This is your periodic reminder to use new chats
This also means keeping your pages/components/files shorter in length and extracting out few things.
E.g. Extract 50 lines of header UI code in diff component and make actual layout file/component lean for AI to read.
Guys, "lost in the middle" context is true.
I experienced or noticed it first time.
this is why you have 8 codex clis running separately, you maximize effective functional non-rotted context even if you blow your usage
it just takes planning to fine tune the orchestration
no token is wasted when every token is in distribution and/or grounded by cross checking
Una guía para entender un poco mejor todo esto aca:
x.com/mattpocockuk/status/19…
arxiv.org/pdf/2510.13928
anthropic.com/engineering/ef…
Una guía para entender un poco mejor todo esto aca:
x.com/mattpocockuk/status/19…
arxiv.org/pdf/2510.13928
anthropic.com/engineering/ef…
You all think they give you the God tool... they just give you toilet paper. The rest on internet is hype. AI for now is just knowledge, that was internet until now, but no one know how to ask.
useful and easy to understand!
The "Context Window"
This is one of the most important things you need to understand in AI to be successful!
Huge shoutout to @mattpocockuk for making it easy to understand, check out his course if you want to learn more and how to use AI SDK, 2 days left 50% off
If you have worked with AI for coding, then you must know about the context window error that pops up after using a few too many questions or pasting a very large code file.
Yes, this issue has tormented us very frequently.
This video is the solution to that problem. The TS wizard @mattpocockuk explains how context window works for LLMs and how you could use measures to reduce the effects and issues like "Lost in the middle" problems.
Please do give it a watch. This will help you code with LLMs in a better way.
This is an awesome explainer video.
That's why I always use @GeminiApp. It literally has a huge context window. No matter how large the chat becomes, it always understands and provides exactly what's needed.
PS: It can be very stubborn sometimes.
For all vibe coders, creators and LLM users:
Do you feel like your AI understands you less the more you talk to it?
This is a great post from Matt on how to learn the idea of context windows when interacting with AI tools.
Wow this was SUPER informative!👏👏
Great video
No this makes sense; I was wondering why the AI get 'tired' after some hard work.
You have skill issues when it comes to LLM context. Watch this to help yourself