would work on a normal computer like mine," and above all, "I wonder if it would work without going into OOM on a computer like mine." So I started working with the help of agents to test this possibility.<p>I started converting the model to int4, understanding MTP usage, and if possible implementing DSA for long context. How it responds in int4 and whether the quality is maintained or not. Until I got to the point, on my computer with 32GB of RAM, I was able to communicate with GLM 5.2 with times that, of course, aren't high in cold start, but even then, we're talking about 0.1 tok/s, but that wasn't important to me. The important thing was the journey to reach this goal. I just wanted it to work at all costs, even slowly.<p>So I created Colibrì, which was born from a very simple idea, to be honest, but tested in every way, where a 744B Mixture-of-Experts model activates only ~40B parameters per token—and only ~11 GB of those change from token to token (the routed experts). So:<p>The dense part (attention, shared experts, embeddings—~17B params) stays resident in RAM at int4 (~9.9 GB); The 21,504 routed experts (75 MoE layers × 256 experts + the MTP head, ~19 MB each at int4) live on disk (~370 GB) and are streamed on demand, with a per-layer LRU cache, an optional pinned hot-store, and the OS page cache as a free L2.<p>The engine is a single C file (c/glm.c, ~1,300 lines) plus small headers. No BLAS, no Python at runtime, no GPU.No GPU or serious hardware because I don't have that hardware so I can't test it on hardware that is more powerful than my computer.Colibrì is a one-person project, written and tested entirely on a 12-core laptop with 25 GB of RAM — the numbers above are the ceiling of what I can measure at home.<p>Any feedback is welcome! (and if anyone wanted to participate in the project I would be delighted)<p>Repo: <a href="https://github.com/JustVugg/colibri" rel="nofollow">https://github.com/JustVugg/colibri</a>
Show HN: Getting GLM 5.2 running on my slow computer
A few days ago I found myself trying out GLM 5.2 and was really positively impressed. The capabilities and security I was getting from this LLM are similar to those I've gotten from models like Claude or GPT, and this really surprised me.<p>But then I thought, "I wonder how it would work on a normal computer like mine," and above all, "I wonder if it would work without going into OOM on a computer like mine." So I started working with the help of agents to test this possibility.<p>I started converting the model to int4, understanding MTP usage, and if possible implementing DSA for long context. How it responds in int4 and whether the quality is maintained or not. Until I got to the point, on my computer with 32GB of RAM, I was able to communicate with GLM 5.2 with times that, of course, aren't high in cold start, but even then, we're talking about 0.1 tok/s, but that wasn't important to me. The important thing was the journey to reach this goal. I just wanted it to work at all costs, even slowly.<p>So I created Colibrì, which was born from a very simple idea, to be honest, but tested in every way, where a 744B Mixture-of-Experts model activates only ~40B parameters per token—and only ~11 GB of those change from token to token (the routed experts). So:<p>The dense part (attention, shared experts, embeddings—~17B params) stays resident in RAM at int4 (~9.9 GB); The 21,504 routed experts (75 MoE layers × 256 experts + the MTP head, ~19 MB each at int4) live on disk (~370 GB) and are streamed on demand, with a per-layer LRU cache, an optional pinned hot-store, and the OS page cache as a free L2.<p>The engine is a single C file (c/glm.c, ~1,300 lines) plus small headers. No BLAS, no Python at runtime, no GPU.No GPU or serious hardware because I don't have that hardware so I can't test it on hardware that is more powerful than my computer.Colibrì is a one-person project, written and tested entirely on a 12-core laptop with 25 GB of RAM — the numbers above are the ceiling of what I can measure at home.<p>Any feedback is welcome! (and if anyone wanted to participate in the project I would be delighted)<p>Repo: <a href="https://github.com/JustVugg/colibri" rel="nofollow">https://github.com/JustVugg/colibri</a>
Show HN: 18 Words
Show HN: 18 Words
Show HN: 18 Words
Show HN: 18 Words
Show HN: Chiptune Radio
I built a chiptune song generator and its broadcasting algorithmically generated chiptune music.
Show HN: Fortress – a stealth Chromium so your agents stop getting blocked
Show HN: Neil the Seal Game
Neil the seal now has a game to destroy Battery Point.