Today's brief
AI Research6 hours ago · 4 min
Retrieval beats longer context on cost, again
A replication study shows retrieval pipelines match million-token contexts at a fraction of the spend.
Depth
Explain simply
HeadlineExplain simplyExpert
You can either hand the model an entire library, or hand it the three right pages. The three pages keep winning, because the model gets distracted by the rest.
What this means for you
You were about to pay for the long-context tier — this suggests waiting a quarter.