LoLCATs: Faster Transformers with Less Cost
LoLCATs introduces a way to make Transformers like Llama 3 and Mistral much faster and cheaper to train. They’ve even made huge 70B and 405B models that work efficiently, saving costs similar to LoRA fine-tuning.
https://github.com/HazyResearch/lolcats/blob/main/lolcats_preprint_v0.pdf
https://github.com/HazyResearch/lolcats
Zamba2-7B Model: The New Best
ZyphraAI and NvidiaAI made the Zamba2-7B model, which is quicker and better than models like Mistral and Llama3. It’s now the best in the 8B class.
AlphaFold2: Game-Changer for Protein Research
AlphaFold2, from DeepMind, can predict the shape of proteins just by their amino acid sequence. This has helped researchers design enzymes, understand diseases, and discover new drugs.
OpenAI’s Multi-Agent Framework
OpenAI launched a tool to build systems with multiple AI agents that can talk to each other and delegate tasks. This framework is lightweight and easy to use.
LangChainAI’s Job Search Helper
LangChainAI has made an AI assistant to help with job searches, writing cover letters, and researching companies. It uses several agents working together to make the process smoother.
Claude 3.5 Sonnet Beats OpenAI’s o1 Model
Claude 3.5 Sonnet outperforms OpenAI’s o1 model in reasoning. Its success is credited to effective hint engineering, as noted by Jeremy Nguyen, PhD.
DiLoCo: Efficient Learning
DiLoCo is a new way to train large models across many computers, cutting down on bandwidth and making the process faster. It showed great results in training a 10B model globally.
LeanAgent for Theorem Proving
LeanAgent combines AI with proof tools to prove complex math theorems. It’s already proven 162 theorems that humans never solved before.
Yi-Lightning Models Released
Yi-01.AI launched its new models Yi-Lightning and Yi-Lightning-Lite, now available via API. These models are featured in a popular chatbot arena.
Adobe Firefly Video Model
Adobe revealed its Firefly Video Model at the AdobeMAX event, bringing advanced video editing capabilities to creators.
AI’s Impact on Jobs
Ethan Mollick discussed how AI is changing jobs and society. He stressed that people need to engage deeply with AI to fully understand its effects on their lives.
Ethereum’s Future and AI
Vitalik Buterin shared updates on Ethereum’s upgrades, which will boost its efficiency and scalability, essential for decentralized apps.
Full-Body Talking Video Generation
A new tool can create full-body talking videos from just audio, available on Hugging Face Space. This tech might soon expand to dance and sports videos.
HuggingFace’s Leaderboard
HuggingFace launched a leaderboard for AI research papers, showing which papers led to the most open-source projects.
AI-Generated Film at UP Summit
An AI-created film was showcased at the UP Summit, showing how AI is making its mark in filmmaking using tools like MidjourneyAI and Runway.
AI & Robotics Weekly Wrap
This wrap-up gives a quick overview of the biggest AI and robotics news of the week, featuring updates from Meta, Tesla, OpenAI, and others.
https://twitter.com/adcock_brett/status/1845738866491207740
CogView3 Diffusers Version
The latest version of ZhipuAI’s CogView3 is now available for testing. It shows how AI can drive impressive image generation.
Voice Cloning with Just 10 Seconds
Voice cloning tech is now so advanced that you can create a voice clone using just a 10-second audio clip, right on your computer.
Elon Musk’s Risk Philosophy
Elon Musk’s approach to risk-taking has helped SpaceX achieve success, especially with reusable rockets. His view is that anything is possible if it doesn’t break the laws of physics.
HuggingFace & Github Updates: AI Models, Tools, and Discussions
AI Model News: The Bllossom-3B model is a great tool for improving both Korean and English. It’s been trained with 150GB of Korean data without hurting its English skills. The model performs well on the LogicKor test and is good enough for businesses. It has also been showcased at big events like AAAI2024.
https://huggingface.co/Bllossom/llama-3.2-Korean-Bllossom-3B
Image Creation Technology:
Meissonic is a new tool that helps make images from text without using too much computer power. It works well even on normal graphics cards, so you don’t need expensive equipment. This makes high-quality image generation easier and more available to everyone.
https://huggingface.co/MeissonFlow/Meissonic
Open-Source Platforms:
Formbricks is a user-friendly, open-source survey tool. You don’t need to know how to code to use it. You can easily create surveys that work on websites, apps, and through email. It also connects to tools like Slack and Notion. You can either host it yourself or use the cloud version. While the basic version is free under an open-source license, some extra features require payment.
https://github.com/formbricks/formbricks
Reddit Insights:
A Reddit post recently discussed how most people don’t realize how much change is coming with AI. The post compared AI to past tech like cars and mobile phones, which changed everything quickly. Some users agree that AI will lead to big changes, while others think the hype might be overblown.
Here are some key points from the discussion:
Tech Impact on Society: New technologies often change the way society works. AI is expected to have the same effect.
Job Changes: Some believe AI will take over jobs, while others think it will help us work better.
AI Consciousness: There’s a debate on whether AI can ever become truly aware or conscious like humans.
https://news.miracleplus.com/share_link/44363
