The best Hacker News stories from All from the past day
Latest posts:
Why I still blog after 15 years
NASA spacecraft to probe possibility of life in Europa's salty ocean
Llama 3.2: Revolutionizing edge AI and vision with open, customizable models
Llama 3.2: Revolutionizing edge AI and vision with open, customizable models
Mira Murati leaves OpenAI
Orion, our first true augmented reality glasses
Orion, our first true augmented reality glasses
Launch HN: Modern Realty (YC S24) – AI Real Estate Agent for Home Buyers
Hey HN! We're Raymond Xu and Raffi Isanians, the team behind Modern Realty (<a href="https://modernrealty.io">https://modernrealty.io</a>). Modern Realty is an AI-powered real estate buying experience that helps you purchase homes without relying on a traditional realtor. Demo: <a href="https://youtu.be/VhsWSVn5blA" rel="nofollow">https://youtu.be/VhsWSVn5blA</a><p>Our mission at Modern Realty is to make home buying easier, faster, and far more enjoyable. We interpret real estate data, giving you an edge on other home buyers.<p>Since 2024, there has been a growing need for innovative solutions in real estate, especially with tools like Zillow making it easier than ever for buyers to find the right house at the right price. However, the process still requires an agent, even though buyers are doing most of the work themselves. Modern Realty changes this by offering an AI self-service solution.<p>Raymond and Raffi came face-to-face with this technology's potential while working on AI at Google and handling billion-dollar asset transactions as an attorney, respectively. The AI tools we've built guide users through every step of the home-buying process, from scheduling showings to making offers, all without the need for a traditional agent.<p>Before we expand further, this summer we've built our core AI real estate agent, which can handle everything from disclosure summaries to competitive market analyses. You can see our platform in action at <a href="https://modernrealty.io">https://modernrealty.io</a>. We're particularly proud of our offer generation tool, which creates offers that are then reviewed by an attorney at no additional cost.<p>We're excited to be launching here on HN and would love to hear what you think! If you're in the market for a new home or just curious about the future of real estate, visit our website or schedule a call with us to get started.
Why Most Published Research Findings Are False (2005)
We fine-tuned Llama 405B on AMD GPUs
Hey HN, we recently fine-tuned the llama3.1 405B model on 8xAMD MI300x GPUs using JAX instead of PyTorch. JAX's advanced sharding APIs allowed us to achieve great performance. Check out our blog post to learn about the cool sharding tricks we used. We've also open-sourced the code: <a href="https://github.com/felafax/felafax">https://github.com/felafax/felafax</a><p>We're a small startup building AI infra for fine-tuning and serving LLMs on non-NVIDIA hardware (TPUs, AMD, Trainium).<p>Problem: Many companies are trying to get PyTorch working on AMD GPUs, but we believe this is a treacherous path. PyTorch is deeply intertwined with the NVIDIA ecosystem in a lot of ways (e.g., `torch.cuda` or scaled_dot_product_attention is an NVIDIA CUDA kernel exposed as a PyTorch function). So, to get PyTorch code running on non-NVIDIA hardware, there's a lot of "de-NVIDIAfying" that needs to be done.<p>Solution: We believe JAX is a better fit for non-NVIDIA hardware. In JAX, ML model code compiles to hardware-independent HLO graphs, which are then optimized by the XLA compiler before hardware-specific optimization. This clean separation allowed us to run the same LLaMA3 JAX code both on Google TPUs and AMD GPUs with no changes.<p>Our strategy as a company is to invest upfront in porting models to JAX, then leverage its framework and XLA kernels to extract maximum performance from non-NVIDIA backends. This is why we first ported Llama 3.1 from PyTorch to JAX, and now the same JAX model works great on TPUs and runs perfectly on AMD GPUs.<p>We'd love to hear your thoughts on our vision and repo!
Move Fast and Abandon Things
Move Fast and Abandon Things
iPhone 16 Pro Storage Expansion 128GB to 1TB [video]
Winamp Legacy player source code
Winamp Legacy player source code
Show HN: OpenFreeMap – Open-Source Map Hosting
Hi HN,<p>After 9 years of running my own OpenStreetMap tile server infra for MapHub (<a href="https://news.ycombinator.com/item?id=11389989">https://news.ycombinator.com/item?id=11389989</a>), I've open-sourced it and launched OpenFreeMap.<p>You can either self-host or use our public instance. Everything is open-source, including the full production setup — there’s no 'open-core' model here. Check out the repo (<a href="https://github.com/hyperknot/openfreemap">https://github.com/hyperknot/openfreemap</a>). The map data comes from OpenStreetMap.<p>I also provide weekly full planet downloads both in Btrfs and MBTiles formats.<p>I aim to cover the running costs of the public instance through donations.<p>Looking forwards for your feedback.
Show HN: OpenFreeMap – Open-Source Map Hosting
Hi HN,<p>After 9 years of running my own OpenStreetMap tile server infra for MapHub (<a href="https://news.ycombinator.com/item?id=11389989">https://news.ycombinator.com/item?id=11389989</a>), I've open-sourced it and launched OpenFreeMap.<p>You can either self-host or use our public instance. Everything is open-source, including the full production setup — there’s no 'open-core' model here. Check out the repo (<a href="https://github.com/hyperknot/openfreemap">https://github.com/hyperknot/openfreemap</a>). The map data comes from OpenStreetMap.<p>I also provide weekly full planet downloads both in Btrfs and MBTiles formats.<p>I aim to cover the running costs of the public instance through donations.<p>Looking forwards for your feedback.
In 1870, Lord Rayleigh used oil and water to calculate the size of molecules
In 1870, Lord Rayleigh used oil and water to calculate the size of molecules
London saw a surprising benefit to ultra-low emissions zone: More active kids