Writer’s GPT-powered CoWrite handles content ‘drudgery’ and leaves creativity to humans – TechCrunch

Writer is an AI-driven tool for examining and coaching content creators in organizations where voice and branding are essential. Its new feature, CoWrite, lets you write your own — but don’t worry, it’s not the content apocalypse we’ve been worrying 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, this means giving a large general-purpose model a specific set of content to mimic more closely than any other language it understands – sort of like telling an image creation model to make a certain style of picture by giving examples.

Writer’s tools already do this to a degree, ingesting style guides and other data to provide a real-time style check service: “use this preferred word instead of that” or “use active voice in the title”, depending on you what the organization likes.

But as founder and CEO May Habib explains, organizations with strong brand presence find themselves in a bind.

“The number of channels they have to persist is exploding. No matter how big the team is, they can’t keep up,” she said.

Writer’s solution is to use a fine-tuned GPT-3 model to directly generate the content in question, but understand that this is very user-friendly. Although Habib prefers the term “content automation” to “generated content” as the latter has some negative connotations.

“The most important thing is to be done by people,” she stressed. “When you put together a newsletter, it’s half the drudgery, right? It’s about freeing up these people to do the most creative parts of their job, the activities, and strategies to win eyeballs, by implementing effective Things. This is for mature, sophisticated content teams trying to do more.”

Image Source: writer

I wouldn’t know about the newsletter because I left that particular chore to my coworkers, but it’s true that there’s a lot of rote work and any writing can be minimized if you work within the existing framework. Even a rough outline would help, but the problem is that someone has to make the outline – again a chore.

CoWrite is not meant to just blast out the final copy, although a little tweaking of GPT-3 might do it, depending on your expectations. In this case, it’s more about making a plausible blog post or newsletter that the content team can look at and say “something like this, but you actually wrote it”.

Animation showing how to use CoWrite to draft a blog post. Image Source: writer

There are other people working in this field – in fact, to be more precise, almost every large company working on large language models is at least working on it. But Habib says it’s not just about having the capability, it’s about integrating it with the existing workflows of companies that are content-focused.

“Most of the people we talked to had access to the GPT-3 API, so it wasn’t the first time they heard about it. They called the models themselves,” she said. “The difference with Writer is that it’s an app designed for content users; it’s integrated with their style guide, brand guide, brand tools, taxonomy. The interface is already there: we’re talking about everything from training titles to generating Three clicks for the title. No one is doing that at the moment.”

While this might be reminiscent of a “bad future full of dystopian ads” where we all dread language generators behaving erratically, Habib believes it’s closer to the early days of robotic process automation.As with automation as a whole, the most likely outcome is that “dirty, dangerous and boring” jobs will be left behind, and in the content world it’s like doing a bulleted outline for a weekly newsletter – dangerous, no, but definitely boring

Leave a Comment

Your email address will not be published.