Weekly Magic AI - Open-Source Grok, MM1 AI model, ScreenAI
Top AI news of the week, Magic AI tool of the week, and article of the week
Hi AI Enthusiast,
Welcome to our weekly Magic AI, where we bring you the latest updates on Artificial Intelligence (AI) in an accessible way. This week’s Magic AI tool is a newsletter powerhouse! Be curious! 😀
Top AI news of the week
🤖 xAI Grok-1 is open-source
xAI has made Grok-1, a large language model (LLM), available as open-source. Grok-1 is a 314-billion parameter Mixture-of-Experts model. xAI has published the weights and architecture of Grok-1 under the Apache 2.0 license, making it accessible on GitHub.
xAI trained the LLM, Grok-1, from scratch. Grok-1 is a raw base model from the Grok-1 pre-training phase, which concludes in October 2023. This means Grok-1 is not an instruction-tuned LLM designed for use as a chatbot like ChatGPT. xAI uses JAX and Rust as its tech stack.
You can start using Grok-1 by following the instructions on the GitHub repository.
Our thoughts
We believe in open-source AI. However, this can only be the first step of xAI! In the future, it would be great if xAI publishes instruction-tuned versions of Grok. Meta has made Llama 2 available as instruction-tuned models on Hugging Face! You can use instruction-tuned LLMs, for example, as a Chatbot. Raw base models like Grok-1 need to be fine-tuned for specific tasks.
More information
- Open Release of Grok-1 - xAI Blog
Apple researchers unveil new MM1 AI models
Apple has developed a powerful AI model that understands both text and visuals. The model can scale up to 30 billion parameters, making it significantly smaller than GPT-4 or Gemini. Despite its smaller size, it delivers competitive performance in established benchmark tests.
Training with a mixture of image-text pairs, mixed image-text data, and text-only data enables MM1 to understand image content and answer questions. Look at the following example:
The Apple researchers plead for more openness in the LLM development in the paper. For this reason, Apple discloses all important results of its model development.
Our thoughts
The new Apple model is small but powerful. We agree with Apple that there needs to be more transparency in the development of large language models. Previous papers have not adequately emphasized the processes involved in the pre-training phase. Apple’s paper is a win for the open-source community.
More information
- MM1: Methods, Analysis & Insights from Multimodal LLM Pre-training - arXiv
- Apple debuts its MM1 multimodal AI model with rich visual capabilities - The Decoder
📊 Google Research presents ScreenAI - A visual language model for UI
Google Research has presented a new visual language model. The model attempts to understand visual graphics like graphical user interfaces (UIs), tables and infographics. LLMs should be able to use this information to understand a website, answer questions about it and navigate its interface.
The blog article on ScreenAI describes nicely how the model works. For more in-depth references, we recommend reading the paper “ScreenAI: A Vision-Language Model for UI and Infographics Understanding” .
Our thoughts
AI is expanding into various fields. In the future, humanoid robots will need to tackle tasks similar to humans, which means they must also understand UI elements. AI development is advancing non-stop!
More information
- ScreenAI: A visual language model for UI and visually-situated language understanding - Google Research Blog
- ScreenAI: A Vision-Language Model for UI and Infographics Understanding - arXiv
Magic AI tool of the week
As content creators, we’re always looking for tools to grow our online business. In the last months, we have tried out several email marketing tools. Beehiiv* has really impressed us. Today, we want to show you the AI writer from Beehiiv Newsletter Platform.
Since ChatGPT was released, more and more companies have integrated AI features into their tools. AI can make you more productive if you use it right. In a Beehiiv Newsletter, you have the option to use the AI Writer as an assistant. To use this feature, you need the Scale Plan.
With the AI Writer, you can generate texts. In our example, we entered the following input prompt.
Write me a text about the current Bitcoin developments.
The AI has generated a short text on the current Bitcoin developments for us. This feature is great for creating a first draft. However, it is crucial to carefully review the generated text. AI applications can sometimes provide incorrect information. Please keep that in mind!
Article of the week
Are you looking for the ideal tool to boost your online presence? Look no further! We’ve sifted through numerous email marketing tools and discovered a standout - Beehiiv*. This newsletter platform has many great features, and we’re excited to share our experience with you.
One of our newsletters has found a new home at Beehiiv*, opening doors to incredible growth opportunities. Would you like to start your own newsletter and get in touch with your customers? So what are you waiting for?
Discover more about Beehiiv and how it can transform your newsletter game in our Beehiiv review.
💡 Do you enjoy our content and want to read super-detailed articles about AI? If so, subscribe to our blog and get our popular data science cheat sheets for FREE.
Thanks for reading, and see you next time.
- Tinz Twins
P.S. Have a nice weekend! 😉😉
Leave a comment