• [Paper] Synthetic customers for purchase intent prediction. Rather than generating a Likert score for intent (and there would be many relevant techniques there, such as sampling methods), generate a freeform text response and match it with reference responses by emb sim to produce Likert scores. The result is “high human test–retest reliability while maintaining realistic response distributions”, with interpretability advantages.

  • [Paper] Think RAG but instead of retrieving context, retrieve reasoning instructions.

  • [Leaderboard] Alpha Arena - Trading as a benchmark.

  • [Blog] Ion Stoica’s story.

  • [Paper] Verbalized sampling - Prompting trick to get more diversity from aligned LLMs.

  • [Paper] Humor understanding abilities seem to be strongly related to STEM training & reasoning.

  • [Blog] Harrison Chase’s view on the paradox of graphical workflow builders. Simpler workflows are easier to build with just no-code agents. More complicated workflows seem to necessitate code.

  • [Paper] There is still hope for under-represented programming languages.

  • [Tweet] Constant mismatch between the ideal and the reality on content recommendation stacks.

Have a great week!

Updated:

Comments

You May Also Enjoy

The week ahead: 20251201

[Blog] HOPE. A model as a set of nested sub-optimizers, each with its own context flow and hyperparams. Relevant paper.