Carlos Rivas

@carlosrivas.bsky.social

Sociologist - Venezuelan | Exploring social & cultural themes through digital data sources | También en español.

"But the point here is that officialese became his language because he was genuinely incapable of uttering a single sentence that was not a cliché." Hannah Arendt, Eichmann in Jerusalem: A Report on the Banality of Evil (1963).

Concepts such as democracy, sovereignty, and self-determination no longer carry the same weight in everyday discourse. It is deeply concerning that they have become empty signifiers today, especially considering how much was sacrificed to achieve them.

Electricity has been restored, though it might only be temporary. Slowly, the neighbors and I are trying to return to our routines. We've been meeting up to provide each other with support.

Once again, we are without electricity. The efforts to restore service have been complicated because the electrical substations were the first to be bombed. When they plunged everything into darkness that night, the horror began.

A medida que pasan las horas, empiezo a procesar la magnitud de lo ocurrido. Mi mirada se detiene en las prácticas sociales de justificación: en cómo se articulan discursos para gestionar profundas contradicciones; en esa negociación social que pretende, sin éxito, preservar la coherencia del yo

The first challenge is: how to narrate an intense lived experience with objectivity? It is, as Jorge Semprún suggests, a perspective tinted by trauma that must reach a certain temporal distance to find some semblance of truth.

40 hours after the bombings here in Caracas, my area is still without power due to the destruction of the electrical stations. The mobile signal is intermittent. Physically we are fine, but emotionally shaken.

Evaluating topic models (and document clustering methods) is hard. In fact, since our paper critiquing standard evaluation practices four years ago, there hasn't been a good replacement metric That ends today (we hope)! Our new ACL paper introduces an LLM-based evaluation protocol 🧵

Screenshot of first page of paper. It is here: https://arxiv.org/pdf/2507.00828

Abstract: Topic model and document-clustering evaluations either use automated metrics that align poorly with human preferences or require expert labels that are intractable to scale. We design a scalable human evaluation protocol and a corresponding automated approximation that reflect practitioners' real-world usage of models. Annotators -- or an LLM-based proxy -- review text items assigned to a topic or cluster, infer a category for the group, then apply that category to other documents. Using this protocol, we collect extensive crowdworker annotations of outputs from a diverse set of topic models on two datasets. We then use these annotations to validate automated proxies, finding that the best LLM proxies are statistically indistinguishable from a human annotator and can therefore serve as a reasonable substitute in automated evaluations

"If desire is repressed, it is because every position of desire, no matter how small, is capable of calling into question the established order of a society..." 𝘋𝘦𝘭𝘦𝘶𝘻𝘦 & 𝘎𝘶𝘢𝘵𝘵𝘢𝘳𝘪. 𝘈𝘯𝘵𝘪-𝘖𝘦𝘥𝘪𝘱𝘶𝘴: 𝘊𝘢𝘱𝘪𝘵𝘢𝘭𝘪𝘴𝘮 𝘢𝘯𝘥 𝘴𝘤𝘩𝘪𝘻𝘰𝘱𝘩𝘳𝘦𝘯𝘪𝘢.

"The everlasting and exclusive coming-to-be, the impermanence of everything actual, which constantly acts and comes-to-be but never is, as Heraclitus teaches it, is a terrible, paralyzing thought." Friedrich Nietzsche on 𝘗𝘩𝘪𝘭𝘰𝘴𝘰𝘱𝘩𝘺 𝘪𝘯 𝘵𝘩𝘦 𝘛𝘳𝘢𝘨𝘪𝘤 𝘈𝘨𝘦 𝘰𝘧 𝘵𝘩𝘦 𝘎𝘳𝘦𝘦𝘬𝘴.