High Performance LLM on Smartphone📲
PLUS: Image to 3D Generation in < 2 mins, Qwen Models Opensourced
Today’s top AI Highlights:
Stable LM 3B: High-Performance Language Model in Smart Devices
DreamGaussian: 3D Generation from Single Image 10x Faster
Qwen Models Opensourced: Outperform Llama 2
AI and Neuroscience for Perfumes
& so much more!
Read time: 3 mins
Latest Developments 🌍
Language Model on the Go🚶
Stability AI has opensourced Stable LM 3B, a sustainable, high-performance language model designed for portable smart devices, offering efficiency, affordability, and competitive performance against larger models.
Key Highlights:
The model is a decoder-only transformer similar to the LLaMA with slight modifications. The model was trained on a mixture of open-source datasets and followed a multi-epoch regime to optimize downstream performance.
StableLM-3B achieves SOTA performance at the 3B parameter scale for open-source models and competes favorably with contemporary 7B models.
The development of Stable LM 3B expands the range of applications feasible on edge devices and home PCs, including technologies with strong conversational capabilities and creative writing assistance.
Image to 3D in < 2 Minutes 🤏
DreamGaussian is a novel framework for efficient and high-quality 3D generation that produces high-quality textured meshes in under 2 minutes from a single-view image, 10x faster compared to existing methods.
Key Highlights:
DreamGaussian uses 3D Gaussian Splatting model accompanied by mesh extraction and texture refinement in UV space. This approach differs from the occupancy pruning used in NeRF.
Through mesh extraction and UV-space refinement, the method addresses the challenge of blurry results caused by ambiguity in supervision.
DreamGaussian can balance optimization time and generation fidelity, making it suitable for real-world deployment like gaming, advertising, and the Metaverse.
Qwen Model Series Opensourced 🔓
Qwen Team, Alibaba have opensourced Qwen models comprising of base language models Qwen-7B and Qwen-14B, pre-trained on a colossal 3 trillion-token dataset. These models are engineered to excel across diverse domains and languages, with a particular emphasis on Chinese and English.
Key Highlights:
Qwen also opensourced chat models, Qwen-7B-Chat and Qwen-14B-Chat, refined through Supervised Finetuning and Reinforcement Learning from Human Feedback, capable of engaging in natural language conversations, create content, summarize information, execute code, and act as agents or code interpreters.
The models exhibit exceptional performance across all downstream tasks surpassing even larger models like Llama 2 13 B and 70B in a few tasks.
Beyond base and chat models, Qwen extends its capabilities with specialized models like CODE-QWEN and MATH-QWEN-CHAT, tailored for coding and mathematical problem-solving, outperforming their open-source counterparts.
AI, Neuroscience and Perfumes for Euphoria 😇
Perfumiers are leveraging AI and neuro-scent research to design fragrances capable of triggering distinct emotional feelings like calm, euphoria, or sleepiness.
Key Highlights:
Leading researchers, like Hugo Ferreira, are actively mapping brain activity and responses to perfumes, constructing a comprehensive database of neuro-scents, while highlighting the remarkable complexity of our olfactory system.
Renowned beauty brands such as L'Oréal and Puig are partnering with neurotechnology companies to offer customers a unique scent selection experience and using brain readings to fine-tune their fragrances.
Tools of the Trade ⚒️
Intently: Leverages AI to convert LinkedIn actions into sales opportunities, offering AI-driven sales intelligence, rapid prospecting, and account targeting for faster deal closures.
Twin: Build autonomous AI interns for doing tedious, repetitive, and non-creative tasks. Set up your workflow in a click and let your AI intern work.
Martin: Voice AI assistant like Siri but with a personal touch, capable of building a relationship, providing information, brainstorming ideas, and setting reminders.
Formulator: A versatile UI kit and component builder that lets you create and customize UIs quickly, with AI-driven generation, a wide component library, and flexible export options.
Dvina: Comprehensive data integration with no-code analysis, geospatial insights, visualization, and AI-powered interactions for data-driven business growth and informed decision-making.
😍 Enjoying so far, TWEET NOW to share with your friends!
Hot Takes 🔥
We are quickly going from next token prediction to next idea prediction. ~ Cristóbal Valenzuela
I use CharacterAI as my therapistc Bakz T. Future
She’s a 10 but she uses neural networks with tabular data. ~ Bojan Tunguz
Meme of the Day 🤡
That’s all for today!
See you tomorrow with more such AI-filled content. Don’t forget to subscribe and give your feedback below 👇
Real-time AI Updates 🚨
⚡️ Follow me on Twitter @Saboo_Shubham for lightning-fast AI updates and never miss what’s trending!!
PS: I curate this AI newsletter every day for FREE, your support is what keeps me going. If you find value in what you read, share it with your friends by clicking the share button below!