The AI trend no one is talking about.
We are constantly being told that improvements in AI capabilities are exponential and we have certainy seen immense advances in just the space of a few months, but there is an exception: Context Windows
Context Windows have been remarkably stagnant (roughly 1M tokens on frontier models) for the past year and there's a good reason for this. The computational cost of attention grows quadratically with context length.
↳ Doubling the size of the context windows means quadrupling the attention computation.
To be fair, AI providers have found ways to compensate: context compression, better retrieval-augmented generation (RAG), task delegation to sub-agents, intermediate planning documents, and simply better models through training and reinforcement learning.
The underlying issue however remains
↳ As model intelligence becomes less of a constraint, context quality becomes more important.
➡️ Context quality matters
It can make the difference between a good AI output and an exceptional one.
As model intelligence continues to advance, context management for AIs is becoming an increasingly important aspect of work.
Two specific challenges stand out:⠀
🔹 Context dilution: Too much information, too little signal.
🔹 Context pollution: Incorrect, outdated, or irrelevant information influencing the model.
The AI labs are working on making models smarter, but every user needs to work on making sure they have the right information to work with.
How is your organization tackling these issues?
I'll be sharing some ideas over the next few weeks, but feel free to DM me if you'd like to discuss.