Practical next steps with AI for publishers

A few episodes back I applied the Gartner Hype Cycle to AI. (I’ll provide a link below to that article.) I thought we were still climbing towards the peak of expectations, but there are signs we might already be heading towards the trough of disillusionment. At least with large language models.

In fact, I might be late to that party. Wired had an article back in April titled “OpenAI’s CEO Says the Age of Giant AI Models Is Already Over.” (Link below.)

Sam Altman – he’s the CEO just mentioned – says the research strategy that led to ChatGPT is already played out, and people will need to come up with new ideas to move things forward. Simply building bigger models is not an option. The costs are enormous, and there are diminishing returns.

Nick Frosst, a cofounder at Cohere who previously worked on AI at Google, … says that new AI model designs, or architectures, and further tuning based on human feedback are promising directions that many researchers are already exploring.

But don’t count on it being under the Christmas tree. Tech advances don’t arrive predictably, or on schedule.

So what do we do right now? AI is already here, and it’s already disrupting the publishing business. We need to build based on what we have, warts and all. Large language models aren’t perfect. They still hallucinate, and say stupid things from time to time. But even at that, ChatGPT is pretty useful.

Here’s an example of what publishers can do now.

Skift built an AI chatbot for their site based on ChatGPT. It took off with a bang, but that might just have been curiosity over the new shiny object. They’ll be incorporating it into their subscription offering. You get three queries for free, but then you have to sign up.

The interesting thing about that story is that it didn’t cost hundreds of millions of dollars, as ChatGPT did. It took two programmers eight weeks. They’ll have to tinker and refine, because it still has some limitations, but that’s not an insurmountable cost. Publishers should be able to manage that.

As I argued in a previous podcast, publishers need to start doing this. Readers are going to expect answers to their questions, not just a bunch of articles.

But don’t make the mistake of assuming that ChatGPT, or Midjourney, define AI. There are still a lot of things coming, so it’s possible we’ll see waves of Gartner Hype Cycles stacked on top of one another as different iterations of AI come on the scene.

Links

OpenAI’s CEO Says the Age of Giant AI Models Is Already Over

AI and the Gartner Hype Cycle

Collect data today for AI wonders tomorrow