- AI Fire
- Posts
- 🛰️ Your WiFi Knows You’re Sleeping
In partnership with
Your WiFi router may already be watching you. RuView just showed how normal home WiFi signals can track breathing, detect falls, map human skeletons through walls, and potentially turn every apartment into a passive sensing system.
What's on FIRE 🔥
IN PARTNERSHIP WITH LENI
Most AI agents sound smart. Until they confidently hallucinate your financial analysis, spreadsheet logic, or investment research. Leni was built to fix that.
It scored 98% on BullshitBench v2, outperforming every public GPT, Gemini, Grok, and Claude configuration on rejecting nonsense before answering. It forces the model to break claims apart, verify if they’re real, and recalculate outputs first. So you get:
Better financial analysis
More reliable investment workflows
Fewer hallucinations
More trustworthy AI agents
Real production-grade outputs
AI INSIGHTS
As EVs become more common, they’re wearing through tires faster due to their weight, yet service shops are struggling to find people willing to do the "back-breaking" labor. ATI is launching SmartBay to turn a 75-minute chore into a 30-minute automated process.
Unlike older automation that followed rigid paths, the SmartBay system uses AI to adapt to every specific vehicle that pulls into the service area.
The system inspects the vehicle, swaps the tires, and balances the wheels by analyzing real-time data rather than following a pre-programmed script.
In an hour and 15 minutes, a human tech finishes 4 tires. In that same time, a single technician managing three SmartBays can process 24 tires.
The system is designed to fit into a standard 12-foot service bay, making it easy for existing dealerships to upgrade without a total renovation.
Serial entrepreneur Andy Chalofsky is betting that dealerships will prefer a lease over a traditional salary. So ATI leases the system for $4,900 per month. This is often less than the total compensation for an experienced technician, but with 3x the efficiency.
And this system collects data during every change, providing dealerships with real-time insights into vehicle health & tire wear patterns across their entire network.
PRESENTED BY HUBSPOT
Want to get the most out of ChatGPT?
ChatGPT is a superpower if you know how to use it correctly.
Discover how HubSpot's guide to AI can elevate both your productivity and creativity to get more things done.
Learn to automate tasks, enhance decision-making, and foster innovation with the power of AI.
AI SOURCES FROM AI FIRE
1. 7 New Claude Features That Feel Almost Unfair to Use (And Most People Still Don't). This guide shows how to use Scheduled Tasks, Dispatch, and Computer Use to delegate your CEO reviews, research, and admin to Claude.
2. Old But Gold: ChatGPT Image 2 vs Nano Banana Pro: Full Test & Comparison Across Every Use Case. The results aren’t obvious & one hidden weakness changes everything. We use both but for different things!
3. Video: One AI Research Workflow I’d Use Before Starting Any Business Idea | NotebookLM + Perplexity. Use Perplexity to find credible sources, then use NotebookLM to turn those sources into a decision-ready research workspace.
4. Video: Gemini × NotebookLM: The 4-Layer AI Workflow for Research, Strategy, and Execution. So you can move from research to insight to execution without losing context every step of the way.
TODAY IN AI
AI HIGHLIGHTS
💸 Anthropic split Claude agent usage into a separate monthly credit pool. Power users are furious over the tiny limits, and devs already threatening to switch to Codex.
👀 Elon Musk just open-sourced X’s “For You” algorithm on GitHub. You can now see how X ranks posts using Grok-based AI, engagement signals, ads, and spam filters.
⚡ Nous Research released Token Superposition Training, a new method that reportedly cuts LLM pretraining time by 2-3x without changing the final model.
💻 DeepSeek-V3 is getting attention again for running locally, staying fully open-source, and reportedly using far less energy than many Western AI models.
🎬 Runway launched Runway Agent, a new AI assistant that helps users brainstorm, edit, and generate finished videos through simple conversations.
💰 AI Money Moves: Cerebras jumped 90% in its Nasdaq debut, giving the AI chipmaker a valuation above $75B in the year’s biggest IPO so far. The company works with OpenAI and Amazon Web Services, as investors bet the AI chip war against Nvidia is only getting started.
HOT PAPERS OF THE WEEK
1. LLM training may scale through tiny adapter updates
MinT from Mind Lab introduces managed infrastructure for training and serving millions of LoRA policies. Instead of saving every model as a full checkpoint, MinT keeps the base model running and moves small adapter revisions. Big impact: teams may train, test, serve, and roll back many LLM variants over shared 1T-class base models with far less storage and movement cost.
2. LLMs can gain long-term memory without huge context windows
δ-mem introduces a compact online memory system for large language models. It stores past information in a tiny 8×8 memory state and uses it to adjust attention during generation. Key shift: better memory may not require bigger context, full fine-tuning, or replacing the model backbone.
3. Qwen is pushing image models toward real creative workflows
Qwen-Image-2.0 unifies high-quality image generation and precise editing in one model. It improves ultra-long text rendering, multilingual typography, photorealism, and complex instruction following. Big impact: image models are moving closer to usable design tools for slides, posters, infographics, comics, and text-heavy visuals.
NEW EMPOWERED AI TOOLS
🧠 OpenHuman is a local-first, open-source AI workspace that remembers everything about you across sessions, with privacy-first memory and one-click setup.
🕸️ HasData is a web scraping API for AI agents that turns any URL into clean JSON or Markdown, with built-in proxies, rendering, and anti-bot handling.
📈 PHBench predicts which Product Hunt launches are most likely to raise a Series A using data from 67,000+ launches and 528 funded startups.
🎯 Lensmor helps sales teams turn exhibitor data into pre-booked meetings by finding events, decision-makers, verified emails, and outreach opportunities.
AI BREAKTHROUGH
The open-source project RuView (formerly WiFi DensePose) recently surpassed 50K GitHub stars by proving that the WiFi signals already in your home can act as a high-resolution radar.
The system relies on Channel State Information (CSI) to map environmental disturbances into actionable data. FYI, key features include:
Skeletal Tracking: Reconstructs a 17-point human skeleton (COCO keypoints) in real time.
Vital Sign Monitoring: Detects breathing (6–30 BPM) and heart rate (40–120 BPM) from subcarrier phase shifts, even while a subject is sleeping.
Through-Wall Sensing: Penetrates up to 5 meters of concrete, wood, or debris, making it functional in total darkness or smoke.
Automatic Fall Detection: Triggers alerts based on rapid changes in skeletal position, designed for elder care and hospital monitoring.
To run the system for real sensing, you need a few ESP32-S3 microcontroller boards (approx. $9 each). These act as the sensors that feed CSI data to a local computer. Full skeletal pose estimation typically requires a 60-second calibration session to "map" the specific geometry of your room.
Because WiFi signals pass through exterior walls, a neighbor or actor outside a building could theoretically monitor your daily routine without you ever knowing.
We read your emails, comments, and poll replies daily
How would you rate today’s newsletter?Your feedback helps us create the best newsletter possible |
Hit reply and say Hello – we'd love to hear from you!
Like what you're reading? Forward it to friends, and they can sign up here.
Cheers,
The AI Fire Team
Reply