Writer is an AI-powered tool for checking and guiding content creators in organizations where voice and branding are essential. Its new feature CoWrite does that writing itself — but don’t worry, this isn’t quite the content apocalypse we’ve been worried about.
CoWrite is the latest in a new wave of tools that use large language models like GPT-3, but modify them using “fine tuning,” a common phrase but with a special meaning in the machine learning world. Basically it means giving the big, general model a specific set of content to imitate more closely than the rest of the language it understands — a bit like telling an image creation model to make a picture in a certain style by feeding it examples.
Writer’s tools already do this to a certain extent, ingesting style guides and other data to provide a live style-check service: “use this preferred word instead of that,” or “use active voice in headlines,” depending on what your organization likes.
But as founder and CEO May Habib explained, organizations with a strong brand presence are finding themselves underwater.
“The number of channels that they have to be present in continuously is just exploding. No matter how big a team is, they can’t keep up,” she said.
Writer’s solution is to use a fine-tuned GPT-3 model to straight-up generate the content in question, but with the understanding that it’s very much human-in-the-loop. Though Habib prefers the term “content automation” over “generated content,” since the latter has something of a negative connotation.
“The most important stuff is being done by people,” she emphasized. “When you put together a newsletter, half of it is drudgery, right? This is about freeing those people up to do the most creative part of their jobs, the campaigns, the strategy to win eyeballs, by operationalizing the things that work. This is for established, sophisticated content teams trying to do more.”
I wouldn’t know about the newsletter since I leave that particular drudgery to my colleagues, but it’s true that there’s a lot of rote work in any writing that can be minimized if you’re working from an existing framework. Even a rough outline helps, but the problem is that someone has to make the outline — drudgery again.
CoWrite isn’t meant to just blast out final copy, though GPT-3 with a bit of tuning could probably do that, depending on your expectations. In this case it’s more about producing a plausible blog post or newsletter that a content team can look at and say “something along these lines, but you actually wrote it.”
There are others working in this space — actually, it might be more accurate to say nearly every large company working in large language models is at least looking into it. But Habib said that it’s not just about having the capability, but integrating it with existing workflows at companies where content is a major factor.
“Most people that we talk to have access to the GPT-3 API, so they aren’t hearing about it for the first time. They’ve played with calls to the models themselves,” she said. “The difference for Writer is, this is an application meant for content people; it’s integrated with their style guide, taking their brand guidelines, branding tools, taxonomy. The interface is there already: We’re talking about three clicks from training headlines to generating them. There’s no one doing that right now.”
While this may bring to mind the “shitty dystopian ad-filled future” we all fear from language generators run amok, Habib believes that is closer to the early days of robotic process automation. The most likely outcome, as with automation at large, is that the “dirty, dangerous and dull” jobs will be left behind, which in the content world is stuff like doing a bullet outline for the weekly newsletter — dangerous, no, but certainly dull.