Apple is finally making its own modems, starting with the iPhone 16e. This means the company is moving away from relying on Qualcomm for this key technology. But despite this shift, Qualcomm’s CEO, ...
Hi everyone,I’ve been working on an open-source tool called Flakestorm to test the reliability of AI agents before they hit production.Most agent testing today focuses on eval scores or happy-path prompts. In practice, agents tend to fail in more mundane ways: typos, tone shifts, long context, malformed input, or simple prompt injections — especially when running on smaller or local models. Flakestorm applies chaos-engineering ideas to agents. Instead of testing one prompt, it takes a “golden pr
We built ViviGener, a free community platform where anyone can discover, share, and contribute AI prompt templates for various models like Nano Banana, ChatGPT, Midjourney, and Stable Diffusion.The Problem: Most valuable prompts are scattered across Reddit threads, Discord servers, Twitter posts, and private notes. There's no centralized place where the community can build and maintain a high-quality prompt library together.What We Offer: • Searchable Library: Browse curated AI prompts orga
I'm a developer who got into flipping/reselling as a side hustle last year. My biggest frustration was standing in thrift stores trying to figure out if something was worth buying while juggling eBay sold listings, Google searches, and fee calculators.So I built Underpriced.app - an AI reseller app that tells you if something is worth flipping before you buy it.The ProblemResearching items takes ~5 minutes per item You waste time and miss deals while researching (someone else grabs it)
Hey HN!Currently, the key parameters for Type 1 diabetes management, like Insulin-To-Carb ratio (ICR) and Correction Factor (CF), are estimated manually by eyeballing historical charts and making a subjective guess about what the "true" values should be.There is a lot of uncertainty and noise in the data itself, so the estimation task is hard even for Diabetic doctors.I thought there had to be a more principled way to make such decisions from the data, so I built a Bayesian ML model th
I’m working on a few LED strip installs where the runs are long enough that the “usual beginner advice” (just pick a PSU and stick the strip up) stops working.Typical scenarios:24V constant-voltage strips for indirect/cove lighting, ~5–20m per runSometimes addressable strips (SPI-style) where data integrity becomes a factorIndoor installs in aluminum channels / diffusers (so heat and wiring neatness matter)I’d love to collect practical rules of thumb from people who’ve done this at sca
Over the last two years we have gone from “LLMs with tools” to genuinely agentic systems that plan, reflect, delegate, retry, and sometimes surprise us in ways that feel uncomfortably close to junior engineers. The ecosystem has matured fast enough that framework choice now meaningfully shapes what your agents can and cannot become.Here is a ground level comparison from someone who has built, broken, and rebuilt agents across several stacks, focusing less on benchmarks and more on lived behavior
Hi HNMonths Ago, we set out with one mission: make high-quality image creation feel effortless for developers, creators, and designers alike. Over countless nights of code, iterations, and real-world testing, we built Imgity, a new kind of AI image platform that blends generation and editing into one seamless experience.At its core, Imgity is powered by Nano Banana Pro — a reasoning-focused engine that gives us unusually high-fidelity results: physics-accurate lighting, studio-grade details, and
Hi HN, I am Kiana, I was an AI researcher for 10 years, and since last November I have been working on Vy. Vy is a desktop app for Mac and Windows. It sees your screen and controls mouse and keyboard to automate workflows across apps, including ones without APIs. I know there is a lot of skepticism about AI agents, and I share some of it. A lot of things in this space look impressive in demos and then fall apart on real, boring work. So here are a few concrete things that Vy does today that we
I built a desktop mock server that lets you update endpoints while it's running. No restart required.The problem: Every time I changed a mock endpoint in other tools, I had to restart the server. WebSocket connections dropped, test flows broke.The solution: Catch-all handlers that read config on every request. Change anything – routes, responses, delays – and it takes effect immediately. WebSocket/SSE connections stay alive.Features that might interest you:• State machine for multi-ste
Today at CES 2026, Jensen Huang stated: "Physical AI requires three computers."An AI Supercomputer (DGX) to train the brain. A Simulation Computer (Omniverse) to simulate the world (Expectation). A Robot Computer (Jetson) to act in the real world (Observation).The core of this architecture is the intentional separation of Simulation and Reality—designed to create a "Sim-to-Real Gap." When the simulation says "this floor is safe" but the robot feels "slippery,&q
Ai Tongue GIF by Respective
day internet GIF
This started as a hobby project, curious if others find it useful. Feedback welcome!botchat is a privacy-preserving, multi-bot chat tool that lets you interact with multiple AI models simultaneously.Give bots personas, so they look at your question from multiple angles. Leverage the strengths of different models in the same chat. And most importantly, protect your data.botchat never stores your conversations or attachments on any servers and, if you are using our keys (the default experience), y
I’ve been meaning to write this up because it’s been surprisingly repeatable, and I wish someone had described it to me earlier.Over the last year or so, my development speed relative to my own baseline from ~2019 is easily 20x, sometimes more. Not because I type faster, or because I cut corners, but because I changed how I use AI.The short version: I don’t use AI inside my editor. I use two AIs in parallel, in the browser, with full context.Here’s the setup.I keep two tabs open:One AI that acts
I’ve been utilizing Google’s Gemini 1.5 Pro via the AI Studio front-end to develop a new platform. The 1M+ context window is, technically speaking, a game-changer for "stitching" together a 55,000-line codebase. However, I recently discovered a predatory billing architecture that I’m calling the "Context Tax."If you use the AI Studio UI, you might be walking into a massive bill without a single warning.Here is how it happened, and the UK/EU privacy "pro-tip" I
Author here.We’ve been frustrated by the "Scaling Hypothesis" (just make the model bigger). We believe the issue with current AI isn't size, it's architecture.We just open-sourced TOPAS-DSPL. It’s a tiny model (~15M params) that uses a Dual-Stream architecture (separating Logic from Execution) to solve the ARC-AGI-2 benchmark.It achieves 24% accuracy on the hard evaluation set, which beats many models 1000x its size.The repo includes the full training code, data augmentation
Hi HN, I’m Oscar, a Year 8 student from Australia who enjoys messing around with computers and AI.I recently built an AI detector to build on my skills in computer science. I entered the prototype into the Oliphant Science Awards which is a local science competition (writing a 4000-word report on the methodology) and ended up winning, so I decided to polish it into a real web service that the world can make use of.I noticed that schools and businesses are rushing to use AI detectors, but most c
HN PHONE DEMO - AI BUSINESS FACTORY ====================================== Running on: Linux localhost 5.10.226-android12-9-g018b9fdb2168 #1 SMP PREEMPT Thu May 8 03:53:46 UTC 2025 aarch64 Android Storage: 0 free Memory: 2.8Gi/5.3Gi used Starting AI Business Factory Demo... This is what's running on my actual phone right now.Demo output: /data/data/com.termux/files/home/HN_DEMO_1767188899Generating AI business concepts...
Show HN: VividManga – AI-based manga coloring focused on line art consistencyHi HN,I built VividManga because most general image colorization models perform poorly on manga-style black-and-white line art — they often blur lines or destroy panel structure.VividManga focuses specifically on manga coloring, using an image-to-image workflow tuned for clean line preservation, flat color regions, and consistent shading.The core idea is to treat manga pages differently from natural images: - prioritize