HeyMeng (@itsheymeng) / X

Pinned

I've been going through @karpathy's "Deep Dive into LLMs like ChatGPT" and noticed a small failure mode in my own self-study loop. When a study session gets interrupted, the source and notes are still there, but the plan often doesn't restart itself. So I tested a tiny tutor

00:00

Anthropic found a small internal workspace in Claude whose contents can be reported, controlled, and used in flexible reasoning. The result I keep coming back to is the difference between the base model and its post-trained version. The J-space already exists in the base model.

we used to write docs so humans could operate software. now we write docs so software can operate the world

HeyMeng

@itsheymeng

Apr 29

ai capability is in 2026. public perception is in 2023. the indie market is the time machine that connects them

HeyMeng

@itsheymeng

Apr 27

used to come back from working with ai tired. now i come back unsure whose ideas were mine. not bad. just new