Hacking Generative AI for Fun and Profit

You hardly request ChatGPT to make a database of reasons wherefore generative artificial intelligence is often little than awesome. The mode algorithms are fed originative work often without permission, harbor nasty biases, and require immense amounts of vigor and water for grooming are each superior issues.

Putting each that speech for a moment, though, it is singular however almighty generative AI tin beryllium for prototyping perchance utile caller tools.

I got to witnesser this firsthand by visiting Sundai Club, a generative AI hackathon that takes spot 1 Sunday each period adjacent the MIT campus. A fewer months ago, the radical kindly agreed to fto maine beryllium successful and chose to walk that league exploring tools that mightiness beryllium utile to journalists. The nine is backed by a Cambridge nonprofit called Æthos that promotes socially liable usage of AI.

The Sundai Club unit includes students from MIT and Harvard, a fewer nonrecreational developers and merchandise managers, and adjacent 1 idiosyncratic who works for the military. Each lawsuit starts with a brainstorm of imaginable projects that the radical past whittles down to a last enactment that they really effort to build.

Notable pitches from the journalism hackathon included utilizing multimodal connection models to way governmental posts connected TikTok, to auto-generate state of accusation requests and appeals, oregon to summarize video clips of section tribunal hearings to assistance with section quality coverage.

In the end, the radical decided to physique a tool that would assistance reporters covering AI place perchance absorbing papers posted to the Arxiv, a fashionable server for probe insubstantial preprints. It’s apt my beingness swayed them here, fixed that I mentioned astatine the gathering that scouring the Arxiv for absorbing probe was a precocious precedence for me.

After coming up with a goal, coders connected the squad were capable to make a connection embedding—a mathematical practice of words and their meanings—of Arxiv AI papers utilizing the OpenAI API. This made it imaginable to analyse the information to find papers applicable to a peculiar term, and to research relationships betwixt antithetic areas of research.

Using different connection embedding of Reddit threads arsenic good arsenic a Google News search, the coders created a visualization that shows probe papers on with Reddit discussions and applicable quality reports.

The resulting prototype, called AI News Hound, is rough-and-ready, but it shows however ample connection models tin assistance excavation accusation successful absorbing caller ways. Here’s a screenshot of the instrumentality being utilized to hunt for the word “AI agents.” The 2 greenish squares closest to the quality nonfiction and Reddit clusters correspond probe papers that could perchance beryllium included successful an nonfiction connected efforts to physique AI agents.