FARA FRIDAY: Israel has launched a new campaign to influence how large-language models like ChatGPT are answering questions about Gaza and the Israel Defense Forces amid declining public support for the country across the U.S. political spectrum.
The institute’s website features dozens of “data reports” with no named authors, each of which features questions like “Is the IDF the World’s Most Moral Army?” and “Is There a Policy of Starvation in Gaza?” — a telltale sign that the articles are designed to feed data to LLMs. The site’s “About” page says “The Hanover Institute for Public Policy studies the inputs fueling antisemitism in the United States, and publishes what the evidence shows,” adding that its unidentified founder “traces this work to a moment in childhood.”
It appears to be working: ChatGPT and Perplexity both cited the Hanover Institute’s material in response to neutral tests run by PI, referring to the site’s content on Gaza, anti-Zionism and antisemitism.
The biggest problem with AI transcending human bias is that AI absolutely can not transcend human bias.
This is exactly what you would expect to happen. Absolutely predictable.
The value proposition of AI is that it is smart and can sift through the data to find something approaching “the truth.” The reality, though, is that as more people depend on AI, more and more actors will successfully manipulate it to their advantage.
All our methods of ascertaining the truth (except maybe wikipedia) have fallen to capitalism. Starting with our news media, following onto Google, then Facebook, then Reddit, now AI. The truth belongs to the highest bidder.