Pangram repeats text so earlier passages can use later context, a user says
A paper linked by the user reports that repeating prompts improves performance for Gemini, GPT, Claude and Deepseek when they aren't using reasoning, without increasing generated tokens or latency.
TLDR
A user highlights a trick they say Pangram uses: repeating text so the early parts of the second copy can draw on the later parts of the first. They link to the model card for Pangram 4, which Pangram describes as an AI detection model, alongside a paper on arXiv. The paper reports that prompt repetition improves performance for Gemini, GPT, Claude and Deepseek when not using reasoning, without increasing generated tokens or latency.
Combined views
—
1 Source, first seen ago
