Bot Eat Brain
Posts
Grok 2 and Grok-2-Mini are faster and more accurate

Grok 2 and Grok-2-Mini are faster and more accurate

PLUS: NVIDIA & Mistral's new model

Michael Parrish
August 26, 2024

TOGETHER WITH

Good morning, human brains, and welcome back to your daily munch of AI news.

Here’s what’s on the menu today:

Elon Musk’s new, faster models 🦾 🤖
xAI's Grok-2 and Grok-2 mini got a major speed upgrade.
Create AI tutorial videos instantly 🎥 🕺
Guidde offers effortless video documentation.
NVIDIA & Mistral sitting in a tree… 👩‍❤️‍💋‍👨 ❤️
They unveiled Mistral-NeMo-Minitron 8B.

New here? Subscribe!😎

Sponsor Bot Eat Brain and reach over 20,000 readers.

(Now 50% off)🤯

Peep today's ‘What Would You Do?’ at the bottom. 👇

MAIN COURSE

Elon Musk’s new, faster models 🦾 🤖

On Friday, xAI's Grok-2 and Grok-2 mini got a major speed upgrade.

Source: xAI

What's the big deal?

Two xAI devs rewrote the entire inference stack in just three days, doubling Grok-2 mini's speed and making the full Grok-2 faster and more accurate.

How'd they pull it off?

They used SGLang, an open-source system that supercharges language model performance.

Where does Grok-2 stand now?

It's nipping at ChatGPT-4o's heels, sitting at #2 on the Lmsys Chatbot Arena leaderboard. Grok-2 mini is coming in at #5 as well.

Why do I care?

This shows how fast xAI’s technology is evolving. Today's speed bump could be tomorrow's breakthrough in AI capabilities.

For more info on Grok 2, click here.

Revolutionize Your Documentation Process.

Tired of spending weeks creating training videos?

Meet guidde, your all-in-one solution for effortless video documentation.

With guidde, anyone in your organization can create, edit, and publish professional-quality videos in minutes.

Whether you're onboarding new hires, training customers, or building a knowledge base, guidde streamlines your workflow.

Key Features:

Intuitive drag-and-drop video editor
AI-powered voiceovers in 35+ languages
Auto-generated subtitles
Branded content with customizable templates
Real-time editing of published videos

Benefits:

Save time and resources
Ensure consistency across your organization
Increase engagement with interactive content
Reduce support tickets with self-serve video guides
Measure impact with detailed analytics

Ready to transform your training and documentation process? Discover how you can create professional video content in less time than it takes to brew your morning coffee.

SIDE SALAD

NVIDIA & Mistral sitting in a tree… 👩‍❤️‍💋‍👨 ❤️

Yesterday, NVIDIA and Mistral unveiled Mistral-NeMo-Minitron 8B.

Source: NVIDIA

What's the big deal?

This AI model is small enough to run on your laptop, but smart enough to compete with the big boys. It tops nine benchmarks for models its size.

How'd they pull it off?

They used some tricks:

Pruning: Trimming the fat from larger models
Distillation: Retraining on a smaller dataset to boost accuracy

Why should I care?

This could bring advanced AI to your personal devices. Think faster responses, better privacy, and AI that works even when you're offline.

YOUR DAILY MUNCH

Cool Tools 🛠️

Mammouth

Get access to the best AI models in one place for $10.

Mammouth gives you access to the best LLMs:

Claude 3.5 Sonnet, GPT-4o, Llama 3, Gemini Pro, Mistral Large, and more.

Together with the best AI generated images:

DALL-E
Midjourney
Stable Diffusion
and more for only $10

Startup News 💰

Microsoft released new open-source AI models in its Phi 3.5 series. The models are called mini-instruct, MoE-instruct, and vision-instruct.

GPT-4 can model basic amino acid and protein structures with surprising accuracy. This is despite it not being specifically trained for structural biology.

Research 👨‍🔬

The Minitron Approach — compresses LLMs like Llama 3.1 8B and Mistral NeMo 12B into smaller models using depth and width pruning techniques.

Transfusion — a multi-modal model that seamlessly handles both discrete (text) and continuous (images) data by integrating language modeling for text with diffusion techniques for images, all within a single transformer architecture.