- AI Fire
- Posts
- 🚨 AI Jobs Flip Overnight
In partnership with
Can AI really replace whole job categories overnight? Early signals say yes. Google pushed a new reasoning model. Meta bought a company that captures your entire day. And AI-written GPU kernels just beat NVIDIA’s best code.
What are on FIRE 🔥
IN PARTNERSHIP WITH LEVANTA
The Future of Shopping? AI + Actual Humans.
AI has changed how consumers shop by speeding up research. But one thing hasn’t changed: shoppers still trust people more than AI.
Levanta’s new Affiliate 3.0 Consumer Report reveals a major shift in how shoppers blend AI tools with human influence. Consumers use AI to explore options, but when it comes time to buy, they still turn to creators, communities, and real experiences to validate their decisions.
The data shows:
Only 10% of shoppers buy through AI-recommended links
87% discover products through creators, blogs, or communities they trust
Human sources like reviews and creators rank higher in trust than AI recommendations
The most effective brands are combining AI discovery with authentic human influence to drive measurable conversions.
Affiliate marketing isn’t being replaced by AI, it’s being amplified by it.
AI INSIGHTS
Right now AI looks like a cheap assistant. Next year it becomes a full worker. And the drop in human jobs happens fast, not slow.
The timeline says everything:
2022 → ChatGPT gives okay answers
2023 → Better writing, better coding, still forgets everything
2024 → Agents start checking their own work
2025 → AI replicas of you that join Zoom calls, work 7 hours straight, never miss
2026? → Any job done behind a screen, done better by an AI
Costs collapse at the same time:
GPT-3: $600 per 1M words
GPT-5: $10
Next wave: $0.10
Cheaper, smarter, persistent.
And suddenly the “just helping me write emails” tool replaces whole workflows.
Early signals:
Amazon exploring automation across 600k jobs
IMF: 60% of roles impacted
Replit jumped from $1M → $250M by letting anyone build full apps with AI
Agents can now run hours of work, not seconds
The tipping point: When an AI can read your emails, join calls, draft replies, update dashboards, and translate across 100 languages for $1,000/year → $100/year → $10/year. Why hire a junior?
Why it matters: This shift arrives like a software update. One week AI is a helper. Next week it’s the most reliable worker in the company. The window to adapt is getting small.
PRESENTED BY BRIGHT DATA
84% Deploy Gen AI Use Cases in Under Six Months – Real-Time Web Access Makes the Difference
Outdated or blocked web data means more time fixing, less time building. Bright Data delivers real-time data to your AI agents so you can scale, improve reliability, and drive growth.
AI SOURCES FROM AI FIRE
1. 10 nodes to master AI automation. Learn the core n8n building blocks that power 80% of real AI workflows and skip months of confusion.
2. 7 advanced nodes that give your automations a brain: Code, Webhooks, AI Agents. Learn the exact stack behind real autonomous workflows.
HOT PAPERS OF THE WEEK
DeepSeek-V3.2: Open LLM Reaches Olympiad Level
DeepSeek-V3.2 narrows the gap with closed models using Sparse Attention and scalable RL. Its Speciale variant wins gold at IMO & IOI, boosting open-agent research. (DeepSeek-AI)From Code Models to Code Agents
A massive multi-institution survey maps the full lifecycle of code LLMs - from training and alignment to safety and deployment - offering practical guides for real-world coding agents. (Monash • CAS • ByteDance • Huawei • NTU)PaperDebugger: Multi-Agent Editing Inside Overleaf
NUS builds PaperDebugger, an in-editor system where agents critique, rewrite, and patch LaTeX docs live. It handles versioning, references, and structured reviews in real time. (NUS)Live Avatar: Real-Time Infinite Streaming Avatars
Alibaba’s Live Avatar runs a 14B diffusion model that streams lifelike avatars at 20 FPS for 10,000 + seconds, syncing perfectly with real-time audio. (Alibaba • USTC • Zhejiang University)
TODAY IN AI
AI HIGHLIGHTS
🍏 ChatGPT is still huge, but its growth just slowed to 6% from August to November, while Gemini jumped 30% in the same window. OpenAI even called a “code red” to push new features. The gap is closing fast.
⚠️ Full Fact uncovered AI deepfake doctors across TikTok, X, Facebook, and YouTube driving traffic to Wellness Nest. Real experts. Fake words. Menopausal women are targeted with unproven supplements. Platforms are scrambling to catch up.
🧠 Google launched Deep Think inside Gemini, a new reasoning mode that runs multiple internal hypotheses before answering. It hits 93.8% on GPQA Diamond and 45.1% on ARC-AGI-2. Ultra users can flip it on in the Gemini app.
🚀 Google Research revealed Titans, a long-context architecture that uses “surprise” to decide what to remember. It beats strong baselines on BABILong and scales past 2M tokens. Think transformers plus deep neural memory.
📝 NotebookLM now supports 10,000 characters for chat personas instead of 500. You can script detailed roles like PMs, teachers, or lab researchers. Google even shared starter prompts you can plug in right away.
💰 Hot AI Acquisition: Meta has acquired Limitless, the AI-wearables startup backed by Sam Altman and A16z. The device’s recording and memory tech will support Meta’s next-gen AI hardware, alongside its Ray-Ban and Oakley smart glasses. Meta also added former Apple design lead Alan Dye to sharpen its hardware push.
NEW EMPOWERED AI TOOLS
📘 Documentation.AI creates and updates product docs with AI, keeping content current and support load low.
🧠 Gemini 3 Deep Think is Google’s most advanced reasoning model, built for math, logic & complex problem-solving.
🏢 Google Workspace Studio lets you build AI agents in Gmail, Docs & Sheets, automating daily workflows in minutes.
🎮 Orca is an AI game-dev agent that codes, debugs, and builds playable prototypes straight from chat.
AI BREAKTHROUGH
CUDA-L2 is an LLM+RL system that auto-writes GPU matmul kernels. It beats cuBLAS/cuBLASLt across 1,000 shapes used in Llama, Qwen, DeepSeek.
Speed gains.
+22% vs
torch.matmul+19% vs cuBLAS
+11% vs cuBLASLt-AutoTuning
And NVIDIA’s tuner already tests 100 algorithms. The AI still wins.
It also finds expert-level tricks on its own: Double buffers, smart swizzles, multi-step prefetch, and switching between WMMA and CuTe.
Method: Pretrain on real CUDA code → RL with speed rewards → strict correctness → benchmark everything.
Bottom line: AI can now write faster GPU kernels than humans.
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