Frontier model pre-training has converged on mixture of experts (MoE), which is fundamentally changing what limits large-scale AI training. As compute per token...
Nvidia’s Vera Rubin platform combines CPUs and GPUs into a single system, reflecting the company’s growing ambition to power every layer of AI infrastructure.
.jpg)
Flipping heck! ScienceAlert stories are written, fact-checked, and edited by humans, never generated by AI. Don't miss a story, subscribe here.
If your AI Factory sells tokens, why not optimize token emission?
<p> 13,000+ MCP servers, skills & plugins for AI coding agents </p> <p> <a href="https://www.producthunt.com/products/remote-openclaw-2?utm_campaign=producthunt-atom-posts-feed&utm_medium=rss-feed&utm_source=producthunt-atom-posts-feed">Discussion</a> | <a href="https://www.producthunt.com/r/p/1202590?app_id=339">Link</a> </p>
A hacker took names, phone numbers, and physical addresses of millions of customers who used AI music generator Suno.
As of July 21, save over $400 on the iRobot Roomba Max 705 at Amazon.
Stellantis will use technology from Intel's Mobileye to power its Level 2 hands-free driving features for its vehicle brands, the companies announced Tuesday. Starting in 2027, Stellantis will begin using Mobileye's EyeQ system-on-a-chip and a data crowdsourcing program called the Road Experience Management, or REM, which uses real-time data from Mobileye-equipped vehicles to build out […]
<p> Let Agents Autonomously Build Better Agents for $0.02 </p> <p> <a href="https://www.producthunt.com/products/penguinharness?utm_campaign=producthunt-atom-posts-feed&utm_medium=rss-feed&utm_source=producthunt-atom-posts-feed">Discussion</a> | <a href="https://www.producthunt.com/r/p/1202577?app_id=339">Link</a> </p>
Useful when implementing AI skills, knowledge bases, MCP descriptions etc - its much easier to use CLI tool than web calculator (especially for multiple files) Comments URL: https://news.ycombinator.com/item?id=48992877 Points: 2 # Comments: 0
Article URL: https://github.com/telemetry-sh/llm-margin-lab Comments URL: https://news.ycombinator.com/item?id=48992841 Points: 2 # Comments: 0
It's now easier to prompt an AI for a specific problem and later release the code on git, without thinking or researching a bigger goal that there could be an existing project where my contribution would be much more meaningful and sustainable... How could this fragmentation be saved in favor of existing projects? Comments URL: https://news.ycombinator.com/item?id=48992797 Points: 1 # Comments: 0
<p><strong><a href="https://blaizzy.github.io/nativ/">Nativ: Run AI models locally on your Mac</a></strong></p> Prince Canuma is the developer behind the excellent <a href="https://github.com/Blaizzy/mlx-vlm">MLX-VLM</a> Python library for running vision-LLMs using MLX on a Mac.</p> <p>I'm really excited about his new project, which wraps MLX in a full macOS desktop application. It's similar in shape to LM Studio, providing both a chat interface and a localhost API server for accessing models.</
Article URL: https://www.plagiarismtoday.com/2026/07/15/study-claiming-ai-helps-students-learn-retracted/ Comments URL: https://news.ycombinator.com/item?id=48992750 Points: 1 # Comments: 0
Does washing actually help? ScienceAlert stories are written, fact-checked, and edited by humans, never generated by AI. Don't miss a story, subscribe here.
The 65-inch LG OLED AI B6 4K TV is on sale at Amazon for $1,599.99, down from the list price of $1,999.99. That's a 20% discount.
Article URL: https://devgeist.com/blog/ai-infra-mutation/ Comments URL: https://news.ycombinator.com/item?id=48992621 Points: 1 # Comments: 0
Article URL: https://www.404media.co/ai-companies-are-buying-tons-of-old-books-because-theyre-free-of-ai-slop/ Comments URL: https://news.ycombinator.com/item?id=48992588 Points: 2 # Comments: 1
Live from the Institute of Law & AI's Workshop on Law-Following AI
Run Qwythos-9B-Claude-Mythos-5-1M locally with llama.cpp, connect it to Pi coding agent, and build fast local coding workflows using MTP speculative decoding and an OpenAI-compatible API.
Spain’s World Cup champions celebrated with custom Oakley Meta glasses. Now, AI glasses are stepping onto the field.
Article URL: https://twitter.com/shl/status/2079387173749714944 Comments URL: https://news.ycombinator.com/item?id=48992436 Points: 3 # Comments: 0
AI's costs are landing in real places, and the people absorbing them are writing the rules after the fact, in council chambers and courtrooms. Source
Article URL: https://github.com/RimantasZ/contextspy Comments URL: https://news.ycombinator.com/item?id=48992375 Points: 2 # Comments: 0
Have I Been Pwned confirms scale for first time
If AI got lucky or not, AI is surely going to assist in solving serious math problems. If you want to spend your resources/weekends contributing, this repository is full of math problems where you can add your thoughts (or proofs - for you Good Will Hunting type people) for others to build on it and we all together can try to change the world of math. I have already extended the research with manual proof of Jacobian Conjecture that was posted on Twitter (with attribution). Comments URL: https:/
Article URL: https://github.com/The18thWarrior/tzro Comments URL: https://news.ycombinator.com/item?id=48992154 Points: 2 # Comments: 0
Hello, while I was learning LLM Inference basics I realized with few tricks I could use my ESP32 for inference if I stream the weights only when they are needed from a host over WiFi. Speed? Really bad. Lol. But this is just a technical demo of how it can be done. I really enjoyed while coding it. It has a sliding window which allows infinite generation ( of course meaningless :) ) Hope you will enjoy! Comments URL: https://news.ycombinator.com/item?id=48992123 Points: 2 # Comments: 0
Exploring GPU acceleration with cuDF, cudf.pandas, and the Polars GPU Engine The post How Much of a Data Science Workflow Can Run on a GPU Today? Part 1: Accelerating Data Preparation appeared first on Towards Data Science .
Deezer said more than 90,000 AI-generated tracks were uploaded daily on the platform in June.
Article URL: https://nutfieldsecurity.com/posts/ai-vacation-physics/ Comments URL: https://news.ycombinator.com/item?id=48992051 Points: 2 # Comments: 0
Article URL: https://antiextinction.substack.com/p/contra-ai-2040-and-the-cult-of-intelligence Comments URL: https://news.ycombinator.com/item?id=48992050 Points: 1 # Comments: 2
Howdy dear folks. Here on hackernews there seems to be very little other than AI related stories. Maybe not all too strange given the crowd who frequents this site. From what I gather the mostly positive sentiment isn't as widely shared among the general population (0: 50% of americans are more concerned than excited about the increase use of AI). How do you think about whether or not to mention, or advertise, the AI capabilities of your own products when marketing to the general population? Do
I use Pi coding agent with OpenAI GPT-5.6 Sol but have been finding that GPT-5.6 is slower than Claude Code. Do you use any other model like one of the hosted OSS ones? Comments URL: https://news.ycombinator.com/item?id=48991997 Points: 3 # Comments: 2
Article URL: https://georgwiese.github.io/posts/formal-verification-ai/ Comments URL: https://news.ycombinator.com/item?id=48991981 Points: 4 # Comments: 0
Hi HN, we're developing Neverbell, an AI agent that combines market analysis with trade execution. One of our beta testers said to Neverbell: “Protect my downside, but don’t be too conservative.” Volatility increased a few days later, and the agent decreased their exposure. Our tester wasn’t happy. The interesting part for us was that both interpretations were reasonable. The agent gave more weight to “protect my downside.” The user thought that “don't be too conservative” was more important. It
<pre style='white-space:pre-wrap;width:81ex'>fix(gateway): keep fresh-state startup responsive with prepared runtimes (#112262) * fix(gateway): keep health responsive during model runtime startup * refactor(agents): isolate prepared runtime startup mode * refactor(agents): keep prepared runtime catalog mode internal</pre>
I built Pinpoint because I was tired of typing out feedback on the visual work my AI agent did. I want my agent to be autonomous, so having it ask for feedback through a UI felt genuinely better. Dropping a contextual comment on each screenshot is much closer to how you'd give feedback on a design in Figma. It also does side-by-side comparison, which is really helpful, having the before/after screens lets the human in the loop (us!) give better feedback. It's just screenshots and comments, the s
<h1> Building Production-Grade LLM Evaluation Pipelines: From Vibes to Metrics </h1> <p><em>How we replaced "looks good to me" with automated evaluation catching 92% of hallucinations before deployment</em></p> <h2> The Problem: Why "Vibe Checks" Fail in Production </h2> <p>Three months ago, our team shipped a RAG-based customer support assistant. It worked great in testing — we'd ask it questions, read the answers, and say "yeah, that looks right."</p> <p>Then it hit production.</p> <p>A custom
