TWIL: March 19, 2023

What a week for AI! Open AI released GPT-4, Microsoft announced Microsoft 365 Copilot, Anthropic launched Claude, Google offers access to PaLM and Baidu unveils Ernie. Also, two awesome podcast episodes from .NET Rocks, new features in Form Recognizer and how to use semantic search in Azure Cognitive Search. Finally, a set of research articles on large language models. Enjoy!


Podcast

.NET Rocks

Episode 1834: Modern Web Front End Development with Amy Kapernick
What does web front-end development look like in 2023? Carl and Richard chat with Amy Kapernick about her work helping companies build web front ends with a vast array of tools. Amy talks about how client frameworks continue to evolve, extending the so-called “big three” of Angular, Vue, and React to focus on different styles. The conversation also ranges over testing web apps, building pipelines for automated testing, accessibility, and more!

Episode 1835: The Next C# with Mads Torgersen
What’s next for C#? Carl and Richard talk to Mads Torgersen about what the team is working on for C# 12. Mads talks about how the language design team is organized to take ideas for C# and explore them, considering all aspects before implementation. The conversation digs into a few of the new features coming and some of the considerations, like breaking changes, that might be necessary to make a feature as good as possible. With C# nearly 25 years old, there is lots of legacy to deal with, but the future looks bright!


Artificial Intelligence

Chain-of-Thought Prompting Elicits Reasoning in Large Language Models
We explore how generating a chain of thought — a series of intermediate reasoning steps — significantly improves the ability of large language models to perform complex reasoning. In particular, we show how such reasoning abilities emerge naturally in sufficiently large language models via a simple method called chain of thought prompting, where a few chain of thought demonstrations are provided as exemplars in prompting. Experiments on three large language models show that chain of thought prompting improves performance on a range of arithmetic, commonsense, and symbolic reasoning tasks. The empirical gains can be striking. For instance, prompting a 540B-parameter language model with just eight chain of thought exemplars achieves state of the art accuracy on the GSM8K benchmark of math word problems, surpassing even finetuned GPT-3 with a verifier.

ReAct: Synergizing Reasoning and Acting in Language Models
While large language models (LLMs) have demonstrated impressive capabilities across tasks in language understanding and interactive decision making, their abilities for reasoning (e.g. chain-of-thought prompting) and acting (e.g. action plan generation) have primarily been studied as separate topics. In this paper, we explore the use of LLMs to generate both reasoning traces and task-specific actions in an interleaved manner, allowing for greater synergy between the two: reasoning traces help the model induce, track, and update action plans as well as handle exceptions, while actions allow it to interface with external sources, such as knowledge bases or environments, to gather additional information. We apply our approach, named ReAct, to a diverse set of language and decision making tasks and demonstrate its effectiveness over state-of-the-art baselines, as well as improved human interpretability and trustworthiness over methods without reasoning or acting components.

Toolformer: Language Models Can Teach Themselves to Use Tools
Language models (LMs) exhibit remarkable abilities to solve new tasks from just a few examples or textual instructions, especially at scale. They also, paradoxically, struggle with basic functionality, such as arithmetic or factual lookup, where much simpler and smaller models excel. In this paper, we show that LMs can teach themselves to use external tools via simple APIs and achieve the best of both worlds. We introduce Toolformer, a model trained to decide which APIs to call, when to call them, what arguments to pass, and how to best incorporate the results into future token prediction. This is done in a self-supervised way, requiring nothing more than a handful of demonstrations for each API. We incorporate a range of tools, including a calculator, a Q\&A system, two different search engines, a translation system, and a calendar. Toolformer achieves substantially improved zero-shot performance across a variety of downstream tasks, often competitive with much larger models, without sacrificing its core language modeling abilities.


Open AI

GPT-4 is OpenAI’s most advanced system, producing safer and more useful responses
GPT-4 can solve difficult problems with greater accuracy, thanks to its broader general knowledge and problem solving abilities. GPT-4 is more creative and collaborative than ever before. It can generate, edit, and iterate with users on creative and technical writing tasks, such as composing songs, writing screenplays, or learning a user’s writing style. GPT-4 can accept images as inputs and generate captions, classifications, and analyses. GPT-4 is capable of handling over 25,000 words of text, allowing for use cases like long form content creation, extended conversations, and document search and analysis.

GPT-4
We’ve created GPT-4, the latest milestone in OpenAI’s effort in scaling up deep learning. GPT-4 is a large multimodal model (accepting image and text inputs, emitting text outputs) that, while less capable than humans in many real-world scenarios, exhibits human-level performance on various professional and academic benchmarks.

Everything You Need To Know About GPT-4
Today we will explore GPT 4 and its brand-new capabilities.

OpenAI vs Azure OpenAI
Recently Microsoft released Azure OpenAI services, which is a new Azure Cognitive Service that allows customers to access OpenAI’s language models like GPT-3, Codex, and Embeddings model series through REST APIs from Microsoft Azure. Furthermore, Azure OpenAI Service has enterprise-grade features such as security, compliance, and regional availability that are only available on Azure compared to the service from OpenAI.

Enterprise Azure OpenAI
Repository detailing the deployment of an Enterprise Azure OpenAI reference architecture.

Chat GPT and Large Language Models: The Future of Natural Language Processing
In recent years, natural language processing (NLP) has made significant progress towards enabling computers to understand and generate human language. One of the most exciting developments in this field is the rise of large language models, such as Chat GPT, which are based on deep learning and artificial intelligence (AI). These models have the potential to revolutionize the way we interact with technology, and the impact they will have on businesses and society at large is immense.


Azure Applied AI Services

Form Recognizer previews document classification, Azure OpenAI integration and lots more
Form Recognizer is an Applied AI service for all your document understanding needs. With the latest update Form Recognizer now adds new capabilities like document classification, new prebuilt models like the 1098 form (with a few variants) and using Azure OpenAI models to extend field extraction with queries.

What’s new in Azure Form Recognizer
Form Recognizer service is updated on an ongoing basis. Bookmark this page to stay up to date with release notes, feature enhancements, and our newest documentation.

Get started with Form Recognizer Studio
Extract text, key-value pairs, tables, and structures from forms and documents using common layouts and prebuilt models, or create your own custom models.


Azure Cognitive Search

Semantic search in Azure Cognitive Search
Semantic search is a collection of query-related capabilities that bring semantic relevance and language understanding to search results. This article is a high-level introduction to semantic search. The embedded video describes the technology, and the section at the end covers availability and pricing.

Configure semantic ranking and return captions in search results
In this article, you’ll learn how to invoke a semantic ranking algorithm over a result set, promoting the most semantically relevant results to the top of the stack. You can also get semantic captions, with highlights over the most relevant terms and phrases, and semantic answers.


Other AI News

Introducing Microsoft 365 Copilot—A whole new way to work
Today, we announced Microsoft 365 Copilot—your copilot for work. Copilot combines the power of large language models (LLMs) with your data in the Microsoft Graph—your calendar, emails, chats, documents, meetings, and more—and the Microsoft 365 apps to turn your words into the most powerful productivity tool on the planet. And it does so within our existing commitments to data security and privacy in the enterprise.

Anthropic launches Claude, a chatbot to rival OpenAI’s ChatGPT
Anthropic, a startup co-founded by ex-OpenAI employees, today launched something of a rival to the viral sensation ChatGPT. Called Claude, Anthropic’s AI — a chatbot — can be instructed to perform a range of tasks, including searching across documents, summarizing, writing and coding, and answering questions about particular topics. In these ways, it’s similar to OpenAI’s ChatGPT. But Anthropic makes the case that Claude is “much less likely to produce harmful outputs,” “easier to converse with” and “more steerable.”

Google opens up its AI language model PaLM to challenge OpenAI and GPT-3
Google is offering developers access to one of its most advanced AI language models: PaLM. The search giant is launching an API for PaLM alongside a number of AI enterprise tools it says will help businesses “generate text, images, code, videos, audio, and more from simple natural language prompts.”

China’s Baidu unveils ChatGPT rival Ernie
Chinese search engine giant Baidu has revealed its artificial intelligence-powered chatbot Ernie, the latest rival to OpenAI’s groundbreaking ChatGPT. Unlike OpenAI’s demonstrations of ChatGPT, Baidu did not demonstrate Ernie’s capabilities live but instead through a series of slides. The chatbot also lacks functions unveiled in the follow-up to Chat GPT, GPT-4, such as the ability to generate text in response to an image.


Have an awesome week!