Podcast: Play in new window | Download
Subscribe: Apple Podcasts | Google Podcasts | Stitcher | TuneIn | RSS
May HabibThe use of artificial intelligence in content design and content operations is emerging and evolving quickly.
As is the case with many new technologies, it might at first look like robots are coming to steal jobs from humans. But, according to May Habib, AI is more likely to create more, and more interesting, work for people.
May and her team at Writer are developing an AI platform that’s designed to support human creativity and improve content operations.
We talked about:
- Writer.com, her startup that makes an AI writing and editing platform
- the difference between large language models and actual AI
- how AI can support human creativity and content productivity
- the give and take between AI and humans across the continuum of creative and writing processes
- the evolution of the traditional style guide with the arrival of AI
- the difference between genuine human creativity and pseudo-creative AI creations
- the implications of AI for the structure of the content workplace
- the kinds of writing that will always be created by humans
- the many ways that AI-driven innovation can be injected into content practice
- one of the most exciting possibilities for the application of AI to content: to achieve the kind of brand consistency you can only get to this point by having the same leadership and the same talent working together for 20 year
May’s bio
May Habib is CEO and co-founder of Writer, an AI writing assistant for teams. She is an expert in natural language processing, AI-driven language generation, and the evolving ways we use language online.
Connect with May online
Video
Here’s the video version of our conversation:
Podcast intro transcript
This is the Content Strategy Insights podcast, episode number 120. Your news feed has included a lot of stories lately about artificial intelligence. From reports on auto-generated news stories to concerns about sentient bots, AI seems to be everywhere. Another place that AI is appearing is in the content workflows at prominent tech companies and other businesses. May Habib is the CEO at Writer.com where they’re building an AI writing and editing platform designed to support human creativity and improve content operations.
Interview transcript
Larry:
Hey, everyone. Welcome to episode number 120 of the Content Strategy Insights podcast. I’m really happy today to have with us May Habib. May is the founder of Writer.com, an AI writing assistant for content folks. Welcome to the show, May. Tell us a little bit more about what you’re up to these days at Writer.
May:
Hi, Larry. I am so excited to be on the show, and I am day seven Covid, so my occasional cough is not contagious, I promise. Writer is an AI writing and editing platform. We are about 50 employees based in San Francisco. We are remote first, so lots of us working all over the place. I myself split my time between London and San Francisco. We are on a super fun mission of really helping infuse AI and technology throughout this writing continuum, a continuum of ideation, writing, editing that all of us go through when we’re working on original content for work. It’s been a super, super amazing ride building out this product with, I think, some of the smartest people in the content world as customers.
Larry:
I know some of them, and I think you’re right. But it’s interesting, and you said that you spent your time between London and San Francisco. One of the things like AI is always in the news these days. Just in the last few days, The Economist experimented with writing their lead story with AI, and there’s also, we’re right around the time of the famous Google engineer who thought his bot had turned sentient. Do you have to deal with that in your job? Like the ongoing glut of news about AI and it’s… Well, I guess I’d love to get your take on how accurate do you think the news is about AI, or is there a lot of like…
May:
Yeah, well, it’s fun because we don’t have to really go to the news to kind of be dealing with a lot of these issues. We are a few weeks from announcing something very exciting that’s been in the works for a few months, and that we’ve had customers on for a few months already. And so, the kinds of questions that are in the headline, really the weekend through this week around large language models at Google, or The Economist kind of in-depth piece on AI over the weekend. We literally are dealing with them every day. I mean, the number of times someone has gotten a demo of Writer and said something like, “Oh, wow, you all are…” Half joking way, half not, “coming for my job.” I mean, it literally happens every day. Of course, the answer is no.
May:
But throughout the last 30 years of innovation, every time a group of people has asked that question, it has meant that we’ve been on the precipice of either new jobs being created in that field or the job itself being reinvented. I do think we are helping usher in a type of content person who is half data scientist, half editorial leader. We can talk more about what that means, but the TL;DR on the exact question, I don’t think large language models are general intelligence. God, if large language models were general artificial intelligence, that would be a real letdown for me, because that is not what I was expecting. I definitely am expecting something more than that. But is it very, very human-like? Is it indistinguishable? Yeah, the answer is yes. Would you trust it more than a seven or eight year old to make decisions that your life depended on? I think not. I would actually still take the eight year old.
Larry:
No, that comes up a lot. This is why I asked about the media hype around AI, because there’s this assumption that we’re close to general intelligence, but that even the most… I had a chance to chat with an AI pioneer at the Knowledge Graph Conference a few weeks ago and was asking him about how close are we to having an AI that’s got the intelligence of a three-year-old child. And he goes, maybe a decade. It’s a ways off.
May:
It’s sounding like you have intelligence. It’s very different than actually having it. And what’s beautiful about these language models is you are training AI to understand human language and to answer really, really specific, hard questions. I mean, it’s just been magical putting it to use for our customers. It is astounding. It is beautiful. The results are poetic. I am so excited for what we are, together with our customers, going to reveal soon. But yeah, there’s a lot of hype, like garbage in garbage out, but a lot of work in, and this is pretty extraordinary technology.
Larry:
You mentioned a minute ago that the continuum of activities that are involved in modern content creation and I don’t know about management as well, but you talked about everything from just planning and ideation and writing and editing, the whole flow, it sounds like we don’t have to worry about our jobs that, if anything, you alluded to maybe creating more jobs or making our jobs more interesting. Can you talk a little bit about how AI kind of can help at each of those stages in the editorial and the writing and creative process?
May:
Yeah, we think about AI supported by humans and humans supported by AI as kind of two separate but super codependent parts of our product. The content generation piece, let’s put it all in that bucket, is certainly AI driven but trained on the content that you want it to produce, which is all not just human ran, but human curated and analyzed and trained and fine tuned. And then, there are the whole host of writing tasks where it is just so much faster, whether it requires a ton of context or it’s a short piece of content or it’s communication, but there’s a whole host of communication where the AI supports the writer in the driver’s seat or the person in the driver’s seat. That is the editorial function or simply the editing function or the branding function.
May:
We think about this continuum of writing tasks that we do from ideation to editing, and depending on where you are in the continuum, it’s going to be more efficient or effective to have AI lead or the person lead with AI assisting. And so, it is this give and take. The beauty that we’re trying to strike is in that layer actually being kind of invisible and us really being able to into it, from a technology perspective, where someone is writing, what they’re writing, what the context is, and when might an edit be appropriate versus an autocomplete versus actually a whole paragraph being suggested versus actually just a spark of an idea.
May:
So, that’s the continuum that we’re referring to. And learning how to use AI or what is the most appropriate and efficient way to use AI without breaking your workflow, actually like your pre-AI workflow, that I think is the new tool belt, is the new skillset that people are developing using our technology. Sorry, some of that was a little vague. I’ll be able to-
Larry:
No, but it inspires a lot of questions. I’ll say. For example, it sounds like that you can make the case that at every stage in that process, there are ways that AI can help you, or that you can go to AI. It sounds some of what you were saying made me think, if you were just crisp and on top of things and you could remember everything that you’d ever learned and had your style guide memorized. Is that kind of how you picture, is that sort of how it works in application?
May:
Yeah, exactly.
Larry:
Yeah. Okay. Yeah. Well, I guess, and to that, things like one of the big concerns of any content designer or UX writer is needing to abide by a style guide or voice and tone – all kinds of guidance is provided. Can you talk a little bit about how AI helps with that?
May:
Yeah, absolutely. I think, first of all, style guide, the word needs a rebrand because this is not a nice to have kind of, like, “Please follow this style guide.”” Most organizations we talk to have pretty mature content teams. And so, this is the body of work that is the result of a deep understanding of the audience and the business objectives and the brand and the language that helps them really nail the messaging to that audience. And so, complying with a set of guidelines for most of our customers is not nice to have. It’s a difference between their content hitting the mark and achieving objectives versus absolutely not nailing why it was written in the first place, whether it was an narrow message or a blog post.
May:
And so, AI there is really about ingesting that style guide in the form of configurations and settings, and then, being worked back into a workflow. It’s AI versus rules based, because most simple rule in a style guide that you could imagine, everything from capitalize the B in black when we refers to the Black community. If you aren’t using artificial intelligence to code up that rule, we’d be asking you to capitalize the B in black when we refers to a blackboard or a black car. And so, the AI that is presented by really deep learning models, we don’t use transformers everywhere. It’s kind of sometimes a nail looking for a hammer or whatever the saying is. But whether it’s deep learning or transformer models, we are taking an AI approach to just about everything in that style guide. Obviously, numbers and dates and things that really are very rules based. So, that’s the AI in it.
May:
To be able to take someone’s 60-, 70-, 80-page style guide and turn that into a Chrome extension, that’s what… We’re world class at that, and that was the start of our business.
Larry:
Cool. You just mentioned the term transformer model, that I don’t think all of our folks might be familiar with. Can you talk a little bit about what that is?
May:
Yeah. A lot of the fuss, GPT-3, Open AI, the big models from Google, everything that you’re hearing as getting the closest to general artificial intelligence really is the work of these LLMs, large language models. Large language models almost always is describing transformer technology. The transformer refers to really a breakthrough in a set of algorithms that together with the principles of deep learning have resulted in, and of course, a shit ton of processing power, have resulted in programs that can answer questions and prompts, really just as well as a human can better actually, in terms of recall and depth and precision.
Larry:
Right. So, that kind of gets at one of the strengths of AI. In terms of, if you had a perfect memory and knew everything in the world, you could have that. We’re at the stage in computing power and programming elegance that you can actually do that, but that’s just basically parroting facts. . .
May:
You’d do it better, because our human creativity is intentional, whereas LLM creativity is by accident. Oh, that was really creative, that literally was by accident. Whereas when a human is creative, thanks to them remembering a lot of what they read or creating something novel, they know that they’ve been creative. You know what I mean?
Larry:
Yep. It sounds like you can get a little bit of a virtuous cycle of human creativity with this insane body of facts and knowledge that an AI bot provides us.
May:
Facts and knowledge are not things that you actually are getting out of GPT-3 or a lot of the language models, like the fact filtering – that’s actually something that we do and are working on doing more of – it’s very dangerous in the wrong hands. I don’t mean elections and like fraud and stuff like that. But I mean, in work settings, when a fact is presented with a shit ton of supporting material, and it really sounds like a fact, people aren’t going to check that it’s a fact. But everything from when was Teddy Roosevelt president to who won the 2016 election, you are getting a lot of misinformation out of these language models, at a steady clip. So, it’s not rare. It is frequent. And so, folks who are using them, our customers included, know what to watch out for, and we structured things in a way that reduces that. But yeah, this is not about facts and knowledge, for sure. Every single actual fact needs to be fact checked.
Larry:
Interesting. Because you’re right, those large learning models, they’re just learning from general sources, which are full of all kinds of misinformation. And they’re making invalid inferences about facts in the world based on what they’re just they’re mimicking, right?
May:
They’re mimicking. Right. And so, the transformer technology is literally what word comes after this word, based on everything you’ve heard and read. So, the facts are constructed to mimic facts that the model has ingested versus actually semantically answering questions. Google’s AI, on that front, in terms of just how they’ve set up their SERPs and just the search technology, to me is actually much closer to generalized AI than some of these LLMs, on that definition. It gets its facts straight. And so, I know they’re using a lot of interesting stuff under the hood that pertains to the big LLMs, but they’re the different applications.
Larry:
That’s right, Because that’s the way you just said. And I think maybe I was conflating because it’s all Google, and Google has that giant knowledge graph that runs their search engine, but that is more semantic and based on something closer to knowledge and facts than the large learning models. Well, I mean, that’s one of those things where I’m fairly well informed about this stuff, but even I’m making big mistakes like that. How can people be good consumers of AI technology and wisdom and lore, and how can we best understand AI in how it can help us as content folks?
May:
Good question. I think the body of work that we’re about to embark on really kind of bring people in, I think is going to help a lot. It’s still very early days. With Ryan Law, from Animalz, we’re actually doing an AI-generated content masterclass to a select group of folks. We’ll be promoting that soon. It is still the provenance of a few, and everything from the SEO implications to how you structure your team when you’re ready, I think there’s a lot of learning that we’ve had, that we are going to be sharing from customers, really doing this at the bleeding edge over the last few months.
May:
I think it’s here and it’s here to stay. We are just at the very beginning. The use cases that I think are going to be the most exciting are ideation and creativity on the one hand. And then, really the kind of the rote production of things, of content types that are less creative but are necessary. Everything from executive bios to employee profiles to case studies. AI is also taking a 3,000-word case study and turning it into 200 words that sales can use much more interactively. And so, those are a lot of the use cases that we’re excited to be talking about soon.
Larry:
When I think of… You’re reminding me of the early use cases of AI in writing, in journalism, sports, and business stories, just for all, almost all immediately done by AI. You’re watching this rollout in the content world and you just mentioned a few of them, like the executive summaries and bios and things like that. How do you see it all unfolding? What do you think is going to be the easy, quick stuff that like, oh yeah, AI can help us with this right away and it’s going to be everywhere, what do you think, from that end of the spectrum to like, boy, it’s going to be hard to ever really do this with AI?
May:
Yeah. I’ll start with the last one. I think the journalistic style, thought leadership, interviewing your customers, interviewing thought leaders, really presenting new information, let’s not forget a lot of these models they’re trained on 2021 maximum. Most of GPT-3 is stuff that’s 2019 or older. And then, certainly not future-focused.
May:
So, I think all of the stuff that we already know sets brands apart is quite safe. How do you get an intro? How do I wrap this up? Sure, you can use ideas for that, but you can use AI for ideating on those kinds of things if you’re stuck. And I do feel like having used it myself now for months, for my own writing, including for our own content marketing and columns and things like that, it is super handy for the kind of… I call it my personal self-loathing when I get writer’s block. It’s really good to keep you in a state of flow.
For those of us who are dopamine addicts, sort of keeping the affirmations going, keep you motivated to finish your damn fucking piece.
May:
So that, I think, is for the neurodiverse among us a underappreciated angle. I think thought leadership and the stuff that is always crème de la crème, that is absolutely safe. I think the beauty is in the things that you can taxonomize. I don’t mean necessarily like intro paragraph and 20 words on this and 30 words on that. We can be more creative. This is what’s beautiful about the way these models work is we can train them on customer content that looks like what they want. It doesn’t have to be exactly replicas. You don’t want that, because then it’s too formulaic.
May:
There’s a nice now result possible that gets you a lot of the way there that is not the baseball scores of yesteryear. This person scored that many, and I don’t know if you score in baseball. I don’t know my baseball terminology, but these are very formulaic articles or earnings reports, things like that, which literally were, kind of think about it as like, painting by numbers. This is how the sentence is going to be, and this is what we’re going to do. These are the stats we’re going to put in. Those companies did fine, and they’re still around and still do that business. So, yeah, I think it’s a whole new world. I’m so excited. We’ll have to come back on this webinar, Larry, when-
Larry:
Yeah. Well, no. What you just said too, it sounds like human creativity is always going to be like a super important part of this, regardless of how far along we get. Another thing you just said, I’m really curious now about you mentioned the training models that generally AI’s built on, but also I’m just wondering specifically, with your product and other writing products, are you using general models to inform the advice you give with your product? Or do you also learn from the corpus of the… Do you look at all the content marketing that a company has done to make sure they don’t do any… Tell me a little bit about the content you’re working with, that you’re training your models with.
May:
Yeah. It’s really cool. It’s trained on customer data and not sharing customer data. So, we are a big privacy shop and actually we don’t ingest any customer data at all or train on customer data, but in the product information that customers put in for the purpose of training, trains their own model.
Larry:
Got it. Yeah. Hey, I’m just curious, you’ve mentioned several applications of this, and I’m wondering now about just the general broad new things in the content world, things like chatbots and virtual agents and voice assistants and that kind of stuff. A lot of those, I know we’re kind of circling around a lot of interesting stuff that’s happening in the technology and content worlds, like that autogenerated stuff, but also the more intelligent NLP kind of stuff, tell me how your product and others are kind of circling all that, the semantic stuff, the NLP stuff, the knowledge stuff, and how much of it is just taking advantage of the general information you get from these large learning models?
May:
Yeah. It seems super easy to be very overwhelmed. I was just overwhelmed listening to the question. So many of us are trying to do more with less in this market. And the table stakes is good work. We’re, in a way, lucky that we don’t actually have too many people wearing strategic hats that sit between content teams, because I think those people would be so overwhelmed just to be paralyzed with all of the different ways they can be introducing AI-driven innovation into their content practices. We almost always are talking to a head of the knowledge base or the head of the blog or head of editorial or head of product writing. And so, companies and teams can really conceive of these use cases with us in very specific ways, and we can grow from there.
May:
The folks who are working on chatbots, and we do have customers who are using Writer for chat content or for bot content and for kind of conversation graphs, have figured out a way to include AI-generated content in that for human review. Folks who are using us in product content have figured out a way to help us accelerate that ideation and content generation piece. So, it is very overwhelming to think about the confluence of how AI is really changing things, but the reality is people in seats are hired to do one or two very specific things. If you can conceive it, if you can think about AI writing it, it probably can.
May:
The question is, what are the ways in which it changes the workflow. And the content generation piece, once you figure that out in a way is just step one. If you have to promote a piece and distribute a piece and the piece needs art, there are so many different downstream parts of this funnel that just accelerating the content production piece alone doesn’t help you meet the business goals. You’ve got to get it into the hands of people.
Larry:
Right. That, and again, you just counter-overwhelmed me when I thought about that, what you just said, about the implications for workflow and the whole process. So, it’s really early days, it sounds like.
May:
Absolutely.
Larry:
You’re kind of pioneering a lot of this stuff. What do you see as the biggest potential? What excites you the most about this emerging world?
May:
I’m most excited for the kind of control consistency and scalability you get. I’m a control freak. I’m the oldest of eight. I’ve always been a control freak. When you can think about with a bird’s eye view, all of the different types of content and communications and copy that your company is producing, and knowing that there is one brain that is helping influence how all of that turns out, the accelerating and just kind of additive impact that has on creating a brand, that’s revolutionary. Unless you have been like American Express, we don’t work with American Express, and you’ve had the same CMO for 20 years and worked with the same very expensive agency for 20 years, you don’t get that kind of consistency.
May:
Most of us aren’t outsourcing everything to one hive mind. It is a bunch of different people and a bunch of different teams with a lot of different writing abilities, all trying to be on brand and actually having a system. We call it the writing OS where you are getting the same set of rules, same set of AI, same brain helping influence all of that. That’s bigger than any one piece that was accelerated by AI. That’s transformational. And I think the companies that figure that out the soonest, it’s an overwhelming competitive advantage. I think a lot of our early customers aren’t going to get there soon.
Larry:
That’s great. I love that the comparison to a 20-year relationship with a really good agency. But if AI can get you there even in a couple years faster, that’s great. Yeah. Well, May, I can’t believe that we’re already coming up on close to time.
May:
Wow. That was fast.
Larry:
Well, it always goes way too quick, and I always want to talk to you all forever, but anyhow. Before we wrap up, is there anything last, anything that we didn’t, you didn’t get a chance to elaborate on fully, or just this on your mind about AI and writing or anything in the content world?
May:
Yeah. If I were an IC, I would want to be exactly in this moment, I think. It is so exciting to think about all of the ways that people can add value to teams right now, thinking about doing things a little bit differently. We’re all trying to stand out with content that delights, that inspires, that informs. A lot of people right now are using AI-generated content to kind of do the opposite. It is just a shit ton of crap, and I can’t wait until Google figures out and penalizes people who are doing that. That might sound strange coming from someone like us, but it’s not at all. I mean, we are all about AI-generated content helping the individual, and fully expect that all of that content is worked through by the time that it’s published. So, yeah, I think there are a lot of open questions, but there always are when you’re at the start of something new. So, exciting days.
Larry:
Well, thanks. Yeah. I’m optimistic that Google will figure out how to sidetrack all the misplaced uses of AI.
May:
Yeah. I mean, just track down the fake facts.
Larry:
Exactly. Yeah. They’re good at that.
May:
Oh yes. They’re good at that. Exactly.
Larry:
Hey, one last thing, May. What’s the best way for folks to stay in touch with you if they want to connect on social media or anywhere online?
May:
Yeah. Well, follow me or add me on LinkedIn. I tend to post my thoughts there more often than on Twitter, and then anything more personal than that, I’m just May at Writer, and we’re at Writer.com.
Larry:
Great. I’ll put that in the show notes as well. Well, thanks so much, May. Really fun conversation.
May:
This was super fun. Great questions.
Leave a Reply