In the ever-evolving landscape of artificial intelligence, there is a prevailing assumption that chat interfaces will reign as the gateway to interact with large language models (LLMs). However, I stand as a sceptic. It is not my intention to kill the chat, but rather to explore its limitations and advocate for a more effective alternative in 2024.
How we got here
I do not doubt that LLMs are here to stay and will have a profound impact on the publishing industry on several fronts including reporting and collecting information, production, etc. A growing concern publishers have is that LLMs will eat up some traffic currently delivered via search. Rather than "Google" a question, people will ask an LLM. This is just one of many situations where I believe chat alone is not the ideal interface.
Google, OpenAI, Anthropic and other LLMs will indeed remain in the privileged position with the audience, providing a base-layer service. Meanwhile, specific information providers (like local news organisations) will need to find ways to squeeze into the conversation. The question is - does this mean we need a three-way chat, provide chat on our sites, or rather start to think outside the [chat]box.
I am sceptical that chat interfaces will be the final form of how we interact with LLMs. If I am right, then much is still up for grabs.
To start, chat interfaces suffer from the paradox of choice. Combine this with the idea that the average user will get frustrated with the nagging notion that they will have to “engineer” a prompt when all they want is a simple recipe and you can start to see where chat limitations come into play. “Was this the best recipe I could have gotten? Did I provide enough instructions, details and nuance in my request? Let me run some A/B tests on this prompt.” These and more are second-order questions nobody wants to contend with on a daily basis.
When faced with a chat box, the mere act of typing or speaking in complete sentences can feel burdensome when a dropdown could suffice and provide more immediate direction. It is not uncommon for users to want a more structured experience, one that offers a clear path forward. You can see this emerging already in Google’s “Search Generative Experience” or even when opening up ChatGPT and it gives examples of queries to ask “Tell me something interesting about the Roman Empire.”
Chat by itself lacks this crucial aspect and I worry that in their rush to reproduce what early generative AI products have shown us news organisations will muddle too long letting folks chat with articles when other UX paradigms would perform better.
This has happened before
To illustrate my point, let us journey back to the early days of the social web. AOL's chat rooms allowed users to engage in discussions on topics of interest, such as philosophy, politics, or San Francisco. The chats were never-ending and free-form. It showed the potential of the “social web” but it was Twitter’s constraint of 140 characters or Instagram’s encircling photos that defined the era.
Our interface with artificial intelligence differs from the social web, as it involves communicating with an AI rather than fellow humans. Nevertheless, I believe we are currently residing in the AOL chat room era of AI interfaces. We are using a low-resolution, catch-all solution that, while brimming with potential, may be overwhelmed by its own vastness. To truly thrive, we need an interface that possesses a more defined perspective.
While I cannot predict what this interface will look like, I am convinced that at the very least, chat-alone is not it.
As news organisations trip over themselves in 2024 to create AI-powered experiences, I hope we can think outside the [chat]box.
David Cohn is the co-founder and chief strategy officer of Subtext, and senior director of Advance Digital, where he leads the 'Alpha Group,' an internal research and development team. He spent seven years as a freelance journalist, writing for titles including Wired, The New York Times, Seed Magazine, Quartz and Columbia Journalism Review.
Free daily newsletter
- New blockchain verification tool to help fight the use of deep fakes in elections
- Trust is not a "useless metric" - we just need to understand it better
- New playbook to help publishers boost content strategy
- How Ruben Reuter challenges disability prejudice in the media industry
- News for All explores how journalism can fix its trust issues