In my nearly 25 years as an analyst, I have to admit that I've never seen advances move faster and competition get more intense than we've seen in the world of generative AI over the last few weeks.
No sooner had Google introduced an impressive raft of AI capabilities for its cloud platform and Workspace productivity suite - or more likely, that was Google trying to steal some of its competitor's thunder - then OpenAI unveiled the latest version of the GPT-4 large language model.
Now, less than 48 hours later, Microsoft is debuting Microsoft 365 Copilot, offering generative AI capabilities for Office, as well as related updates for its Power Platform for low code/no code development. In between, Microsoft also confirmed that it had been using an early revision of GPT-4 in Bing Search. The company is using GPT-4 to power Microsoft 365 Copilot as well.
You could argue this latest Microsoft news were widely telegraphed by the company with the release of Bing's AI chatbot a few weeks back. There were some who wondered why Microsoft didn't bring these capabilities to Office first. But now that it's been officially unveiled, it's worth diving into what they offer and what their potential impact could be.
The integration of Copilot into Word, Excel, PowerPoint, Outlook, and Teams delivers a number of the features we've quickly come to expect from these types of tools. You can use Copilot to generate new text-based content, analyze numeric data, create new designs, summarize email, chat discussions, or other content, and much more.
The word "Copilot," which the company very intentionally chose, is meant to clarify the type of role that AI content is meant to play and, at the same time, emphasize the continued importance that humans still play in creating and controlling it.
In some cases the new capabilities build on some simpler AI features in previous versions of Office, such as the incredibly useful Designer feature in PowerPoint, which can turn basic slides into an elegant presentation.
With Copilot, the ability extends to building the entire presentation from something as simple as a few ideas in a different Word document. In addition, there are completely new features, such as Excel data analysis, visualization, and natural language formulas.
What really makes these capabilities so potentially impactful is that Copilot can use your own existing documents and other data through what Microsoft has long been calling the Microsoft Graph, making the outcomes immensely more personal.
Technically, Microsoft Graph is an API that developers can use to build customizable applications. The company's been talking about the Graph for years now, but I'm guessing that up until now, most people didn't know much about it. The Graph lets applications "see" things like your emails, calendar, documents, usage patterns and other information that's stored in Microsoft's cloud such as Outlook.com, OneDrive, Office 365, Teams and more.
For the purposes of a generative AI-based tool, that information becomes incredibly valuable, because it provides important context to content generation. In other words, Copilot doesn't just create content or make suggestions from scratch, it does so knowing the contents of the documents you've been creating, the emails you've sent and received, your meeting schedule and summaries, and more.
The net result is the ability to function as an incredibly intelligent personal digital assistant, as well as a useful generative content tool. For example, Copilot can find trends in Excel data, generate emails based on past messages, create PowerPoint designs from other documents you've worked on, etc.
Honestly, it's what we always dreamed Cortana, Siri, Alexa, and other digital assistants might eventually become if they were truly intelligent - but that we sadly knew they never could. In its best form (and most importantly, if it truly works as advertised) Microsoft 365 Copilot offers a new way of working and thinking about productivity in general. It really could be that big of a deal.
As with Google's related announcements, there's still a lot more vision in Microsoft's latest news than reality. There are also lots of things that will take time to unpack and fully understand. Some of the demonstrations of what 365 Copilot can do, however, are pretty mind-blowing.
Like Google's AI offerings, pricing and business models have yet to be determined and initial availability will be limited to a small set of trusted testers (20 companies in Microsoft's case). Though the underlying GPT-4 model's capabilities have been further refined, it's still possible for Copilot to create false information ("hallucinate") and potentially incorporate biases or other issues in its responses. This is an industry-wide problem that everyone continues to work on, but it's an important caveat to bear in mind when using any of these AI tools from Microsoft, Google, OpenAI, or any vendor.
The promise of generative AI tools continues to amaze, however, and that's undoubtedly what has everyone so excited about their potential. The incredibly competitive environment for generative AI is also great for all of us, because it's bound to push these and other companies into building better and better tools. Let's not forget that we've heard virtually nothing from Amazon, Apple, and other large tech companies about their AI plans, and we're bound to see a bunch more exciting startups enter the fray as well.
Microsoft's Copilot highlights how we're starting to see new types of applications for generative AI as well. The new Business Chat tool that will be available via Teams, includes the ability to apply the personal digital assistant-like capabilities that you can use as an individual to group communications (including the ability to incorporate files from multiple people in a chat, help organize and summarize meetings, and more).
The latest AI extensions for Power Platform expand earlier announcements to include the ability to simply describe how you want an app or macro to work and have the tool create the code for you. As I described a few weeks back (see "The surprise winner for generative AI"), this really takes the notion of a citizen developer to a whole new level and, in the long term, could end up being one of the most impactful outcomes of these developments in LLM-powered generative AI tools.
The ultimate success of any of these tools will eventually come down to how effective and useful they prove to be in widespread, real-world use, and that has yet to be seen (ChatGPT has painted a very promising picture nonetheless). There are still enormous questions to be answered about reliability, stability, usability, and more when it comes to LLMs, the AI tools and services they enable. Only with the benefit of time will we truly be able to evaluate how impactful (or harmful) these developments prove to be. But it's hard not to get excited about the kinds of possibilities they might enable, and for now, that's going to drive a lot of advances in a field that is bringing a once-in-a-generation sense of vigor to a previously moribund tech market.
Bob O'Donnell is the founder and chief analyst of TECHnalysis Research, LLC a technology consulting firm that provides strategic consulting and market research services to the technology industry and professional financial community. You can follow him on Twitter @bobodtech.