This AI newsletter is all you need #32
Author(s): Towards AI Editorial Team Originally published on Towards AI. What happened this week in AI by Louis Following recent advancements in image, code, and text generation, there have been new developments in AI-generated music and text-to-speech. In 2019, everyone was already impressed with OpenAI’s Musenet, built on GPT-2 techniques applied to MIDI files. However, with recent advancements in AI, much more flexible and comprehensive music models are now possible. This week, AI has noted several interesting AI-generated music and text-to-speech models, including MusicLM, a model announced by Google Research that generates high-fidelity music from rich text descriptions. Although the model isn’t released yet, dataset MusicCaps, consisting of 5.5k human-written music-text pairs, is available. The paper “Make-An-Audio” was also released this week, describing Text-To-Audio Generation with Prompt-Enhanced Diffusion Models. We expect to see a wave of AI music startups and open-source models released going forward, especially as annotated music data sets become accessible. We hope progress in AI music models can benefit musicians exploring new concepts and lower the cost and obstacles for new musicians entering the industry. Besides new music models this week, we also discovered an impressive and flexible text-to-audio model from elevenlabs. Excited about the potential of such models to increase accessibility of written content, we also see growing risks in text-to-audio, including voice cloning for fake quotes and voice-protected logins and verifications. Hottest News 1.ChatGPT is ‘not particularly innovative,’ and ‘nothing revolutionary’, says Meta’s chief AI scientist Recently, there has been much discussion about the potential of OpenAI’s ChatGPT program for generating natural language responses to human prompts. However, AI scholars have a different view. During a Zoom meeting with press and executives last week, Yann LeCun, the Chief AI Scientist at Meta, stated, “In terms of underlying techniques, ChatGPT is not particularly innovative.” 2. The inside story of ChatGPT: How OpenAI founder Sam Altman built the world’s hottest technology with billions from Microsoft Sam Altman believes that the future of AI could be exceptional — unless things go astray. It is important to know the story of OpenAI’s ChatGPT chatbot, which has been used for activities such as debugging code, writing recipes, scripts, and more, and how it has sparked a revolution. 3. AI adoption: is it obvious yet? Regardless of one’s opinion of ChatGPT, its release last year generated another buzz in the AI community. Its launch likely did more to promote the value of machine learning to non-experts than any other event. It is crucial to assess if we are ready for AI adoption in terms of technological and product maturity and to comprehend the excitement surrounding AI and the arguments for and against incorporating it into products. 4. AI21 Labs has created a co-writing bot that can suggest quotes, statistics, provide citations and more AI21 Labs, a research lab specializing in NLP and generative AI, announced the launch of Wordtune Spices, a new feature for its popular Wordtune editing platform, to enhance the writing experience for writers of all types. Wordtune Spices is an AI-powered writing tool that comprehends content and meaning to assist users in expressing their ideas more effectively and compellingly. 5. The Human-AI Partnership In this podcast, Reid Hoffman speaks with ChatGPT about the partnership between humans and AI. He delves into the unique ways in which AI can enhance human capabilities as part of a miniseries focused on the future of AI and chatbots. Three 5-minute reads/videos to keep you learning 1.A Brief History of Artificial Intelligence This article explores the rich history and evolution of AI, from its early origins to the current ethical debates surrounding its development. The article traces the journey of AI, starting with its first concepts and leading to the seminal conference organized by Allen Newell, Cliff Shaw, and Herbert Simon, which marked the beginning of its proof of concept. The article provides insight into the past, present, and future of AI. 2. Introduction to embeddings — The bread and butter of language models Word and sentence embeddings are the backbones of language models. This Twitter thread by Cohere offers a clear and simple introduction to these essential concepts, including examples, applications, and additional resources. It also explains how embeddings work and how they can be used in language models. 3. Getting started with LLMs using LangChain Recently, there has been a surge of interest in generative AI and language models (LLMs). This article introduces LangChain, a library that enables the creation of advanced applications around LLMs such as OpenAI’s GPT-3 models and the open-source alternatives provided by Hugging Face. It begins with a discussion on the simplest component offered by LangChain. 4. Five pieces of advice for those building in AI right now In this Twitter thread, Nathan shares his thoughts on the process of building products. Some advice he shares in his thread: avoiding generalizations, recognizing that AI is not a unique advantage, treating AI-powered products as more than just “wrappers”, disregarding hype, and acknowledging that AI-powered applications are not primarily focused on AI. 5. Manipulating Tensors in PyTorch PyTorch is a deep-learning library that operates on numerical arrays known as tensors. This article provides a brief overview of what PyTorch offers for tensors and how to use them. It gives insight into how to create and perform operations on PyTorch tensors and the common functions available in PyTorch for manipulating tensors. Enjoy these papers and news summaries? Get a daily recap in your inbox! The Learn AI Together Community section! Upcoming Community Events The Learn AI Together Discord community hosts weekly AI seminars to help the community learn from industry experts, ask questions, and get a deeper insight into the latest research in AI. Join us for free, interactive video sessions hosted live on Discord weekly by attending our upcoming events. 1.Convolution Networks: The Neural Network Architecture Seminar (#4) This is the fourth session of a (free) nine-part series on Neural Networks Architectures presented by Pablo Duboue (DrDub), covering Convolution Networks. The session focus on CNNs, DL image processing, YOLO, U-Net, Retina-Net, and SpineNet. Find the link to the seminar here or add it to your […]