<iframe src="https://www.linkedin.com/embed/feed/update/urn:li:share:7470136878755889152?collapsed=1" height="624" width="504" frameborder="0" allowfullscreen="" title="Embedded post"></iframe> ^post > Knowledge about yourself is your biggest asset in the age of [[Artificial intelligence]]. The advantage is being able to move your own data quickly into new [[LLM|AI tools]]. ^karwatka-self-knowledge-asset Still, there is a catch: he uses [[Obsidian]] as a bucket and leaves the structure to the [[LLM]] (the [[Karpathy method]]). As the pile grows, the [[Retrieval|semantic search]] has more to scan - slower, and the results get worse. The five-minute context load confirms it: pushing the *entire* [[Personal knowledge base|base]] into an [[LLM|model]]'s context on every switch is not optimal - that is what [[Retrieval|semantic search]] is for. And if 5,000 files genuinely fit in one context, either the files are tiny or something is off about how much each one actually says.