Farhan Hammad 一 7 Jan 2024
Pilot Edition #3
Note: This newsletter is published under the pilot edition until all segments are integrated.
New Sections: Highlights & Under The Radar (one project worthy of recognition)
Today’s Edition at a glance:
Mobile Aloha - The ChatGPT Moment for Robotics
Microsoft Phi-2 Allowed for Commercial Use
New Linux Release
New model releases, a new AI agent framework & more.
Read Time: 6 Minutes
Jan 3rd, 2024 - Robotics got its first major breakthrough of the year with Mobile ALOHA, the first ever hardware capable of carrying out everyday tasks. The paper was recently published and the code for the hardware and the learning algorithms used to train the model was open sourced.
Mobile ALOHA learns through various imitation learning algorithms and is trained on 50 human demonstrations for every task. The paper demonstrates the robot carrying out various chores from cooking shrimp, washing dishes to high-fiving humans.
Jan 7, 2024 - Microsoft' Phi-2, one of the smallest but smartest thinking transformer models changed its license recently to MIT. The model was previously operating under a non-commercial license but given the model’s capabilities on code, QA & chat-related tasks, it’s a change that allows the community to build apps on top of it.
3. Linux updates its Kernel to 6.7
Jan 7, 2024 - The newest release of Linux kernel 6.7 is out with new security improvements and a new file system. The new bcachefs file system rivals Btrfs and ZFS while maintaining the speed of EXT4 and XFS. The release includes notable networking enhancements, improved file system features for Btrfs and EXT4, and various hardware support updates. Additionally, security improvements such as AES/XTS performance enhancements and updates to the AppArmor security module are incorporated. The Linux 6.7 kernel is now available for download, with a short-lived support window before being succeeded by Linux kernel 6.8, expected to be released in mid-March 2024.
Tencent released a new expansion of LLaMA2-7B, the LLaMA-Pro-8B. This advanced model, equipped with an impressive 8.3 billion parameters, is an expansion of the LLaMA2-7B, having undergone further training on extensive code and mathematical datasets, collectively amounting to 80 billion tokens.
LiteLlama-460M-1T was released, a reproduction of Meta AI”s LLaMa2 was released t open-source reproduction of Meta AI's LLaMa 2. However, with significantly reduced model sizes, LiteLlama-460M-1T has 460M parameters trained with 1T tokens
Crew AI is a new framework dedicated to building and deploying autonomous AI agents for various tasks. Similar to Microsoft’s AutoGen, various agents can be deployed that communicate together to accomplish a complex task.
The project of the week, worthy of recognition.
It has been an incredible start to the new year with breakthroughs in AI and new model releases. The Consumer Electronics Show (CES) takes the stage this week in Las Vegas where the focus will be specifically on robotics & AI. Expect a lot of news to come out in next week’s edition of Open Sourced Sunday.
If you have any feedback or suggestions, we’d love to hear from you on our Twitter.
Thanks for reading.