🚀 Mistral AI Unveils 123B Open Model

PLUS: OpenAI Free Fine-Tuning

Author

The Summary AI
July 23, 2024

Welcome back!

In the escalating AI arms race, Mistral AI has unveiled its 123B open model, directly competing with Llama 3.1 and GPT-4o. This major shift towards open-source AI is pushing the boundaries of innovation and challenging closed-source leaders. Let's unpack this...

Today’s Summary:

  • 🦾 Mistral AI launches 123B open model

  • 🆓 OpenAI offers free GPT-4o Mini fine-tuning

  • 🚀 Meta AI assistant now using Llama 3.1

  • 🎨 Adobe updates Firefly AI tools

  • 🤖 Elon Musk sets Optimus sale date

  • 🌍 KLING AI generator goes global

  • 🆕 2 new AI tools: dating and avatars

TOP STORY

AI arms race escalates: Mistral AI drops 123B open model, takes on Llama 3.1 and GPT-4o

The Summary: In a major push for open-source AI, Mistral AI has released Mistral Large 2, a 123B parameter model with similar performance as GPT-4o and Llama 3.1 405B. This follows closely on the heels of Meta's Llama 3.1 405B release. These open models now compete at the frontier of AI capabilities, accelerating innovation.

Key details:

  • Outperforms all previous versions in code generation and math

  • Trained to acknowledge knowledge gaps, reducing hallucinations

  • Supports function calling and advanced retrieval capabilities

  • Fits on a single H100

  • Available through Mistral platform and major cloud providers

Why it matters: The current open-source push is challenging the dominance of closed AI models, democratizing access to frontier AI. This could launch a new wave of innovation. With open models closing the performance gap, we may see increased competition, faster progress, potentially forcing fast reactions from closed-source leaders.

OPENAI

OpenAI fires back with free GPT-4o Mini model customization

The Summary: OpenAI is offering free fine-tuning of GPT-4o Mini for custom applications. This move comes right after Meta Llama 3.1 release. The offer, available to top-tier users with plans to expand, includes 2 million free training tokens daily until September 23.

Key details:

Why it matters: This move shows an acceleration in AI business models. By offering freemium AI customization, OpenAI may be setting the stage for an AI ecosystem where models become progressively commoditized, but advanced features and compute remain premium.

META

Meta AI assistant leaps forward with Llama 3.1

The Summary: Meta has released the new version of its assistant Meta AI across WhatsApp, Messenger, Instagram, Facebook, and Ray-Ban Meta powered by the new open-source model Llama 3.1, with 405 billion parameters, rivaling closed-source competitors like GPT-4o. Meta AI now offers enhanced creative tools, multilingual support, and advanced reasoning capabilities for complex tasks such as coding and math problems.

Key details:

  • Meta AI available in 22 countries (not EU), adding 7 new languages

  • New "Imagine me" feature creates personalized images based on user prompts

  • Llama 3.1 405B model offers performance comparable to GPT-4o across various tasks

  • Meta AI to be integrated into Meta Quest devices next month

Why it matters: Meta's open-source approach is challenging the AI industry status quo, democratizing access to advanced AI. This move will accelerate innovation in the field and pressure competitors to reconsider their strategies.

QUICK NEWS

Quick news

TOOLS

🥇 New tools

  • Hey AI - No-swipe AI match dating

  • HeyGen - Chat with cutting-edge avatars in real-time

That’s all for today!

If you liked the newsletter, share it with your friends and colleagues by sending them this link: https://thesummary.ai/