You almost don’t need ChatGPT to generate a list of reasons why generative AI often isn’t amazing. The way algorithms are fed innovative work, often without permission, carry nasty biases with them, and require huge amounts of energy and water to train are grave problems.
But all that aside, it’s remarkable how powerful generative AI can be in prototyping potentially useful novel tools.
I had the opportunity to see it for myself when I visited this place Sundai Cluba generative artificial intelligence hackathon held one Sunday each month near the MIT campus. A few months ago, the group kindly agreed to let me participate in the discussion and decided to spend this session exploring tools that might be useful to journalists. The club is supported by a Cambridge non-profit organization called Ethos that promotes the socially responsible apply of artificial intelligence.
The Sundai Club crew includes MIT and Harvard students, several professional developers and product managers, and even one person who works for the military. Each event begins with a brainstorming session on possible designs, which the group then whittles down to the final option they are actually trying to build.
Notable journalism hackathon proposals included using multimodal language models to track political posts on TikTok, automatically generating freedom of information requests and appeals, or summarizing video clips of local court hearings to support cover local news.
Finally, the group decided to build a tool this would support AI reporters identify potentially engaging stories to which they were sent Arxiva popular server for reprinting scientific articles. My presence here probably convinced them, considering I mentioned at the meeting that scouring Arxiv for engaging research was a high priority for me.
After coming up with a goal, the coders on the team could create word embedding– mathematical representation of words and their meanings – Arxiv AI articles using the OpenAI API. This enabled the data to be analyzed to find articles relevant to a given term and to explore connections between different areas of research.
Using other word embeddings in Reddit threads and Google News searches, developers created a visualization showing research articles alongside Reddit discussions and relevant news reports.
The resulting prototype, the so-called AI News Houndis general, but shows how huge language models can support extract information in novel and engaging ways. Here’s a screenshot of the tool used to search for “AI agents.” The two green squares closest to the news article and the Reddit cluster represent research articles that could potentially be included in an article about efforts to build AI agents.
Compliments from Club Sundai.