Supercharging Journalism with AI: A workshop for the US Office of Foreign Broadcasting

I love doing workshops. It’s the hands-on approach and the interaction with the audience that makes it so fun. There is nothing like discussing a topic with a group of professionals, journalists in this case, and to be able to get their perspective and experience about a topic that’s so new that you can still find pieces of the packaging tape on it, as is the case forArtificial Intelligence as being used by the masses.

I had the honor of teaching a 3-part workshop for the US Office of Foreign Broadcasting on the advancement of AI tools and their uses for journalists.The workshop was in Spanish, which was hard enough because although Spanish is my native language, I’ve been functioning in English in the tech space fort almost 30 years so I had to learn the terminology in Spanish and the tongue-twisting gymnastics that some of these tech terms are when said in the language of Cervantes.

The Role of AI in Modern Journalism

There are many roles that AI can take in modern journalism that can assist a capable journalist to become more efficient and produce more in-depth content and connect seemingly distant concepts more efficiently than ever before. It was mind blowing to explore many potential applications of these tools for journalists that are diverse and interesting but at the same time they are changing the paradigm of what it is to analyze and report the facts. 

Large Language models such as Chat GPT have so many potential uses for journalists that the book is being written in real-time but many journalists and media organizations are acting in an abundance of caution, and rightfully so, after the recent New York Times lawsuit against Open AI where they claim that many of their works are being used by the Open AI language model and the result of the text output for Chat GPT is not transformed enough to constitute fair use.

At the heart of the matter is the use of the Common Crawl dataset which spans over 250billion pages and it’s constantly being refreshed by adding 3 to 5 billion new pages monthly. A big part of those pages originates from The New York Times.

I’m not a copyright attorney and don’t know who’s going to win the legal battle but what I can tell you is that the cloud of litigation will affect adoption of many of these systems in newsrooms for the foreseeable future.

LLMs in a nutshell 

These large language models are incredibly complex but at the core, they are basically a jumble of billions of texts that are analyzed by the model and the outputs are created based on probability and statistical models of what’s likely to complete the idea or sentence. Of course, I’m trying to simplify how these neural networks encode and decode information and the rigorous training process but the ability to interact with natural language prompts and produce human-like results are the game changer and what’s blowing everyone’s minds.

Prompting and interaction with Large Language Models

For the first time, people are accessing a tool where they interact with natural language asking it to produce results and that’s basically what prompts are, the instructions that you provide the LLMs in natural language to ask it to produce a result. The refinement of this process to produce a better-quality response is what is called Prompt Engineering, which is basically the skill to guide the output of these generative AIs to produce a better-quality outcome. In a nutshell, better prompts generally equal better results.

Practical applications for AI and Large Language Models for Journalists

Putting the content conundrum aside, journalists today can use LLMs such as Chat GPT to supercharge their work and there are so many potential applications that it’s hard to wrap your head around it. It is imperative to incorporate AI into journalists’ workflow not only for efficiency’s sake but also to provide new perspectives and delve into new work models that are transforming journalism as we speak. Here are some that we discussed in the workshop.

Brainstorming

AI is incredibly efficient in generating ideas on pretty much any topic and can be used effectively to generate infinite approaches and variations on an idea, exploring it from different angles or even listing many ideas or approaches on a topic. When brainstorming, you want the ability to generate ideas and iterations of those ideas quickly and given the iterative nature of these models, beware of what you wish for because you might hit analysis paralysis sooner than you think.

Data Analysis

This is a part where they really excel by loading a large amount of data in a document or database in the LLMs, you can start questioning the data and ask it in natural language to create reports, charts, correlations, you name it. Journalists are now more capable than ever to find correlations between distant datapoints and to interrogate the data to uncover insights hidden in plain sight.

This capability is ushering a completely new era in data-driven journalism and the results that will be produced will transform reporting forever. The barrier to entry for data-driven journalism has historically been high given the requirements of understanding how to handle large amounts of data and extracting information but, with the assistance ofLLMs and the ability to interact in natural language, we are going to see the barrier to producing data-rich journalism lowered even further.

Summarizing

Given a large document, journalists can create executive summaries of the work by just inputting the actual text, uploading the documents and asking the LLM to write a summary of it.

Graphs, illustrations, charts, diagrams, videos, etc.

In the workshop, we did a few exercises on how the ability to create graphics from data is a game changer for journalists. You no longer have to be an excel guru to come up with advanced charts and see how easy it is to ask the LLMs to produce high-quality charts based on data. The part I personally loved was getting ourselves out of the way and asking the LLM which types of charts could be created with the data we uploaded and watch it provide us a menu of options that was truly mind-blowing. 

When it comes to illustrations, we explored some of the capabilities of DALL-E and Mid journey in creating images from text and how the process worked. The ability of these models to generate incredibly realistic graphics from natural language descriptions is truly transformative and will augment journalism in ways we can’t even comprehend at this point.

We also discussed the newly unveiled Sora, the new text-to-video model and how it will revolutionize video content creation as we know it.

Transforming Social Media Interactions

A big part of journalists jobs these days is to share and promote their work in social media. This can be a tedious and tiresome process that can be made more efficient by using AI. We discussed many potential social media applications and demonstrated practical ways to share content, identify trends, create effective titles and descriptions for tweets, videos and how to simplify the process of sharing content on social media.

Translations

Although there are many translation tools already, LLMs are quite good at interacting in several languages and its translation capabilities shouldn’t be ignored.

AI Content Detectors / Plagiarism Checkers

We discussed and demonstrated how some AI content detectors and plagiarism checkers worked and how they can check thousands of lines of text in just seconds. These systems are incredibly sophisticated in checking potential copyright issues or potential plagiarism that can diminish trust in the organization and potentially result in litigation.

Ethical considerations

We had a lively discussion of ethical considerations when it comes to using Large Language Models and Generative AI in general and we addressed the elephants in the room of misinformation and how to identify potential conflicting content as well as how to address potential biases and how to potentially navigate these risks.

Creation of Custom GPTs

Custom GPTs is a recent development where you can create a specialized instance with a specific set of instructions provided in natural language that can be catered to be laser-focused for a specific purpose. This approach can save a great deal of time by avoiding repetitive prompting and generating better quality results overall for a specific purpose.  We also discussed how to access the already vast library of custom GPTs with a wide set of capabilities that are relevant to journalists, from copy-editing, fact-checking, transcriptions, video summaries, tweet generators and more.

Parting thoughts

It was a humbling experience to conduct this workshop for a group of experienced journalists, to exchange ideas and to get their perspectives about these advancements and the challenges posed by the speed and increased frequency of how these transformational changes are happening.Artificial Intelligence is now a part of journalism, and its efficient, effective, and ethical use is in every journalist’s best interest.