First OpenSource 7B Model to Challenge ChatGPT 💥
PLUS: LLMs with 4x Context Size of GPT-4, Text-to-3D by Luma, Generative AI for Developers and Analysts
Today’s top AI Highlights:
OpenChat: Open-source Language Models with Mixed-Quality Data
YaRN: Efficient Context Window Extension of LLMs
Fast, Easy and Secure LLM App Development With Snowflake
LinkedIn Debuts AI Features for Enhanced Networking
Luma AI Launches Text-to-3D Model
& so much more!
Read time: 3 mins
Latest Developments 🌍
OpenChat Dominates Much Larger LMs 💪
OpenChat is a library of open-source LLMs fine-tuned with innovative strategies that outperforms even the larger 70B models and gives a tough competition to ChatGPT. It leverages mixed-quality data and a novel strategy, C-RLFT, to fine-tune language models, avoiding the need for costly human preference labels.
Key Highlights:
C-RLFT allows for simple and RL-free training, making the fine-tuning process less complex and more stable compared to typical RLHF. It also has low requirements for the quality of the reward and does not necessitate expensive human feedback collection.
OpenChat-13B outperforms other popular LLMs on various benchmarks, including Alpaca-Eval, MT-bench, and Vicuna-bench, demonstrating its superiority over larger models like GPT-4 and llama-2-chat-70B.
The AGIEval accuracy of OpenChat-13B highlights its measure of generalization performance, indicating that the model performs well without overfitting.
Mistral and Llama 2 with 4x Context Size of GPT-4
Introducing new variants to Mistral and Llama 2 models with 128k context window, 4x the size of GPT-4. Utilizing the novel YaRN (Yet another RoPE extensioN method), both models have been fine-tuned, extending their context windows to an impressive 64k and 128k tokens.
Key Highlights:
YaRN is a compute-efficient method approach allowing for significant context window extension with 10x less tokens and 2.5x fewer training steps compared to previous methods.
Overcoming the limitations of positional encodings in transformer-based models, YaRN demonstrates the ability to effectively utilize and extrapolate to context lengths much longer than the original pre-training would allow.
YaRN's successful performance in various evaluation metrics, including the handling of long sequence language modeling, passkey retrieval tasks, and standardized benchmark comparisons.
Generative AI for Developers and Analysts 📊
Snowflake announces Snowflake Cortex and Snowpark Container Services, bringing the power of generative AI to developers, seamlessly integrating into their analytical processes while ensuring unparalleled security and governance.
Key Highlights:
Snowflake Cortex offers access to industry-leading AI models and LLMs, enabling organizations to analyze data and build AI applications swiftly. This includes serverless functions for inference on LLMs such as Llama 2 and task-specific models, along with advanced vector search functionality.
Snowpark Container Services facilitates deployment, management, and scaling of custom containerized workloads and models within the secure Snowflake-managed infrastructure, including support for GPU instances.
Snowflake's initiatives allow any user to leverage cutting-edge LLMs through UI-based experiences, Snowflake Cortex Functions, and Streamlit in Snowflake, without the need for custom integrations or front-end development expertise.
LinkedIn Adopts AI to Keep You Engaged and Informed 💡
As LinkedIn will hit 1 billion users this month, it is unveiling AI-powered reading and writing tools. The AI tools will initially be available for Premium users and can be used to summarise content, write smart responses, and enhance the job-hunting experience.
LinkedIn is using OpenAI APIs from Azure, taping GPT-4, combined with its proprietary data to generate personalized AI outputs. These AI tools aim to enhance social media engagement on the platform and keep you more active, especially in a professional network like LinkedIn.
Tools of the Trade ⚒️
Genie by Luma AI: Text-to-3D generative model designed for quick and easy creation and customization of 3D objects and prototypes, currently in research preview.
YouTune: Fine-tune SDXL on images from YouTube videos. It downloads the video, takes screenshots of every 50 frames, remove near duplicates and very light/dark images, and creates a training for you.
Chatd: A desktop app that lets you chat with your documents locally using AI.
Knime: A data science platform that enables comprehensive data analysis, skill development, and scalable deployment of data solutions without the need for coding.
Yack: A fast, lightweight MacOS app for accessing ChatGPT from your menu bar, along with multiple themes, keyboard-first design, cross-app integration and prompt templates.
😍 Enjoying so far, TWEET NOW to share with your friends!
Hot Takes 🔥
I’m calling it. Creative software is dead. It’s the end of an era.
Creative software 1.0 was about separating specific tasks into domains. Vector graphics, NLE, motion graphics, image editing, 3D, audio editing, compositing, etc. are highly specialized fields. ~ Cristóbal ValenzuelaI suspect that Andrew Ng and Yann LeCun have missed the main reason why the big companies want regulations. Years ago the founder of a self-driving company told me that he liked safety regulations because if you satisfied them it reduced your legal liability for accidents. ~ Geoffrey Hinton
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!