Inside LinkedIn’s AI strategy

While some may think of cheesy job postings or the cutthroat job market when they think of LinkedIn, the networking site has also become an unexpected oasis for young people—and a place where artificial intelligence has long been welcomed.

IT Brew caught up with Juan Bottaro, a principal staff software engineer at LinkedIn in Madrid, Spain, to discuss AI features, booted projects, and challenges he and his team faced when tapping into generative AI.

Many people use LinkedIn for networking or for job searching. What made you want to incorporate AI into the platform?

“First of all, I think the term ‘AI’ is—it’s too hard to really put your finger on exactly what it means nowadays,” he said. “But I can tell you, LinkedIn has been doing AI-related work—in particular, machine learning—for more than a decade.”

“When you go to LinkedIn, and you…see your feed—the way that we decide what goes first, second, etcetera—it’s all using AI and machine learning,” Bottaro added. “The same way when we decide what notification to send you, or which people to recommend to you [for connections].”

Bottaro said generative AI is what “really put AI at top of mind for everybody in the past year and a half.” LinkedIn—which, according to Bottaro, has been experimenting with generative AI for close to a year—uses a combo of traditional and in-house LLMs, such as OpenAI models via Microsoft Azure OpenAI APIs.

Can you pull the curtain back a bit and explain how LinkedIn uses generative AI in some ways that readers wouldn’t expect?

“So, we’re trying to look at all these verticals and see where we can really make a difference by incorporating this technology,” he said. “We don’t want to just splash it everywhere and see what sticks.” One way it’s using generative AI is to simplify the job-hunting process for users. With the new AI-powered experience, premium users can click on prompts like, “Am I a good fit for this job?” and “How can I best position myself for this job?”

The AI tool will then analyze that person’s profile to respond, but instead of flat-out saying the job is a no-go, the team wants the feature to be factual and empathetic. If users aren’t a good fit, the tool will then suggest courses for them to take “to get the skills that are needed.”

That’s where a second key area comes in: “knowledge seeking.”

What are a few features that you ended up leaving in the cutting room?

“There’s a few—I mean, initially, we actually started with this idea of a very open-ended chat; that’s what pretty much every company seems to be doing nowadays. [We started with] this idea of, OK, you’re in a chatroom with this magical machine, and then you can ask whatever,” he said. “But what we found is that, the same thing…happens when you want to write a piece or a blog post—facing that blank page is a challenge, right? It’s not easy to [start] from scratch, figure out what you should ask, or how you should write it.”