Media Party New York

Answering: How has the space evolved for both AI and hallucination detection. How GPTZero has grown and learned from both writers, journalists, and enterprise customers in adopting AI responsibly and keeping our information ecosystems safe.

In an industry set on rapid information flow and shortened attention spans, generative AI offers a powerful tool to enhance the value of journalism, particularly during the high-stakes environment of an election. By leveraging AI to delve into archives, we provide crucial historical context, draw meaningful connections between past and present events, and set a more relational tone for public discourse that people can make sense of. Moreover, the integration of generative AI with human-centric interactions, social listening, location data, sentiment analysis, and other indicators (economy, employment, health…) for each community, creates a multidimensional view of the election context is specific areas that makes it easy for the user to find familiar ground from where to absorb the information. This synthesis of information allows journalists to uncover patterns, anticipate trends, and provide more comprehensive coverage tailored to specific demographics or regions which also allows for richer, deeper and longer interactions. By harnessing these technological capabilities, we can take election journalism beyond the “we and them” horse-race reporting or some outlandish quote, offering citizens in-depth, contextualised information that they can relate and build upon from, so that they can make an informed decision at the ballot box.

Using the case on Zac Brettler's visual investigation as an example. This talk will delve into how data-driven visual storytelling, how we combined new evidence with intuitive visual storytelling to bring back to life Zac’s case and how a multidisciplinary team was key for the end result. By combining new evidence and data and fact checks with intuitive visual storytelling, this project aimed to take the user step by step on how that night unfolded allowing them to engage with the information on a personal level and where in the final result collaboration was key.

I want to highlight the inequality on Wikipedia. The Wiki community has been working hard to solve for this problem: one Wikiproject at a time.

The adoption of Generative Artificial Intelligence (GenAI) by journalists, researchers, and educators is growing across various disciplines. However, its broader uptake remains limited to tech-savvy populations, leaving marginalized communities in the USA underrepresented despite its potential for meaningful impact. A 2019 Pew Research Center report highlighted that the U.S. had about 208,000 legal Bangladeshi immigrants, with only 55% proficient in English. Similar trends are seen across other immigrant communities, where language barriers limit access to information. Ethnic media plays a crucial role in bridging this gap by providing news in the native languages of these communities, such as Bangla in the USA. Journalists often rely on tools like Google Translate to deliver local news in ethnic media, but these tools frequently fall short in providing contextually accurate translations. Emerging large language models (LLM) and GenAI, like GPT-4, offer the potential to improve these translations and generate content in diverse languages. Bangla AI is a project aimed at transforming the Bengali ethnic media landscape using the power of artificial intelligence. Our goal is to empower US-based Bengali journalists with seamless access to critical news and information in their native language, bridging the gap between mainstream media and ethnic media through innovative AI solutions. Bangla AI's primary objective is to develop an AI-based tool that allows Bengali ethnic media journalists to access vital information in Bengali, eliminating the language barrier they often face when gathering news. By offering a personalized newsfeed based on relevant keywords for Bangladeshi immigrants, this tool will enhance the quality and relevance of their news coverage. With advanced search functionality, Bangla AI empowers journalists to discover stories that resonate specifically with the lives of ethnic community members, ensuring comprehensive and accurate news coverage. To bridge the language gap, Bangla AI's dedicated team is tirelessly working on a robust translation feature that will seamlessly convert news articles from English to Bengali. This breakthrough not only connects mainstream English media with ethnic media but also ensures that crucial information reaches the Bengali-speaking audience, strengthening the role of ethnic media as a trusted source of news. The team is actively developing an AI-powered summarization feature that distills lengthy news articles into concise and meaningful summaries in Bangla. Journalists can now keep abreast of current events without compromising on accuracy or depth of coverage, catering to their audience's evolving reading habits.

The Internet is designed to be hard to kill. It was specifically meant to survive a nuclear apocalypse, after all. So it feels only right that when the Internet is under an existential threat, it's going to be by accident. While the Canadian Bill S-210 is meant to tackle some serious issues online, the Bill's poor writing threatens to kick the entire Canadian Internet offline. And did I mention it could soon become law? In this lighting talk, we will: - Run through how Bill S-210 threatens to break encryption in Canada; - How the broad scope of the Bill would make it impossible for the Canadian Internet to connect with the global Internet; and - What you can do to keep the Internet safe.

Santiago will present on what content provenance and content credentials are, why the CAI exists and what problems the C2PA standard solves, and how anyone can get involved in the movement to build a culture of trust and transparency in digital content in an era of rapid progress in generative AI.

We will show how we’re approaching getting more journalism research into everyday newsroom practice through a product/tech solution. Specifically, we’ll share more about how Modifier, Trusting News and Stylebot are working together to expand newsroom style and standards manuals with research-backed strategies for responsible election coverage. This collaboration is in its early stages, and we want to share challenges and questions that can help us iterate as we go.

I would like to propose a new type of HTML markup that would allow LLMs to track "units of knowledge" as they travel through AI systems so that provenance can be tracked and credited. This will enable publishers to share in the fruits of the coming AI explosion rather than block against it.

While traditional tools like dedicated election website pages, data visualization tools, and fact-checking remain standard, I didn't notice any groundbreaking new technologies being employed by the media. This doesn't necessarily mean innovations weren't present; they just didn't catch my attention. However, I found some non-media tech usage quite interesting: Many social media users turned to generative AI platforms like ChatGPT and the Meta bot for advice on voting. People used these tools to ask questions like "Who should I vote for?" and shared the responses. Others requested summaries of election manifestos or comparisons between different party platforms. This highlights the importance of ensuring AI training data is current and accurate, a role where I believe media could be pivotal. The electoral commission attempted to use voter management devices to prevent double voting, a practice first introduced in the 2021 local elections. Unfortunately, this did not go smoothly—the devices malfunctioned, and many election officials seemed inadequately trained, which led to unprecedented questioning of the election results. Also noteworthy was the continued use of bulk SMS technology by political parties and the electoral commission, as in previous years.

The goal will will be to reflect on the current AI and digital tool landscape for independent media outlets, globally. The speaker will suggest that data-driven insights put power back in to the hands of media outlets, so that they can make informed decisions about their business success. The speaker will speak about some of the top-line successes of the Media Viability Accelerator and the place that it holds as a digital tool - built for media, by media.

Imagine using software to transcribe your interview into text, unaware that the privacy policy you agreed to allows the company to use your unfinished work to train their AI model. This could be alarming for an investigative journalist! In this lightning talk, I will share findings from privacy policies of commonly used journalism tools, including those for online conferences, transcribing, social media, and more. I'll highlight which tools are safe and which ones raise concerns based on a thorough analysis of their privacy policies, so you don’t have to sift through them yourself. These findings are part of our PrivaSee project, which won a grant from the Columbia Venture Challenge. We analyze the latest privacy policies from app stores, identifying patterns in privacy and AI. We want to use this project to first protect journalists' privacy and share our findings with you. Here’s the structure of the talk: 1. The Problem: Companies are leveraging user data to train their AI models, embedding this practice within their privacy policies. 2. Why It Matters: Investigative journalists need to protect their unpublished work, and newsrooms often sell content to large companies for revenue. 3. Key Findings: An overview of commonly used tools for journalists, identifying which are safe and which may pose risks. 4. Solutions: The best strategies we've identified for keeping your content safe while working efficiently. 5. Methodology and Limitations: How we conducted this project and how it will continuously monitor and update privacy policies. 6. Project and Acknowledgments: An introduction to our PrivaSee project, a simple demonstration of the prototype, and thanks to the Brown Institute for Media Innovation.

My 5-minute lightning talk will touch upon a red-teaming exercise we did ahead of elections in India to test the limits of LLMs, and how easily they can be manipulated by persons or organisations with malicious intent to harm your news brand. Our experiments involved changing the narrative of a news story, changing the bias in a news story, changing the sentiment of a news story, cloning a news site - all to spread misinformation.

At Media Party (Buenos Aires) last year, ICFJ launched the first of four global workshops to ask diverse media stakeholders: "What can we do to create a more sustainable system for independent journalism?" Every region had particular obstacles but there were some common needs and solutions. This talk presents preliminary findings from these sessions and a blueprint for possible solutions.

In this lightning talk, we will briefly discuss the Amazônia Vox project, specifically focusing on the media literacy work being done, with 15-16 year-old students reviewing our news articles in advance as a way to combat misinformation and engage new audiences. The Amazônia Vox project is supported by the ICFJ through the Knight Fellow program. It is developing a knowledge source bank of the Amazon, a network of communication freelancers from the Amazon, and content based on Solutions Journalism. The news reports are reviewed in advance by teenage students, as part of media literacy work and the development of new audiences.