- Micro Empires
- Posts
- How To Use A Million AI Experts
How To Use A Million AI Experts
Weekly #111 | The rise of Small Language Models (SLMs), PEER & MoME
Table of Contents
Overview
Over the last week, I stumbled upon some interesting work in the AI field.
Pioneering research in Large Language Models (LLMs) has given rise to PEER (Parameter Efficient Expert Retrieval) – a new approach that unleashes the power of “Millions of Micro Experts” (or MoME’s), based on leading research by Xu Owen He from Google DeepMind.1
Imagine a million-strong army of micro experts, each a master of their specific niche, working tirelessly for your business. That's the PEER advantage. It's not science fiction; it's the latest in AI research, empowering you to outperform industry giants with unparalleled efficiency and precision.
It also brings forward discussions about Small Language Models (SLMs) becoming more efficient than LLMs — similar to what we can see in the real world — especially with OpenAI Labs’ recent release of ChatGPT-4o mini:
🤯 GPT-4o mini is a game changer for SLMs and guardrails!
We've benchmarked @OpenAI 4o mini's performance on our upcoming set of Guardrail evaluations, and here's what we saw:
If you need to run guardrails or evals at inference time, 4o mini looks like a cheap, fast, comparable… x.com/i/web/status/1…
— Aparna Dhinakaran (@aparnadhinak)
11:25 PM • Jul 22, 2024
But here’s how PEER dismantles the advantages of LLMs and larger players (key for micropreneurs building their own micro-empires):
Traditional AI often struggles with the mountain of customer data you possess. PEER's micro-analysts act like bloodhounds, sniffing out hidden trends and preferences with laser focus. You'll gain an intimate understanding of your audience, allowing you to craft targeted messaging that resonates deeply and drives sales. You can see this starting to be applied with products like CommonRoom (via “signal stack” tracking):
2 — Inventory Nirvana Achieved
Mixture of Million Experts (MoME)
My team and I are captivated by this new AI model architecture. MoME's modular design unlocks never seen before potential for P2P machine learning systems. While improving efficiency.
1. What it is
1.1. MoME is made of millions of specialized… x.com/i/web/status/1…— bidhan roy 🥯 (@bidhanxyz)
4:01 PM • Jul 18, 2024
Stop wasting money on overstocking or scrambling for missing items. PEER's micro-inventorians are like tireless elves, meticulously analyzing sales patterns, seasonal fluctuations, and supplier performance. They ensure you have the right products at the right time, keeping customers happy and profits soaring.
3 — Operational Efficiency Unleashed
From streamlining projects to automating workflows, PEER dissects your operations with a surgeon's precision. Its micro-optimizers pinpoint bottlenecks and inefficiencies, then prescribe the perfect solution. Imagine a project management mastermind ensuring projects stay on track and within budget. That's the PEER difference.
Google Deepmind's new PEER architecture uses a million tiny experts to boost AI efficiency.
— Techpresso (@techpresso_en)
9:49 PM • Jul 19, 2024
The Trend is Clear: Small is the New Big
Just like PEER's Small Action Models (SLMs) outperform their bulky LLM counterparts, micropreneurs have a built-in advantage – agility. You can adapt, innovate, and outmaneuver the competition with laser focus.
And if you’re wanting to apply all the above today, check out this great series from Riley Brown, where he goes through how to build your own web app using AI, with no prior coding skills:
Longer Video:
Using Claude to make a fully functional note-taking app with a real-time backend in just 4.5 hours without writing a single line of code.
------------------------------------------------------
[Time stamps]
00:00 Designing the App Layout with @AnthropicAI Claude… x.com/i/web/status/1…— Riley Brown (@rileybrown_ai)
10:18 PM • Jul 18, 2024
The future is micro-powered. Apply PEER and watch your micro-empire flourish. Remember, in the business world, David with a team of PEER experts can conquer any Goliath.
Google DeepMind’s groundbreaking research levels the playing field, and you hold the winning ticket.
Reply