The media industry is poised to become one of the biggest beneficiaries of the recent boom in AI technologies. Organisations worldwide are exploring advancements in personalization and user experience, with a significant focus on how Generative AI in media is reshaping the landscape. Central to this transformation are Large Language Models (LLMs) like ChatGPT, which are revolutionising content creation and distribution.

In the midst of the growing interest in Generative AI in the media, it’s crucial to delve into the mechanics of these technologies. Understanding the foundation and functionality of Generative AI models, such as ChatGPT, is key for media professionals and journalists looking to leverage these tools effectively. At its core, ChatGPT, like other LLMs, is an advanced statistical model built upon machine learning principles. Its ability to generate text is based on extensive data training, rather than an abstract concept like magic.

LLMs, the backbone of AI in journalism, operate as artificial neural networks trained on large datasets. Their primary function involves analysing input text and predicting subsequent words, thereby enabling them to understand context, generate coherent sentences, answer questions, and even create compelling narratives like stories or poems.

However, the development and training of LLMs in media and journalism represent a significant investment. As noted by AI expert Andrej Karpathy, the process involves two critical stages: Pre-Training and Fine-Tuning. The Pre-Training Stage, which involves processing immense volumes of internet text data (up to 10 TB), can cost around two million dollars. This stage is akin to assembling a vast puzzle, where the goal is to make sense of diverse information pieces. The Fine-Tuning stage, on the other hand, involves refining the model through supervised learning, focusing on specific topics or linguistic styles to enhance the LLM’s proficiency.

Effective prompt writing is key to harnessing the full potential of LLMs in journalism. This involves crafting detailed, context-rich prompts that guide the AI to perform desired tasks accurately. Although these models process information sequentially and lack true intelligence or emotional comprehension, their capabilities are rapidly advancing. Future developments in Generative AI in media suggest that LLMs will become more autonomous in following and interpreting instructions, reducing operational costs and increasing efficiency.

Implications of AI on Journalism Jobs and Industry Practices:

The integration of AI into journalism heralds significant changes for job roles and industry practices. AI’s ability to streamline routine tasks like data analysis and basic reporting is undeniable. This shift prompts journalists to increasingly focus on producing high-quality, creative content, delegating data-heavy and repetitive tasks to AI. In this rapidly evolving media landscape, journalists are urged to develop new skills and adapt to an environment where AI is an increasingly dominant force.

A key insight from the ‘Changing Newsrooms 2023’ report by Reuters Institute and the University of Oxford reveals the perceived impact of AI on journalism. The report highlights that while 74% of survey participants believe Generative AI will enhance efficiency without altering the essence of their work, 21% anticipate a transformative change in workflows and roles within the newsroom.


Code to Embed

Furthermore, AI’s role in combating misinformation and managing large data sets is pivotal for upholding journalistic integrity. This technological progression makes the media industry more resilient and adaptable to digital era challenges. However, it also necessitates vigilance against AI’s potential misuse in media manipulation and misinformation. The growing need for an ethical AI framework in journalism has led to initiatives like Reporters Without Borders and the Partnership on AI developing guidelines. Notably, the ‘Changing Newsrooms 2023’ report indicates that only 16% of media outlets have detailed guidelines for generative AI use, with 35% actively working on them.

Applications of LLMs in the Media Industry:

Generative AI in media is not just a futuristic concept but a present reality, as evidenced by several pioneering applications. For instance, Artifact, a news aggregator launched by Instagram co-founders Kevin Systrom and Mike Krieger, utilises a homegrown Large Language Model to curate personalised content for users. This app exemplifies the innovative use of LLMs in media, offering features like content summarization, voice narration, and even AI-assisted headline editing to counteract clickbait tendencies in journalism.

As of now, Artifact has garnered over 400,000 downloads and is continuously evolving, adding user profiles, commenting features, and link sharing. This growth trajectory highlights the increasing relevance of AI in journalism and media, particularly in terms of personalization and user engagement.

Another notable example is Norwegian Media Group Schibsted‘s utilisation of LLMs to enhance editorial workflows and create new media products (discussed in AP’s Webinar). In 2022, Schibsted embarked on integrating GPT-3 for content summarization, developing an API for their partners to efficiently process and draft external news articles. This initiative underscores the versatility of LLMs in media, catering to diverse journalistic needs from content generation to editing and proofreading.

In conclusion, Generative AI and Large Language Models are rapidly becoming indispensable tools in the media industry. As media houses and publishers continue to experiment with and train their staff on these innovative technologies, the landscape of journalism and media is set for a significant transformation, paving the way for AI-powered platforms tailored to the unique demands of media professionals.