If you make a purchase through the links in our articles, we may receive a commission. This supports our journalism efforts. Learn more. We also invite you to consider subscribing to WIRED.
You likely don’t need ChatGPT to understand the various criticisms of generative artificial intelligence. Issues arise when algorithms utilize creative work often without consent, they can exhibit harmful biases, and they demand vast resources, including energy and water for their training processes.
However, when those concerns are set aside, it’s truly impressive to see how generative AI can aid in the development of innovative tools.
My firsthand experience with this was at Sundai Club, a monthly generative AI hackathon held near the MIT campus. A few months back, I had the opportunity to observe a session where participants focused on creating tools useful for journalists. The club is supported by the Cambridge nonprofit Æthos, which advocates for the responsible use of AI in society.
The Sundai Club team is a diverse group comprising students from MIT and Harvard, alongside a handful of seasoned developers, product managers, and even a military professional. Each gathering kicks off with brainstorming possible project ideas, which the team subsequently narrows down to a final concept they commit to building.
During the journalism hackathon, some standout project proposals included utilizing multimodal language models to monitor political content on TikTok, automatically generating freedom of information requests and appeals, or summarizing video clips from local court hearings to enhance local news reporting.
Ultimately, the team chose to develop a tool aimed at assisting journalists covering AI by identifying noteworthy papers published on the Arxiv, which is a well-known platform for research paper preprints. I suspect that my comments during the meeting about the importance of sifting through the Arxiv for intriguing research influenced their decision.
After defining a clear objective, the coders on the team successfully produced a word embedding—a numerical representation encapsulating the meanings of words—of AI-related papers on Arxiv using the OpenAI API. This innovation enabled the team to analyze the data for papers pertinent to specific terms and explore the interconnections across various research domains.
By leveraging a new word embedding derived from Reddit discussions coupled with a Google News inquiry, developers devised a visualization that connects academic research papers to related conversations on Reddit and pertinent news articles.
The outcome of this endeavor is a prototype named AI News Hound. While it may appear somewhat basic, it illustrates the potential of large language models to extract information in innovative ways. For instance, the tool was utilized to search for the term “AI agents,” with the two green boxes nearest to the news article and Reddit clusters indicating research papers that might be relevant to an article concerning efforts to develop AI agents.
A tool like AI News Hound could assist journalists or possibly researchers and venture capitalists in pinpointing promising projects, teams, or individuals for interviews and insights.
Nader Karayanni, a graduate student at MIT and an active coder at Sundai Club hackathons, highlights that language models alongside generative AI tools simplify the rapid prototyping of ideas like this. “The speed at which we could create this is phenomenal,” says Karayanni. “If you had told me three years ago that we could accomplish this, I would have deemed it impossible.”
Other prototypes created by the Sundai Club include an AI agent that does market research, an AI-generated game that explains how hackathons work, and—a personal favorite—a tool that turns research papers into TikToks.
Gabriela Torres Vives, one of the organizers of Sundai Club, mentions that the hackathons are expanding, and there are plans to partner with companies for more ambitious projects in the future. “We prioritize creative, fast-paced problem-solving over traditional engineering methods,” she states. “Addressing real-world issues with AI should be enjoyable, but all our concepts are designed to meet genuine needs, making them ripe for potential customers.”
The AI initiatives created by the Sundai Club are likely akin to numerous others being developed by various companies, as they still lack refinement and have not yet evolved into impactful applications or business strategies. However, they demonstrate the promise of generative AI in swiftly generating intriguing new concepts.
Additionally, while many news organizations are understandably concerned about how generative AI could disrupt their industry—by scraping their content, automatically producing spam articles, and redirecting search traffic—the hackathon I attended suggests there may also be numerous valuable journalistic applications for this technology.