- NLPlanet Newsletter
- Posts
- Weekly AI and NLP News — April 3rd 2023
Weekly AI and NLP News — April 3rd 2023
The Twitter Algorithm, a Letter to Pause AI, and OpenAI History
Here are your weekly articles, guides, and news about NLP and AI chosen for you by NLPlanet!
😎 News From The Web
Twitter's Recommendation Algorithm. Twitter's algorithm uses ML to pick preferred tweets from user timelines in < 1.5 seconds. Includes candidate sourcing, ranking, and filtering stages.
The Only Way to Deal With the Threat From AI? Shut It Down. Open letter urges AI labs to halt training models more powerful than GPT-4 for 6 months. AI progress outpacing alignment and understanding raises concerns of approaching catastrophe.
The secret history of Elon Musk, Sam Altman, and OpenAI. History of OpenAI, transformation to for-profit, power struggles & controversies (incl. Musk & Altman), ChatGPT language model, Microsoft collab.
GM wants to bring ChatGPT-like assistant to drivers. GM is developing an AI assistant with Microsoft Azure & OpenAI's ChatGPT to aid drivers with advanced voice commands & useful guidance.
Jupyter X Hugging Face. Hugging Face enhances Jupyter notebook support on their machine learning platform for better reproducibility of models and datasets among the community.
Grammarly adding ChatGPT-like AI to create text in your writing style, outlines, and more. GrammarlyGO is a new AI tool being launched this April for users. It helps compose text, rewrite, ideate, personalize voice, and much more.
📚 Guides From The Web
labml.ai Annotated PyTorch Paper Implementations. Labml.ai has PyTorch paper implementations with explanatory notes, neural network & other related algorithms, updated regularly.
Open Stanford Course on Large Language Models. Stanford's CS324 focuses on large language models' fundamentals and practical applications. Students get hands-on exp with GPT-3/BERT models, ethical/scalability challenges covered.
Malleable software in the age of LLMs. The power of Large language models: LLMs (like GPT-4) mean anyone might soon author code & LLMs will revolutionize tool support for end-user programming.
Superhuman: What can AI do in 30 minutes? AI was used to create a marketing campaign in just 30 mins. It conducted market research, created a website and social media campaign, plus more. Its potential and disruption were discussed.
Not Boring, by Packy McCormick. OpenAI is becoming a significant tech position and business model, with millions of users on its new interface. It may become an aggregator and platform.
🔬 Interesting Papers and Repositories
Vicuna: An Open-Source Chatbot with Quality Similar to ChatGPT. Open-source chatbot Vicuna-13B rivals ChatGPT and Bard with 90% quality on 70K conversations using LLaMA fine-tuning. Costs ~$300.
A Comprehensive Survey of AI-Generated Content. This paper surveys Artificial Intelligence Generated Content (AIGC), its recent advancements in complex modeling and large datasets, and explores new ways to integrate technologies such as reinforcement learning.
LLaMA-Adapter: Efficient Fine-tuning of Language Models. LLaMA-Adapter is an efficient method to fine-tune models into instruction-following ones using learnable prompts. It generates high-quality responses and achieves great reasoning capacity with multi-modal inputs.
ChatGPT Outperforms Crowd-Workers for Text Annotation Tasks. ChatGPT beats humans in text annotation tasks like relevance, stance, topics, and frames detection, is more cost-effective, and has a higher intercoder agreement.
Meet in the Middle: A New Pre-training Paradigm. New language model pre-training method uses bidirectional sequence info for training and inference, achieving better results than strong baselines.
Thank you for reading! If you want to learn more about NLP, remember to follow NLPlanet. You can find us on LinkedIn, Twitter, Medium, and our Discord server!