Nearly two years into the global journey with generative AI, news media companies are moving beyond the experimentation phase and looking at how these new tools can move them into the future.

During this week’s South Asia Webinar, The role of AI and analytics at HT Media, INMA members heard from Yudhvir Mor, chief product officer at HT Media, who shared his insights into how GenAI has improved operations, what hasn’t worked, what media companies should keep in mind as they invest in GenAI, and what challenges lie ahead.

Learning what GenAI can (and can’t) do

Today, adoption of GenAI is high across newsrooms, Mor observed. But despite the evolving sophistication of the algorithms, they still can’t compare to the work of a journalist.

There are, however, ways it has worked “beautifully” and significantly improved operations.

For example, data analytics has become democratised, allowing anyone who can access and use large language models (LLMs) to perform deeper data analytics. Content summarisation has also become sophisticated, SEO is more efficient, and design review processes are streamlined.

“Whenever we are making any big UI/UX changes, the first two levels of design reviews are pretty much automated now,” he explained, adding that the discussions have become more refined and objective because the data is more structured and offers a better view into the specific user experience. Bugs have also been reduced significantly.

For all the upsides and benefits of GenAI, however, there have also been things that haven’t worked, such as auto-generation of complete articles. User engagement-related activities didn’t progress as well as expected at HT Media because the company started experimenting, he said: “I think this is one area where we probably could have done better when it comes to using different generative AI models.”

Where do we go from here?

 What matters now, Mor noted, is how the industry evolves from here: “In 2023, the total investments in generative AI-related applications and the ecosystem is close to $50 billion,” he said, sharing that initially much of that went to infrastructures and building applications.

Now, the focus is on applications.

“This is the key shift that you will be seeing, and I think it has already started playing out,” he said.

HT Media has successfully leveraged GenAI for multiple use cases, including analytics, research, fact-checking, image generation, and more.

A pleasant side effect is that it has helped create more well-rounded editors: “They pretty much know what to write, the performance of the content, the different keywords they need to use, [and] the quality of their content. They’re pretty much managing the end-to-end pipeline on their own.”

As companies look at how to structure their investments in GenAI, Mor cautioned against going all-in on one specific technology and urged news media companies to continue exploring their options:

“It’s still a very fast-evolving space. My opinion and many of my colleagues’ opinions about specific AI models are changing every week because of the amount of things that we see these models are bringing in. My suggestion would be to experiment with most of them because it’s still a little early for you to go ahead with one specific generative AI technology.”

Prompt engineering is another area that warrants exploration, and Mor said it is one of the fastest-evolving skills related to GenAI. HT Media is training its editorial teams and developers how to ask different prompts.

“There is a massive difference when it comes to using a generic prompt and a specific prompt; the outcome is vastly different,” Mor said. “This is an important skill. You need to keep experimenting with it, so really invest in your prompts.”

He also advised that, when investing in learning prompts, an enterprise account offering such a service be used so that the information teams are learning based on their data will stay private.

“It’s still an evolving journey. There are certain use cases where it worked well; there are certain use cases where I think it’s far from what we were expecting,” Mor said.

Tackling the challenges of GenAI

As GenAI continues evolving, Mor pointed out there are challenges along with opportunities. The technology is not fully understood, and ethical concerns persist.

“As we know, many of these models are trained or created on data that belongs to somebody else,” he said. Supervised learning, where humans label data, can introduce biases into the models, affecting their predictions and outputs.

Accountability is another significant issue. Responsibility for the output of these models is shifting from providers to consumers, creating an evolving and perhaps confusing landscape as consumers must become more aware of the implications of using these AI tools and the potential biases they may carry.

Data privacy is also a critical concern; Mor said AI models often use publicly available Internet data, including personal information. As these models improve, they require more data, raising concerns about how much personal information is being shared and how it is being used.

“Are you giving personal accounts to everyone to do these experiments? How are we partnering with your IT teams to make sure that tools are available to the people, but also that the employees and the partners understand the responsibility with these tools?”

As frequently emphasised in conversations about media and AI, Mor said human oversight remains crucial in developing and deploying AI. Despite advancements, AI and human intelligence operate differently, and the combination of both is essential for the foreseeable future.

“Human beings don’t think using linear regression, we don’t think using neural networks or unsupervised learning,” he said. “I think, as of now, whatever is happening in artificial intelligence is primarily because of massive amounts of computers, a great amount of data, and some amazing mathematicians.”

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .