Confluence for 1.14.24
Confluence for 1.14.24
OpenAI introduces ChatGPT Team. State of Pennsylvania partners with OpenAI. Duolingo cites AI as reason for contractor reduction. Large Action Models and Rabbit R1. Exploring ChatGPT's new GPT Store.
This was a big week in the world of generative AI, with OpenAI launching its much-anticipated GPT Store as a collecting point for user-created custom GPTs (more about that below). We’ve shared a number of our own custom GPTs in this space over the past few months, and plan to make a number of them available via the store (without fee), and via a menu tab here at Confluence, as well. Look for that in the next few weeks. With that said, here’s what has our attention at the intersection of generative AI and communication this week:
OpenAI Introduces ChatGPT Team
State of Pennsylvania Launches Partnership with OpenAI
Duolingo Cites AI as Reason for Contractor Reduction
Rabbit R1 Introduces Large Action Models (LAMs) into the Conversation
Exploring ChatGPT’s New GPT Store
OpenAI Introduces ChatGPT Team
Smaller teams can now use ChatGPT securely.
Since the launch of ChatGPT in November of 2022, data protection has been one of the biggest concerns of organizations and individual users. Organizations have been rightly concerned about employees inputting sensitive, confidential, or proprietary data into publicly available tools like ChatGPT, Microsoft Copilot, or Google Bard, for fear that those providers could use that information to train their underlying models, and for reasons of data security and privacy. Arriving in the second half of 2023, Microsoft 365 Copilot and ChatGPT Enterprise (among others) allayed these concerns with promises of enterprise-grade data security. The catch, until now, has been that these enterprise-level solutions have only been available to larger organizations. Microsoft 365 Copilot, for example, requires a minimum of 300 subscriptions.
That changed this week with OpenAI’s introduction of ChatGPT Team, which offers a solution for smaller teams and organizations (you need a minimum of three seats at $25 or $30 per month per seat, depending on the contract, to start). While for us the headline is the data security and removal of data from training, Open AI notes several other features included with ChatGPT Team, including the ability to create and share custom GPTs within your organization. As a boutique consultancy with around 100 colleagues, ChatGPT Team is what we’ve been waiting for, and we’ve already begun piloting it with a handful of users across our firm. We will share what we learn here as we go.
State of Pennsylvania Announces Partnership with OpenAI
Governor Josh Shapiro wants to use AI to make government run better.
This week, the Shapiro Administration and OpenAI announced a partnership to bring ChatGPT Enterprise to Pennsylvania’s Office of Administration. The partnership will begin with a pilot, with the potential to expand to more employees and more departments over time. Even at this initial scale, we think the partnership is notable for several reasons.
First, it’s a public example of a government of size embracing the potential of generative AI. While governments can be slow to adopt new technologies, this partnership suggests that Pennsylvania understands the potential of this technology to shape how people work in and with organizations, and to create upside for state employees and Pennsylvanians.
Second, the Shapiro administration is transparent about where they expect to see the benefits of AI. The administration’s press release shares the use cases the Office of Administration will explore:
“The ChatGPT Enterprise pilot will begin in January 2024 and is initially limited to OA employees who will use the tool for tasks such as creating and editing copy, making outdated policy language more accessible, drafting job descriptions to help with recruitment and hiring, addressing duplication and conflicting guidance within hundreds of thousands of pages of employee policy, helping employees generate code, and more.”
Based on the research and our own experiences with ChatGPT, these seem like good places to start, though we believe there is as much — if not more — benefit from using generative AI tools as thought partners as there is as a means of outsourced labor. Depending on how transparent the state government is with what they learn and how quickly they move, this effort could serve as a blueprint for how organizations introduce generative AI.
Finally, there is — for now at least — a commitment that Pennsylvanians will not interact with AI when engaging with government services. We’ll be mindful of whether this commitment holds long term, but it does signal that the administration views AI as an augmentation to state employees rather than as a replacement.
Given the operational scale of state governments and the diversity of services they offer, we’ll be paying close attention to this pilot.
Duolingo Cites AI as Reason for Contractor Reduction
Spokesperson: “We just no longer need as many people to do the type of work some of these contractors were doing. Part of that could be attributed to AI.”
While generative AI was a dominant theme in the mainstream press and in most organizations in 2023, those conversations were largely exploratory: What is this technology? What can we do with it? What might it mean for me, my team, my organization, and my industry? In 2024, while we don’t expect generative AI to immediately affect every business overnight, we do expect to see an acceleration of the application and mainstream impact of these technologies. One area to watch, of course, will be workforce decisions, a case in point of which is Duolingo’s recent contractor workforce reduction of 10%.
As generative AI is more widely adopted and applied across organizations in 2024 and beyond, we’ll need to continually recalibrate the value we add as humans. This is particularly true in corporate communications, where we believe the value people add will shift away from content creation and toward more nuanced skills like judgment, strategic counsel, and curation. We don’t see this as an immediate cause for alarm, but we think it’s cause for proactive thinking (and action) about the unique value humans can add in corporate communications in a world where generative AI tools are increasingly commonplace.
Ultimately, as organizations find, test, and validate use cases for generative AI, the decision-making needs to shift from technical and training concerns (what do we use and how do we use it) to matters of organizational design. Over a broader sweep of years, the adoption of tools like GPT-4 — with it taking over some tasks people used to do, and augmenting others — will affect skill proficiency, skill development, the acquisition of judgement, role definitions, talent development, the definition of career paths, and more. When the generative AI can write a press release as well as (or better than) any junior staffer, how do we create editorial judgement, the ability to write under pressure, and other foundational skills in junior talent? This is only one small example, and there are dozens more. We’re facing into this work now in our firm, and we are beginning to plan for the implications of generative AI on the development of expertise and judgment in our own people. Ultimately, all organizations will need to wrestle with the organizational design implications of generative AI, as the tools shape us just as much as we shape the tools.
Rabbit R1 Introduces Large Action Models (LAMs) into the Conversation
AI as operating system is becoming a reality.
One idea that has quite a bit of traction is that of AI as an operating system: using a large language model (LLM) as a way to engage with the applications you use, rather than as the application itself. An example is being able to ask ChatGPT to find and book travel for you, with it doing the work of interfacing with your travel and hotel applications. This idea is one step closer to reality with a new, standalone AI device called the Rabbit R1.
The device, developed by an AI startup called Rabbit, is well-designed if unremarkable in appearance, but it’s the software that really captures our attention. Rabbit uses what it calls a Large Action Model (LAM) to allow AI to interact with applications we use every day. It’s similar in concept to Siri or Alexa but powered by the combination of a LLM (to understand human instruction) and LAM (to act on those instructions). The video below summarizes the technology well and includes a demonstration of how it works in practice.
An LAM trains by watching people interacting with applications to get things done. LAMs are distinct from LLMs that power ChatGPT, Claude, and Copilot. If Rabbit realizes its stated potential, we expect to hear and learn much more about them in the future. What we’ll pay attention to is less about the success of the Rabbit R1 but rather how the underlying idea develops. If Rabbit proves we can use LLMs with LAMs to make it easier to use applications that already exist, it very well may introduce another step-change in how we leverage AI in our work and in our lives.
We expect this to be the case. One of the things that’s so powerful about LLMs is the ability to engage with the technology using your language, not the computer’s. The creation of the mouse and the graphical user interface made computers and computer applications much more accessible to billions who didn’t have the skill to interface with them via keyboard and computer languages. LLMs will take that to the next level. Prepare for a world where you talk to your technology.
Exploring ChatGPT’s New GPT Store
OpenAI aims to create the “app store” for custom chatbots.
After first being announced at OpenAI’s developer conference in November, the GPT Store officially launched this week in ChatGPT. You can read OpenAI’s post introducing the GPT Store here, and The Verge sums it up nicely in this article. For our readers, the key thing to know is that with the GPT Store, any (paying) user of ChatGPT now has immediate access to millions of GPTs1 created by users across the world. Previously, users could access GPTs created by other users if the creator designated them as public and shared the link, but the GPT Store — like the Apple and Android app stores — allows direct publishing to a central location.
How useful the GPT Store is will depend on the quality and utility of the GPTs it houses. We expect that, as with other app stores, quality will vary and that there will be a handful of standout GPTs, a lot of mediocrity, and a lot of junk. We hope to identify some of the standouts in the coming weeks and months. And on that note, one GPT which has particularly impressed us thus far is Books, which we found vastly superior to a similar GPT that one of us had previously tried to create on our own.
Many of the categories in the GPT Store — Writing, Research & Analysis, and DALL-E (image creation) — are directly relevant to the work of corporate communication. Time will tell how valuable and widely-used GPTs in these categories become, but they will certainly be worth monitoring for everyone in the profession. Consider the top four GPTs in the “Writing” category at the time of our publishing this:
Write for Me — Write tailored, engaging content with a focus on quality, relevance, and precise word count.
Fully SEO Optimized Article Including FAQ’s [sic] — Yoast and Rank Math SEO Optimized | Create a 100% Unique | Plagiarism Free Content with | Title | Meta Description | Headings with Proper H1-H6 Tags | up to 1500+ Words Article with FAQs, and Conclusion.
All-around Writer (professional version) — A professional writer📚 who specializes in writing all types of content (essays, novels, articles, copywriting)...
Automated Blog Post Writer — I craft professionally written, and researched, blog posts in your unique voice.
We haven’t had much time to experiment with these particular GPTs or compare their quality to similar GPTs we’ve created ourselves. The most important thing to know for now is that these tools (and millions of others) are immediately available to anyone willing to spend $20/month for a ChatGPT Plus account (or who has ChatGPT Team or ChatGPT Enterprise). Even before GPTs focused particularly on writing use cases, we were expecting an explosion of content within organizations because the “base” version of ChatGPT and similar tools made creating content so easy. The GPT store will likely turbocharge those dynamics.
Less than a week in, it’s certainly too early to predict the ultimate impact of the GPT Store. While the primary analogue for the GPT Store will be the Apple and Android app stores, we’ll note a key difference between a GPT and a typical mobile application in the Apple or Android stores. Anyone can create a GPT in minutes, whereas traditional applications require a higher level of technical skill. The barrier to entry for GPT creation is lower. Why does that matter? Since any user or organization can create a GPT of their own, the recurring question will be “Should I create this myself, or should I see if there’s something in the GPT store?” The answer to this question over time will go a long way toward determining what the long-term value of the GPT Store will be.
We’ll leave you with something cool: talking cars. As we wrote earlier, prepare for a world where you talk to your technology.
AI Disclosure: We used generative AI in creating imagery for this post. We also used it selectively as a creator and summarizer of content and as an editor and proofreader.For those not yet familiar with GPTs, they are essentially custom chatbots or “applets” that anyone can create in a matter of minutes. We’ve written about GPTs and our creation and use of them here and here, and we’ll certainly have more to say about them as the technology evolves and as we continue to get smarter about them.