Mr.D

Just call me Mr.D
Supreme Leader of Degenestan
Old job Title: E-beggar @monad
New job Title: Digital AI Artist

[Mr.D @mr0xed]

@theafterseed @RobinhoodApp EARLY!

[Mr.D @mr0xed]

its funny thou i can cantinue in code my code session but whenever i try to send any message on claude or cowork it says usage limit reached lol

[Mr.D @mr0xed]

I've barely used half my quota. Did Claude forget how to do math? Getting rate limited while clearly having usage left is ridiculous. Definitely time to cancel. Sorry guys but you are losing the AI War 👋 @claudeai @AnthropicAI

[Mr.D @mr0xed]

@thsottiaux is there a limit on luna ? I have usage left but getting this error when i try luna, all other models are working fyi

[Mr.D @mr0xed]

@198iki Herkeseborsa iş yatırımın onu deneyebilirsşn.Ayeıca enpara da saklama komisyohu etc kaldırdı

[Elon Musk @elonmusk]

Wow, the situation in Spain looks crazy!

[Mr.D @mr0xed]

The part that costs the most is that every new session on the same codebase starts the learning curve over. Same explanations, same constraints, same architecture. Compaction does the same thing mid-session.

The real cost isn't the wasted time. It's that the model picks a different approach this week than it did last week for the same problem, or quietly contradicts a constraint we already settled. I don't always catch it, because a forgotten constraint doesn't announce itself. It shows up as a decision that looks reasonable and isn't. Over a long project, that is what architectural drift actually is.

[Mr.D @mr0xed]

Three things I keep needing in Codex.

Project memory built on an index and a matrix, not a summary.

Karpathy's LLM Wiki pattern already showed this works. You point an agent at raw sources, it incrementally compiles a linked Markdown wiki, and you stop re-reading the raw material every time. I've been running that pattern and it holds up. What I want is the same idea native to Codex, with the raw material being the whole project environment. Not only code. Financial models, architecture documents, strategy notes. That is what a real project actually consists of.

The index maps each file or document to its functional role: what it is responsible for, not what it contains. The matrix maps where each decision, constraint or variable is actually expressed across the project, and separates the source of truth from every place that mirrors or derives from it. That second part is what makes a change safe. When something moves, the matrix shows everything that has to move with it, including the copies that only look correct because nothing tests them.

This memory should stay separate from personal or profile memory. It should record why a decision was made and not just what it was, and connect linked projects so a decision recorded in one is visible from the others. It should survive compaction. It should also be purgeable: when a project ends I want to wipe its memory completely, so stale context doesn't leak into the next thing I work on.

Most of my back-and-forth with these tools isn't model weakness. It's context that already exists somewhere the model can't see.

An opt-in deep inspection mode.

Not the default. A mode I turn on deliberately, where the model stops using grep, AST queries and batched reads, and instead reads every file in the declared set completely and in sequence, along with its full dependency tree. It should validate structural and logical coupling before applying any edit, then report coverage at the end: which files were read fully, which were sampled, which were never opened.

This matters because grep and batch reads fail silently. A partial read still produces a confident summary. Confidence doesn't drop when coverage does, so nothing signals that something was missed. You find out later, from the breakage.

Connect the two, and let me watch the reasoning.

The deep inspection pass is what should build the index and the matrix in the first place. Run it once to map a project properly, store the result as project memory, then re-inspect only when things drift.

I also want to see the reasoning transcript as it streams. When the model takes a wrong turn mid-thought I can stop it there, before an answer gets built on top of it. A wrong assumption rarely appears in the final output. It just quietly shapes it.

[Mr.D @mr0xed]

@thsottiaux How much should i wait my codex still have %0 usage ??

[Mr.D @mr0xed]

@yapayzekahocasi Eğer gece bir projede ard arda 2-3 kere fikir değişikliği yaşadıysan bunu geceye ve yorgunluğa vuruyor. Akşam saatlerinde hazırladığım bir mimari için 2-3 kere fikir değişikliği yaşayıp yapalım dediğim şeyden vazgeçtim veya başka bir şeye döndürdüm ondan beri bana da aynı

[Mr.D @mr0xed]

@thsottiaux It says i have %87 usage left but also says i reached ?

Is there a problem with codex ? Btw i am pretty sure i have already used all my usage

[SOLOFÚTBOLMEMES⚽ @SoloFutbolMemes]

Resumen de Argentina en el mundial

[Mr.D @mr0xed]

@nikitabier @ForgiatoBlow47 Tham nam @MrBeast too almost every content is scheme or bait

[Mr.D @mr0xed]

@thsottiaux thanks to you guys nothing. APP became retarted after update!!!

[Mr.D @mr0xed]

@thsottiaux Giving reset without user control lol you are no diffetent from them. What happened to banking resets?

[Mr.D @mr0xed]

@enorton1903 Siparişi teslim alıp almadım diyen insan dolu memleket kamera olmazsa kanıtlanmıyor da qr veya kod ile teslimata geçilmesi gerekli

[Mr.D @mr0xed]

@thsottiaux Hellon there, i get the reset but it didnt banked it used automatically. I remember this because i had %90 usage left but it %100ed.

[Mr.D @mr0xed]

@elonmusk Grok OAuth on Openclaw is a massive W, but the 6h expiry is way too short. Constantly renewing is a pain. Any chance we could get 3-5 days? Or is it capped at 6h just because I'm on standard X Premium and not Super Grok? @grok

[Mr.D @mr0xed]

@kaayaa1923 @tarih101YT Yutubır

[Troll Football @TrollFootball]

After nine decades, England has finally been Germanised.