Home › Hard problems › Inside the model

Inside the model: the core of LLMs

Using long context, not just fitting it

Solo: strongOne person can run a real test of this in about 8 weeks.

A MILLION TOKENS FIT INhow well it noticesABCstart: noticedend: noticedthe middle sagsthe answer needs A + B + C together: missed
A long document fits in the window, but facts in the middle get missed, and facts spread far apart are not combined.

Why it is still open

Models accept a million tokens but miss facts in the middle and can't combine facts spread far apart.

An experiment one person can run in 8 weeks

Build a test where the answer needs 3 facts from different parts of a 200-page document. Compare long context against retrieval against a memory agent.

Verdict: Solo: strong

Cheap, measurable, and directly useful to buyers.

The solo test behind the verdict

  1. Can it be measured? There is a number that shows you did better, such as an error rate, a catch rate or a cost per answer.
  2. Is it cheap to try? It runs on a laptop, free open models and open data, or under about ₹50,000 of rented GPU time.
  3. Is there a public yardstick? A benchmark or dataset others use, so your result can be compared and believed.
  4. Would a small win get noticed? A lab, a funder or a customer would care about a 10% improvement, not only a full solution.

Four yeses means Solo: strong. Two or three means Solo: possible. Fewer means Follow, don't build.

PreviousContinual learning: learn after deployment without forgettingNextPersonal memory across months of conversations

All 79 hard problems · Working on this, or know of a result we missed? basant@cleandataengineering.ai