You often hear people like Sam Altman or Dario use words like "scaling laws" "AGI"... If you don't understand the field of AI, you probably believe that BS... So here are some research papers on the importance of data quality that'll give you perspective on the current state of AI:
* Muennighoff et al., “Scaling Data-Constrained Language Models” (2023, updated 2024):
3× more high-quality data beat 10× more parameters on the same total compute budget.
* Ivgi et al., “Data Engineering for Scaling Language Models” (2024):
Carefully curated 1 trillion tokens beat 15 trillion unfiltered tokens by a huge margin.
* DeepSeek-V3 technical report (2025):
They spent more engineer-time on data cleaning, clustering, and synthetic data than on cluster bring-up, and that’s why their 400B model beats much larger frontier models