Post-event: "Is This Thing On?" Large Language Models and the Cultural Risks of Bad Jokes

On June 25th, the Digital Media Research Centre and the GenAI Lab hosted scholars Prof. Aram Sinnreich and A/Prof Victoria Simon for an engaging seminar titled “Is This Thing On? Large Language Models (LLMs) and the Cultural Risks of Bad Jokes.” Organized and chaired by Dr. Ariadna Matamoros-Fernández, whose DECRA project examines the impact of harmful humour on women’s wellbeing online, the seminar delved into the intersection of Jewish epistemology and the affordances of LLMs, exploring the nuances of cultural context in joke-telling.

Scholars Victoria (left) and Aram (right) presenting at the event.

Aram and Victoria discussed the importance of shared epistemology, self-deprecation, and the nuances of what is unsaid in Jewish humor. Jewish jokes, they noted, often rely on a shared cultural understanding between the teller and the audience. For instance, jokes about personal experiences or historical contexts resonate deeply within the community, but might fall flat or even be misinterpreted by outsiders.

Scholars Victoria Simon and Kevin Witzenberger in discussion at the event.

A critical point that follows from this is the fundamental difference in the epistemologies of Jewish humor, and of LLMs. For instance, Jewish humor often starts with the ‘second meaning’ of a word or phrase, whereas LLMs tend to regress to stereotypical representations due to their positivist, extractive nature. As a result, (they argue) LLMs are fundamentally incapable of telling Jewish jokes.

Several examples illustrated this challenge. For instance, when prompted about minority communities, LLMs frequently default to positive stereotypes, such as joy, pride, and empowerment, rather than capturing the full spectrum of cultural nuances. This tendency underscores the limitations of LLMs in representing culturally rich and context-dependent forms of humor.

GenAI Lab and DMRC members discuss the seminar around a table.

The speakers also discussed unique methodological practices drawn from Jewish cultures of knowledge production, including futzing (hands-on black-box experimentation with no fixed goal), and kibbitzing (which provokes deeper understanding through playful, informal dialogue). These methods highlighted the communal and iterative nature of knowledge-building in Jewish culture, contrasting sharply with the formal, structured approach of LLMs.

In addition to being a lot of fun, the seminar underscored the importance of understanding the cultural risks and limitations of LLMs in context-dependent tasks such as joke-telling.