Ask HN: Where does your AI/ML data pipeline hurt the most? (2025)1 points·by phukon·geçen yıl·0 commentsI’m gathering quick, honest feedback from people who wrangle data for models and post-training/ML teams at LLM labs.Share your thoughts/anecdotes on:- Biggest recurring bottleneck (collection, cleaning, labeling, drift, compliance, etc.)- Has RLHF/synthetic data actually cut your need for fresh domain data?- Hard-to-source domains (finance, healthcare, logs, multi-modal, whatever) and why.- Tasks you’d automate first if you could.0 commentsPost comment—
Share your thoughts/anecdotes on:
- Biggest recurring bottleneck (collection, cleaning, labeling, drift, compliance, etc.)
- Has RLHF/synthetic data actually cut your need for fresh domain data?
- Hard-to-source domains (finance, healthcare, logs, multi-modal, whatever) and why.
- Tasks you’d automate first if you could.