Master LLMs with our FREE course in collaboration with Activeloop & Intel Disruptor Initiative. Join now!


This AI newsletter is all you need (#37)
Latest   Machine Learning

This AI newsletter is all you need (#37)

Last Updated on July 17, 2023 by Editorial Team

Author(s): Towards AI Editorial Team

Originally published on Towards AI.

What happened this week in AI by Louis

This week developments in LLMs and Generative AI continued at pace — ranging from more cutting-edge research, to open-source accessibility to general affordability of mass-market products.

Following Meta’s somewhat open-source release (to researchers only) of its 65BN parameter Llama model last week — the full model weights were promptly leaked on torrent websites. With the latest estimates at just $0.5–1m training cost, we also think it’s only a matter of time now that an open-source model with size and quality equivalent to GPT-3.5 is trained and released fully open source by a nonprofit or tech company. Increased flexibility for researchers and entrepreneurs to experiment with these models should accelerate progress — but also comes with questions on the risks of what could be built with these models if put to bad use.

OpenAI released its ChatGPT and Whisper APIs with significantly lower pricing than anticipated, partly in response to progress in open-source LLMs and competitor models. ChatGPT is available via GPT-Turbo for $0.002 per 1k tokens, which is 10x cheaper than its Da-Vinci model (which itself was reduced 3x in price in September) and in line with pricing for OpenAI’s smaller Curie model (Curie has been estimated at 7–10bn parameters vs Da-vinci’s 175bn). In our view, this pricing makes GPT-Turbo accessible and affordable in many applications and is only going to accelerate the rollout of Generative AI products and startups. It’s not clear how GPT-Turbo achieved its lower pricing relative to Da-Vinci — it could be trained as a smaller model, and it could benefit from cost-efficiency gains from methods such as pruning, sparsity, and model distillation. We are interested in hearing your thoughts!

The week concluded with Google announcing several new models, such as its Universal Speech Model, which advances speech AI for over 100 languages, and the Palm-E model, an “embodied” multimodal language model that merges PaLM-540B and ViT-22B. This model’s early success with positive transfer across multiple domains indicates we will be hearing more about embodied multimodal language models in the future.

Hottest News

1.Introducing ChatGPT and Whisper APIs

Open AI has recently made the ChatGPT and Whisper APIs available for developers to integrate the ChatGPT and Whisper models into their applications and products. The API features a range of system-wide optimizations that enable developers to use the open-source Whisper large-v2 model with quicker and more cost-effective outcomes.

2. Anthropic begins supplying its text-generating AI models to select startups

Anthropic, an AI startup co-founded by ex-OpenAI employees, has started providing partners with access to its AI text-generating models. The company believes its technology is distinct and can compete with rivals such as OpenAI, Cohere, and AI21 Labs, all of which offer paid access to their text-generating AI via APIs.

3. Microsoft brings its new AI-powered Bing to the Windows 11 taskbar

Microsoft is releasing a major update to Windows 11 today, which includes the company’s new AI-powered Bing search on the taskbar. Alongside a range of new features, such as improvements to widgets, a better touch mode, a screen recording feature, and tabs inside Notepad.

4. Why AI Won’t Cause Unemployment

There are various debates surrounding the impact of new AI technology on the replacement of human labor. This article puts forward the argument that AI cannot result in overall unemployment, as it is already illegal for most of the economy, and will be for virtually all of the economy.

5. YouTube’s new leader teases AI tools that can virtually swap creators’ outfits and locations

In his first public address since taking over as the head of YouTube, Neal Mohan outlined his key priorities and shared some upcoming features for the platform. He also teased one particularly interesting new one: YouTube is developing generative AI tools for content creators.

5-minute reads/videos to keep you learning

1.The Best of NLP: February 2023’s Top NLP Papers

This post curates a selection of the top NLP papers for February 2023, covering various topics, such as the latest developments in language models, text generation, and summarization.

2. What’s the big deal with Generative AI? Is it the future or the present?

The current developments in Generative AI and its future have captured the popular imagination. This article provides an overview of the current state of generative AI and discusses how we should approach its development in the future.

3. An overview of the latest open-source AI image and video editing neural network models on HuggingFace

The video shares an overview of the latest AI models for image and video editing, exploring their possibilities, capabilities, and more.

4. The technology behind GitHub’s new code search

This article provides insights into the development of the world’s largest public code search index, including the factors that contributed to its creation, the technologies utilized, and the process involved.

5. Must read: the 100 most cited AI papers in 2022

Who Is publishing the most Impactful AI research right now? This article analyzes the 100 most cited papers per year for 2020, 2021, and 2022, providing early insights into the direction of AI technology.

Papers & Repositories

1.Check Your Facts and Try Again: Improving Large Language Models with External Knowledge and Automated Feedback

The paper introduces the LLM-Augmenter system, which can augment a black-box LLM with a set of plug-and-play modules. The results show that LLM-Augmenter significantly reduces ChatGPT hallucinations without sacrificing the fluency and informativeness of its responses.

2. A Prompt Pattern Catalog to Enhance Prompt Engineering with ChatGPT

The paper introduces a catalog of prompt engineering techniques in the form of patterns, which have been applied to address common challenges in conversing with LLMs. It proposes a framework for documenting patterns that structure prompts, presents a catalog of patterns used to enhance LLM outputs, and more.

3. Directed Diffusion: Direct Control of Object Placement through Attention Guidance

This work proposes a simple solution to guide text-based diffusion models to focus on specific objects by injecting “activation” at desired positions in cross-attention maps corresponding to the objects of interest while attenuating the rest of the map. The approach is a significant step towards extending the applicability of text-guided diffusion models from single images to collections of related images.

4. WhisperX

A repository for Whisper-Based Automatic Speech Recognition (ASR). It refines the timestamps of OpenAI’s Whisper model via forced alignment with phoneme-based ASR models (e.g. wav2vec2.0), multilingual use-case.

5. ChatGPT: A Meta-Analysis after 2.5 Months

The paper analyzes over 300,000 tweets and more than 150 scientific papers to investigate how ChatGPT is perceived and discussed. The findings indicate ChatGPT is generally viewed as of high quality, however, this perception has slightly decreased since its debut.

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. Network Architecture Search (#9)

This week’s session in the (free) nine-part Neural Networks Architectures series led by Pablo Duboue (DrDub) focuses on Network Architecture Search. This is the last session before the seminar presentation! During this session, he will explore NEAT, Knowledge/Network Distillation, and GDAS. Find the link to the seminar here or add it to your calendar here.

Interested in presenting a neural network architecture at the TAI Discord Server? Let us know more by signing up for the seminar presentation here.

Date & Time: 7th March, 11 pm EST

Add our Google calendar to see all our free AI events!

Meme of the week!

Meme shared by jacek#5724

Featured Community post from the Discord

Misbah#5820 asked ChatGPT to write a self-help book titled “The Book of Infinite Wisdom” by condensing the knowledge and wisdom of thousands of books into one concise volume. The Book of Infinite Wisdom distills centuries of accumulated human knowledge, offering readers a comprehensive guide to achieving their fullest potential. You can find the book here or reach out to Misbah#5820 for a free copy (for select readers). Join the thread here and share your thoughts on a book entirely authored by AI.

AI poll of the week!

Join the discussion on Discord.

TAI Curated section

Article of the week

Clustering A-Z Briefly Explained ChatGPT Powered by Gencay I.

This article provides a comprehensive overview of Clustering A-Z, including its algorithms, boundaries, clusters, distance metrics, evaluation, feature selection, hierarchical clustering, k-means clustering, and quality metrics. As machine learning techniques can be broadly categorized into Regression, Classification, and Clustering, this article aims to cover the various aspects of Clustering.

Our must-read articles

The Most In-Demand Metaverse Skills for AI Professionals by Jennifer Wales

Precision, Recall, F-1 score- Must Know Before Your Next Project! by Anmol Tomar

If you are interested in publishing with Towards AI, check our guidelines and sign up. We will publish your work to our network if it meets our editorial policies and standards.

Job offers

Senior Software Engineer I, Analysis Pipeline @Freenome (Remote)

Senior Software Engineer @Tiliter (Sydney, Australia)

Clinical Data Analyst — R — #2915–6 Month Contract @GRAIL (Remote)

Senior Software Engineer (Open-Source) (f/m/d) @Jina AI (Berlin/Hybrid)

Machine Learning Engineer @Caption Health (Remote)

Senior Data Scientist/ML Engineer @Rivet (Remote)

Senior MLOps Engineer @Labelbox (Remote)

Interested in sharing a job opportunity here? Contact [email protected].

If you are preparing your next machine learning interview, don’t hesitate to check out our leading interview preparation website, confetti!

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI

Feedback ↓