1000 subscribers!!! 🎉, Google released Gemini, the biggest rival for GPT-4 🌌, Sam Altman, Times CEO of the Year 🏆, Purple Llama, a safer AI by Meta 💜, and new Linkedist e-book about AI TOOLS 📚
AI Connections #43 - a weekly newsletter about interesting blog posts, articles, videos, and podcast episodes about AI
I'm over the moon to share that AI Connections has hit 1000 subscribers! 🎉 Thank you all for your incredible support and enthusiasm; it means the world to me. 🌟 Let's continue this amazing journey into the world of AI! 🤖
JOIN 🪐
Cohere for AI: Aya - An Open Science Initiative to Accelerate Multilingual AI Progress: HERE
Contribute 500 submissions in any language and get Cohere for AI swag 🚀
How to do it:
1) Go to: https://aya.for.ai/
2) Push the "Sign In/Sign Up" button, and use Discord or Gmail to log in.
3) Push on your picture and go to settings
4) Choose the language in which you want to contribute
5) On the contribution page, "Rate Plan Performance."
6) Check if prompt and completion are in your selected language.
HUGE THANKS FOR OUR PAID SUBSCRIBERS:
edita.perku ⭐️
ausra.pock ⭐️
donatas.jasas ⭐️
rima.viniarskaite ⭐
RECOMMENDATION ⭐🥳
LINKEDIST released a FREE E-BOOK about AI TOOLS for MARKETING!!!
The ebook has 21 tools for copywriting, visual creation, website curation, and project management, as well as a guide to creating effective prompts and a sneak peek at the future of AI in marketing; this ebook is something not to miss.
GOOGLE RELEASED GEMINI 🤩🧠🦾
Gemini is Google’s latest large language model, which was first teased at the I/O developer conference in June and is now launching to the public.
Gemini comes in three sizes: Ultra, Pro, and Nano (Very similar to Apple products) for different purposes, from datacenters to mobile devices.
Gemini it's multimodal; it can understand and operate across text, images, audio, video, and code - and achieves state-of-the-art performance across many tasks.
Bard will now use Gemini Pro, making it better at understanding and summarizing, reasoning, brainstorming, writing, and planning
From December 13, developers can access Gemini Pro via the Gemini API at Google AI Studio, a free web-based developer AI tool.
The model will also be available in more Google products and services like Search, Ads, Chrome, and Duet AI.
It has the most comprehensive safety evaluations of any Google AI model to date, including for bias, toxicity, and other potential risks.
Technical Report: READ
Blog Post: READ
READ 📚
“CEO of the Year 2023: Sam Altman” article by Time magazine: READ
“Purple Llama: Towards open trust and safety in the new world of generative AI” blog post by Meta AI: READ This blog post is about the launch of Purple Llama, an initiative to foster responsible development and usage of open generative AI models through community collaboration, providing tools and evaluations for cybersecurity and input/output safeguards, and emphasizing the importance of safety and trust in the rapidly evolving field of AI innovation.
“AI Alliance Launches as an International Community of Leading Technology Developers, Researchers, and Adopters Collaborating Together to Advance Open, Safe, Responsible AI” blog post by AI Alliance: READ
“Early impressions of Google’s Gemini aren’t great” article by TechCrunch: READ This article is about the challenges and shortcomings of Google's new generative AI model, Gemini Pro, highlighting its inaccuracies in basic facts, struggles with translation, outdated information, and vulnerability to "jailbreak" prompts, despite Google's claims of its advanced capabilities compared to other AI models like GPT-4.
“Anthropic’s latest tactic to stop racist AI: Asking it ‘really really really really’ nicely” article by TechCrunch: READ This article is about a study by Anthropic researchers demonstrating that appending pleas to AI prompts can significantly reduce biases in decision-making related to race and gender, but it also emphasizes that such language models, including their Claude 2.0, are not suitable for high-stakes decisions like finance and health, highlighting the ongoing challenge of aligning AI models with ethical and legal standards.
“Long context prompting for Claude 2.1” blog post by Anthropic: READ This blog post is about Claude 2.1, an AI model with a 200K token context window, which shows excellent recall ability for well-integrated sentences but hesitates with out-of-place sentences, a limitation that can be overcome with minor prompt modifications.
“Introducing PPLX Online LLMs” blog post by Perplexity: READ This blog post is about the introduction of Perplexity's new PPLX models, pplx-7b-online and pplx-70b-online, which address limitations of current LLMs by providing helpful, factual, and up-to-date information through online access to internet knowledge, enhanced by in-house search technology and fine-tuning, outperforming other models in evaluations focused on helpfulness, factuality, and freshness.
“StableLM Zephyr 3B: A New Addition to StableLM, Bringing Powerful LLM Assistants to Edge Devices” blog post by Stable Diffusion: READ This blog post is about the release of StableLM Zephyr 3B, a new lightweight LLM with 3 billion parameters, designed for efficient text generation in Q&A and instruction-following tasks, building upon the StableLM 3B-4e1t model and inspired by HuggingFace's Zephyr 7B model, suitable for edge devices.
2023 AI Research from Lithuanian researchers 🚀🦾🧠
As 2023 nears its end, I want to take a moment to acknowledge and express gratitude to the extraordinary AI researchers from Lithuania for their significant achievements this year. Their work is a testament to the exceptional AI talent from Lithuania.
READ (RESEARCH PAPERS) 📚
“AlphaCode 2 Technical Report” research paper by Google DeepMind: READ This research paper is about AlphaCode 2, an advanced AI system enhanced by Gemini, which significantly outperforms the original AlphaCode by solving 1.7× more problems and exceeding the capabilities of 85% of human competitors in competitive programming tasks involving advanced mathematics, logic, and computer science.
“Nash Learning from Human Feedback” research paper by Google DeepMind: READ This research paper is about introducing Nash learning from human feedback (NLHF), an alternative approach to fine-tuning large language models (LLMs) using human feedback, which involves learning a preference model and optimizing policies to generate responses that align with human preferences, aiming to reach a Nash equilibrium, demonstrated through the innovative Nash-MD algorithm and experiments in text summarization tasks.
“Elo Uncovered: Robustness and Best Practices in Language Model Evaluation” research paper by Cohere for AI: READ This research paper is about the analysis of the Elo rating system, originally used for ranking chess players, for evaluating Large Language Models (LLMs) in NLP, questioning its reliability and transitivity when applied to entities with constant skill levels like LLMs, and providing guidelines to enhance the evaluation methods based on extensive evaluations and the study of system's axioms.
“Monarch Mixer: A Simple Sub-Quadratic GEMM-Based Architecture” research paper by Stanford University: READ This research paper is about the development of Monarch Mixer (M2), a new architecture that addresses the quadratic scaling issue of Transformers by using Monarch matrices for sub-quadratic scaling in both sequence length and model dimension, demonstrating its efficiency and effectiveness in various modeling tasks including BERT-style language modeling, ViT-style image classification, and GPT-style language modeling, with fewer parameters and higher throughput.
“Evaluating and Mitigating Discrimination in Language Model Decisions” research paper by Anthropic: READ This research paper is about a method for evaluating the potential discriminatory impact of language models in high-stakes societal decisions, using generated prompts across diverse scenarios to reveal patterns of discrimination in the Claude 2.0 model, and demonstrating how prompt engineering can mitigate such biases, aiding developers and policymakers in safer language model deployment.
WATCH🎥
AI TOOLS 💎
“AI Upscale tool” by Magnific AI: TRY
MagicAnimate: Make your photo dance: TRY
COOL THINGS 😎
StreamDiffusion - super fast image generation! The age of synthetic data is starting 🎭
Motion capture powered by AI by Move AI 🕺
New AI Video by Martin Haerlin using Pika 🚀