📚 It may be exam season, but not to worry, WatchTower 🏰 has you covered with all the thrilling developments in AI! 🤖

This week's letter 💌 takes you into the world of AI-powered learning with LLaMA 2, Meta's latest Large Language Model trained on 2 trillion tokens! 📚🧠

Further, eager to explore new horizons? 🚀 "Neural Text-to-Speech Synthesis" 🗣️ is your guide.

And finally, for all the music enthusiasts 🎵, Meta's AudioCraft lets you compose like a pro with AI-generated music and open-sourced models—unleash creativity! 🎹🎼

Best of luck in your exams, you've got this! 💪🍀

Weekly Dose of AI

Well, have you ever wanted to infuse the essence of a book into your responses? Or maybe even a person? Or maybe both ;) 

Lets start with a completely hypothetical situation, where you’re signing up to a hypothetical society called the Intelligently Artificial Society and they ask you this question, but you don’t know what to do!

Why do you want to join IA Society as a subcommitee member?

Write a paragraph answering this question. IA Society is a student-run society at UNSW. Fabricate wherever necessary. Use principles from Robert Cialdini's book Influence and 48 Laws of Power to be as persuasive as possible.

In my experience, it didn’t get it right the first time so follow up with “Now write it, but remove all mention of both books.“

But this is purely hypothetical. We know if you use AI😁😁😁😁 

Have fun!

🌐 Fine-tuning Your Own Chatbot

Meta, the tech giant, has just unveiled LLaMA 2, their latest Large Language Model, and it's a game-changer! Trained on a massive 2 trillion tokens, these cutting-edge models excel in reasoning, coding, proficiency, and knowledge tests. Plus, they offer three options - 7B, 13B, and a jaw-dropping 70B parameter size, freely available for research and commercial use.

But here's the best part - LLaMA 2 allows you to unleash your creativity! Thanks to QLoRA, a groundbreaking technique, fine-tuning these models with your own text becomes a breeze. Want a personalized AI study buddy? You got it! Train LLaMA 2 to be your private GPT and create your smart learning notebook!

Now, how can this affect you? Picture this: Imagine a chatbot that knows exactly what you need to study, reminds you of important dates, and helps you revise like a pro. LLaMA 2 can be your academic sidekick, helping you ace exams and stay organized!

So, don't miss out! Request access to LLaMA 2 and dive into the world of AI-powered learning. Embrace the future of technology and create your very own smart learning companion. Let's level up together! 🚀

_{Published by Li Li, August 7 2023}

Unveiling "Neural Text-to-Speech Synthesis"  

We're thrilled to introduce a remarkable leap in understanding speech synthesis – the debut of "Neural Text-to-Speech Synthesis." Authored by Xu Tan, a prominent scholar in the field, this book illuminates the future of human-computer auditory interaction.

Journey through over six decades of tireless research, tracing speech synthesis from its roots to the current realm of modern neural networks. From foundational insights into spoken language processing to the intricacies of deep learning in text-to-speech synthesis, the narrative advances to dissect essential building blocks: text analysis, acoustic models, vocoders, as well as fully end-to-end models.

But this book isn't limited to the technical realm. It ventures into advanced territories, exploring the interplay between data efficiency, expressiveness, and their fusion with the immersive metaverse. It also glimpses the horizon, where speech synthesis reshapes auditory engagement.

That's not all! This book assumes zero prior knowledge in the field or AI. Starting from scratch, it gradually builds up, making it a self-contained read. By the end, you'll be up to speed with one of deep learning's trendiest topics. Whether you're a newbie or a seasoned pro, this book has something for you. It's not just about tech – it's your guide to mastering neural text-to-speech synthesis.
_{Published by Solomon Na Muangtoun, August 7 2023}

Harmonizing Innovation: AudioCraft, Meta's Symphony of AI-Generated Music and Sound

Hey there, musical maestro! 🎶 Ever tapped your fingers on a desk, dreaming up a symphony? Well, get ready to compose like a pro with Meta's AudioCraft! As someone who has never touched an instrument in his life, I'm all ears for this innovation, and I bet you will be too! 🎧 🤣 

Introducing AudioCraft, a generative AI tool that's like a virtual orchestra at your fingertips. Comprising three models: MusicGen, AudioGen, and EnCodec, it turns your text prompts into melodies and effects with a quality that would make Beethoven nod in approval. 🎵

But wait, there's more! Meta's sharing the love by open-sourcing these models. Imagine the possibilities as researchers and creators like you (yes, you!) train and tweak these models. It's like a global jam session, and you're invited! 🎷

Creating high-fidelity audio isn't a walk in the park; it's more like a dance with intricate signals and patterns. But AudioCraft makes it look easy, offering a simplified design that works for music, sound, compression, and generation—all in one codebase.

_{Published by Andrew Suryanto, August 7 2023}

Closing Notes

In the exam season, we hope that this week's AI-powered newsletter has sparked some inspiration for you all🥳 . Best of luck to everyone during this time!🎉 May your hard work and dedication lead to successful results. Keep pushing forward and embrace the power of technology as you continue to excel in your studies🤗 . You've got this!💪

_{Published by Li Li, August 7 2023}

🥫Sauces 🥫

Llama 2: Open Foundation and Fine-Tuned Chat Models

Neural Text To Speech Analysis

Meta Introduces AudioCraft

Meta Launches AudioCraft, an open source music generator