- AI Fire
- Posts
- ⚙️ Huawei’s Chip Could Break Moore’s Law
Sponsored by
A new Stanford HAI study exposing how automated AI hiring platforms are creating algorithmic black boxes of systemic racial bias, locking us out across entire industries.
What's on FIRE 🔥
IN PARTNERSHIP WITH SECTION
AI has shortened the time to ship for engineering teams significantly. But how do you scale good judgement along with the codebase?
Find out on June 9 at this free event where you'll learn about the new operating model for engineering teams when AI becomes the primary developer.
AI INSIGHTS
Instead of fighting TSMC and Intel on transistor size (the classic Moore's Law), Huawei is introducing the Tau (τ) Scaling Law (also nicknamed Her’s Law). Their idea is to start shrinking the time it takes for electrical signals to travel inside the machine:
LogicFolding: This is their new circuit-level trick. They are folding and placing related logic blocks much closer together.
They massively reduce electrical resistance and parasitic capacitance just by making the critical internal wires shorter.
They plan to reach 1.4nm-class density by 2031, without owning the world's best lithography tools.
The next Kirin phone chip will be the first full test of this LogicFolding architecture in the wild.
Huawei is essentially saying, "If we can't make the cars any smaller, let's just make the highway perfectly straight." It is a massive gamble, but they claim to have already built 381 chips using this thinking.
Being cut off from TSMC forced Huawei into a corner, and they decided to engineer a completely new way out. Whether this becomes an industry standard or just a clever survival tactic, that next Kirin chip is going to be incredibly fun to benchmark.
PRESENTED BY LINDY AI
Get 2 hours back. Every day.
The average professional spends 28% of their workday on email. The other 72% is spent recovering from it.
Lindy is an AI assistant that reads every email, sorts out the noise, and drafts replies that sound like you. Before calls, it texts you a brief over iMessage with context from your last conversation. You text it back like a friend. And it only takes one minute to set up.
AI SOURCES FROM AI FIRE
1. Free: [AI Personal Brand Playbook] Introduction: Kickstart Your AI Branding Journey with Claude & Canva. Step by step, we’ll guide you to create a complete AI-powered personal brand workflow with Claude, Canva,…
2. Pro: From Idea to $1M ARR: Claude Code’s High-Value Problem-Solving Strategy. The real AI SaaS playbook for 2026. Use Claude Code to discover profitable problems, generate solutions, simulate ideas, and ship enterprise software fast.
3. Free: Claude Agent Team Setup: What I Learned After Getting It Wrong. A step-by-step guide to setting up roles, prompts, and workflows that actually produce better results.
4. Video: How to Convert a Coding Project from Claude Code to Codex. Claude Code is great for starting projects, but long sessions can make it stupid. We show how to transfer your work to Codex so it continues smoothly.
Thank God, Google Flow just dropped its biggest update yet, character consistency is finally as easy as eating a cake.
We all know that Runway is famous for character consistency and performance capture, while Sora also showed off reusable characters before it shuts down. But Flow locks in your character's identity and voice from the start.
→ You blend real footage with generated content, tweak conversationally, and iterate fast. That's the part Runway still makes you work hard for. How to set it up:
Lock in a base reference image and add secondary angles correctly
Build a custom voice profile from scratch
Write Character Info that actually controls on-screen behavior
Call your character into any image or video prompt
Test character consistency across static images
Handle policy violation errors (or you can’t do anything with Google’s tools)
TODAY IN AI
AI HIGHLIGHTS
😂 Google’s AI Overview still can’t reliably count letters in simple words. At one point, it claimed “Google” has 2 Ps and spelled “Trump” as “t-r-p-u-m.” Real proof here.
🤦 Funny enough, we have a new AI detector apparently. People are calling out this obvious phrase as the fastest way to make writing feel AI-powered. Not the em dash.
📱 Instagram, Facebook, WhatsApp & Meta AI now have paid plans. These unlock extra creator tools, custom profiles, analytics, even AI features across the ecosystem.
🚀 Qwen’s smartest model, Qwen3.7 Max, landed in Go plan. It handles 2,000-page contexts, runs agents for 35 hours straight, even plugs directly into Claude Code.
🧠 Anthropic researcher Sholto Douglas revealed Mythos just solved the famous Erdős Problem #90 too. Same result as GPT-5.5, but with a "cute, simple” proof.
🏭 Figure AI just signed a major deal to deploy humanoid robots inside Catalyst Brands’ logistics network. JCPenney and Brooks Brothers warehouses may be next.
💻 Former researchers from OpenAI, Apple, Google DeepMind raised $15M for a startup called Trajectory. Their goal: AI that improves itself from real user feedback.
💰 Big AI Fundraising: OpenRouter raised $113M after exploding to 8M developers. Their single API now handles a staggering 1.5 quadrillion AI tokens annually as model demand keeps surging.
Willo lets you launch a real online business with AI agents running everything for you. No coding. No hiring. No figuring out 15 different tools. Just describe your idea, and Willo starts building everything you need:
No technical skills needed
Free to start
No credit card required
Built-in Growth Engine that improves automatically
Launch faster than most startups build a logo
You get a full 7-agent AI team handling your business 24/7 while you stay in control. This feels like hiring an entire company instantly!
NEW EMPOWERED AI TOOLS
⌚ Bluedot 2.1 can record conversations directly from your Apple Watch, then sync them with Claude through MCP.
🧠 Powabase builds AI apps with Postgres, RAG, agents, memory, workflows, and automation primitives in one platform.
🔌 zero.xyz gives your AI agent access to ~8k tools, APIs and services: Claude Code, Codex, Gemini, OpenClaw, Hermes, other CLI agents.
🌐 Oasis Browser is the privacy-first AI browser you can train (anonymously). Your data is your data.
AI BREAKTHROUGH
Letting AI scan resumes sounds like the ultimate time-saver for any growing company. But a massive new study out of Stanford showed that it is not a good look.
The researchers dug into 4 million job applications across 156 employers using an AI screening tool (specifically Pymetrics). Here’s a breakdown of what’s happening:
Around 10.6% of the job positions showed a clear negative bias against Black applicants, and 5.3% against Asian applicants.
Because 42 of these AI models were shared across different employers, getting rejected by one algorithm could essentially trigger a rejection at a completely different company using the same system.
Roughly 4% of candidates who applied to 10 different positions were rejected from all of them. That is a much higher fail rate than if human recruiters were making independent choices.
Researchers rightly point out that this data is from 2018 to 2022. Today's hiring wave is increasingly powered by LLMs, which function completely differently.
When multiple companies rely on the exact same third-party AI vendor, they also inherit the exact same blind spots. Always keep a human in the loop when it counts.
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