Generative AI - Short & Sweet
Posts
How Grok Outpaced r1 and o1 in Real-Time Reasoning

How Grok Outpaced r1 and o1 in Real-Time Reasoning

Plus: Quantum Breakthrough, Google Shake-Up, and the AI Song Everyone’s Talking About

Martin Musiol
November 22, 2024 • Estimated Reading Time: 7 minutes

In partnership with

Speed is key in intelligence—see how Grok left r1 and o1 behind in real-time reasoning.

This issue also explores THE quantum computing milestone, Google’s possible shake-up, and the must-listen AI song. The quality is getting insane.

10x Your Outbound With Our AI BDR

Imagine your calendar filling with qualified sales meetings, on autopilot. That's Ava's job. She's an AI BDR who automates your entire outbound demand generation.

Ava operates within the Artisan platform, which consolidates every tool you need for outbound:

300M+ High-Quality B2B Prospects
Automated Lead Enrichment With 10+ Data Sources Included
Full Email Deliverability Management
Personalization Waterfall using LinkedIn, Twitter, Web Scraping & More

Book a demo and supercharge your sales team.

(✨ If you don’t want ads like these, Premium is the solution. It is like you are buying me a Starbucks Iced Honey Apple Almondmilk Flat White a month.)

DeepSeek’s R1-Lite—China’s answer to OpenAI’s o1

First OpenAI’s o1, now DeepSeek’s R1—the AI reasoning race is heating up.

Chinese AI firm DeepSeek just launched R1-Lite-Preview, a reasoning model that rivals OpenAI’s o1 and offers a transparent, real-time chain-of-thought (CoT) process.

Unlike o1’s condensed summaries, R1-Lite-Preview lets you peek into its full reasoning as it happens.

It shares even its feelings 😄

I tested it with a tough math riddle:

“Three positive integers form a geometric sequence. Their sum is 91, and their product is 1,365. What are the numbers?”

R1-Lite-Preview took over two minutes, sharing an extensive (and sometimes amusing) internal dialogue. It even admitted, “This is frustrating.”

In the end, it couldn’t solve the problem.

OpenAI’s own reasoning model also couldn’t. And Google’s Gemini? Forget it.

Surprisingly, Elon Musk’s Grok 2 nailed it quickly, giving the correct answer: 7, 21, and 63. The comparison video:

I've tested the reasoning models o1, and DeepSeek's r1 by giving them a hard math riddle.
Results: both (+Gemini) were wrong, but xAI's Grok 2 solved it with its first go
(To my knowledge, Grok 2 isn't a reasoning model architecture.)
@OpenAI@deepseek_ai@grok@elonmuskx.com/i/web/status/1…
— Martin Musiol (@musiol_martin)
4:01 PM • Nov 22, 2024

Why it matters

DeepSeek’s advancement shows how rapidly AI reasoning is evolving globally. With an estimated 50,000 H100 chips, they’re matching Western AI labs in compute power.

By open-sourcing R1, DeepSeek accelerates innovation across the AI industry, challenging closed US AI labs.

As reasoning models become widespread, the competition isn’t just about intelligence—it’s about speed.

Inference latency is the new battleground.

In a world where powerful reasoning models are accessible to all, the edge goes to those who can think and act faster!

The AI arms race is shifting from who thinks best to who thinks best, but fastest.

Try out r1 yourself.

Release statement & more info.

AlphaQubit: DeepMind’s Quantum Error Correction Breakthrough

Google DeepMind’s AlphaQubit is revolutionizing quantum computing by improving error correction accuracy using AI.

Qubits, the building blocks of quantum computers, are highly sensitive to noise, requiring error correction to ensure reliability.

AlphaQubit employs GPTs (it’s basically a language model) to predict and correct errors better than traditional methods. Tested on Google’s Sycamore processor, it achieved:

6% fewer errors than tensor networks.
30% fewer errors than correlated matching.

While AlphaQubit excels, challenges include real-time speed improvements and scalability to millions of qubits.

Why it matters: This innovation brings quantum computing closer to real-world impact, enabling, e.g., drug discovery and material science breakthroughs.

I used to spend four hours a day answering emails; now, it’s done in 30 minutes

I’m getting 3 1/2 hours back daily because I use Superhuman AI.

While I answer most emails with Shortcuts, I have AI writing the answers for me in my voice.

Feel free to check it out here.

It’s a game-changer for productivity, allowing me to focus on what truly matters while maintaining a personal touch in my communication.

DOJ Eyes Google’s Search Dominance—What’s Next for Chrome?

The U.S. Department of Justice (DOJ) is intensifying its antitrust scrutiny of Google’s search monopoly.

While there’s no official mandate yet, industry chatter suggests that Google might be pressured to separate its Chrome browser to level the playing field.

Speculation is rife about who could acquire Chrome if it becomes available.

OpenAI’s name has surfaced, though no concrete evidence of their interest exists. Tech giants like Microsoft, Amazon, Apple, and Meta could also be interested.

A new owner for Chrome would accelerate AI integration. Exciting times ahead.

ChatGPT “Creative Writing“ Mode + Suno AI = Absolute Banger

Hey @OpenAI your @ChatGPTapp 4o "Creative Writing" upgrade today is INSANE!
My test prompt for two years has been some version of "write an Eminem style cipher about quantum mechanics."
ChatGPT has always been the best at this, but NO LLM has ever captured the sophisticated… x.com/i/web/status/1…
— Kyle Shannon 🍓 (@kyleshannon)
9:56 PM • Nov 20, 2024