Microsoft’s OmniParser and Apple’s Ferret-UI
Microsoft has launched OmniParser, a new tool that makes understanding web interfaces much better than before. This model is even better than GPT-4V at parsing, and it’s aimed at helping with web automation. You can find it on Hugging Face!
At the same time, Apple has released Ferret-UI, a special language model made for iPhones and iPads. It’s great at recognizing icons and translating text on screens. You can find more information about it in their research paper and demo.
Google and UNC’s Generative Infinite Games
In another exciting partnership, Google and the University of North Carolina (UNC) have introduced Generative Infinite Games. This new type of interactive experience combines game mechanics with generative models. It’s a fresh way to create immersive and dynamic gaming. You can check out the details in a tweet by Nataniel Ruiz!
Cohere’s Aya Expanse Multilingual Models
Cohere has released Aya Expanse, a series of top-notch multilingual models that cover 23 languages. These models aim to improve AI language understanding and are available for anyone to use on Hugging Face.
AIatMeta’s Llama 3.2 Quantized Version
AIatMeta has introduced a new, smaller version of the Llama 3.2 model. This new version uses a special training method that makes it faster (2 to 4 times quicker) and takes up less space. It’s perfect for devices with limited resources. You can read more about this on MiraclePlus.
New Analytical Tools for Claude
The AI model Claude has some fantastic upgrades! It can now write and run code better, thanks to a new analytical tool. This tool helps provide clear and accurate answers. Plus, it can create interactive data visualizations using Artifacts. Learn more about this update on MiraclePlus!
Meta Open Materials 2024
Meta FAIR has launched Meta Open Materials 2024, an open-source model and dataset. This tool is aimed at discovering new inorganic materials. It uses a special architecture called EquiformerV2 and comes in three sizes. The dataset includes over 100 million calculations. You can find more details on Hugging Face and in their announcement.
https://huggingface.co/fairchem/OMAT24
Unbounded: A Generative Game
A new paper has introduced Unbounded, a character life simulation game created entirely with a generative model. It uses a streamlined LLM for dynamic game mechanics and includes a regional IP adapter for consistent visuals. Check out more on Twitter!
3D Printing Gaining Traction in Hardware Companies
There’s a notable trend happening in the hardware industry. Every hardware company in the latest Y Combinator batch is using 3D printing in their products. This shows that 3D printing technology is finally being recognized for its potential, as highlighted in a tweet by Paul Graham.
Waymo’s Self-Driving Cars
Despite the excitement around self-driving cars, Waymo has fewer than 1,000 self-driving vehicles on the road. It’s surprising given that the technology has been around for over two years. Paul Graham discussed this slow rollout in a tweet.
HuggingFace & GitHub: Language Model and Innovation
Aya Expanse is a powerful multilingual language model made by Cohere For AI. It supports 23 languages, like English, Chinese, and Spanish. With 32 billion parameters, it’s hosted on Hugging Face. The model uses a special Transformer architecture designed for different languages and is shared under a CC-BY-NC license.
https://huggingface.co/CohereForAI/aya-expanse-32b
Interaction Between Games and AI
The Mindcraft project, created by kolbytn on GitHub, lets AI interact in Minecraft using the Mineflayer API. This project allows AI models to write and run code on a computer, but it comes with security risks. Users must have Minecraft Java Edition and Node.js to use it.
https://github.com/kolbytn/mindcraft
Live TV and Streaming
Guovin/TV is an IPTV tool that helps you watch live TV. It includes channels from CCTV, Satellite TV, and areas like Guangdong, Hong Kong, Macao, and Taiwan. The tool updates automatically two times a day and allows you to add your own channels and icons. It works with playback software like TVBox and can be set up in different ways, like using workflow, Docker, or the command line.
Virtual Machines and Infrastructure
OpenVMM is a virtual machine monitor written in Rust. It’s mainly for the OpenHCL paravisor and is hosted on GitHub. The project is open to contributions but requires a signed Contributor License Agreement (CLA). It follows the Microsoft Open Source Code of Conduct.
https://github.com/microsoft/openvmm
Reddit Insights
Ray Kurzweil suggests that we might soon be able to back up our brains to survive accidents. This raises many questions and discussions about the idea of backing up human consciousness:
Concerns About Consciousness Transfer: Some people worry if consciousness can really be transferred. A backup might create a copy, but the original would still be lost. This brings up tricky questions about identity and continuity.
Economic and Accessibility Issues: Initially, this technology might only be available to the wealthy, making it less accessible to others. This could lead to unequal access to life-extending technologies.
Technological Skepticism: Many people are unsure about what consciousness really is. Since we don’t have a clear definition, the idea of saving consciousness seems unclear and confusing.
Ethical Implications: Creating a copy of oneself brings up important questions about identity and the rights of both the original and the copy, especially if both exist at the same time.
