Claude code is 98% not AI

7 views
Skip to first unread message

Ahmed Halil Oğuzhan Gül

unread,
Apr 28, 2026, 4:01:43 AMApr 28
to yz-ve...@googlegroups.com
Bu yt short videosunda anlatıldığına göre en başarılı LLM'lerden Claude'un kodunun %98'i AI değil. Aslında modelleri tamamen otonom yapmaya çalışmanın başarıyı arttıran esas faktör olmadığı, AI'ın etrafında örülen ve onunla birlikte çalışan "altyapı"nın önemi ile alakalı bir görü sunuyor. Ne dersiniz?

https://www.youtube.com/shorts/18JD2_bUK34


Akif Eyler

unread,
Apr 30, 2026, 3:54:11 AMApr 30
to yz-ve...@googlegroups.com
On Tue, Apr 28, 2026  Ahmed Halil Oğuzhan Gül
Bu yt short videosunda anlatıldığına göre en başarılı LLM'lerden Claude'un kodunun %98'i AI değil. ... Ne dersiniz?
Bu doğru bir kıyaslama değil. LLM modelinin yüklü olduğu mekanizma Claude kodundan milyon kere büyük olabilir. YZ çiplerini ihmal ederek sadece bu koda bakmak adil görünmedi.

"AI'ın etrafında örülen ve onunla birlikte çalışan altyapı"yı anlamıyorum, bilenlere havale ediyorum.

__Akif Eyler__

Gürsu Gülcü

unread,
May 1, 2026, 5:21:20 PMMay 1
to yz-ve...@googlegroups.com, Ayla Gülcü
Claude'u değişik LLM'ler ile entegre etme imkanı bulunduğundan ve bu LLM'lerin bazılarının model parametrelerin toplam boyutları Claude'u meydana getiren kodların toplam dosya büyüklüklerinden defalarca büyük olduğundan (onlarca hatta yüzlerce GByte), neyin oranını incelediğimize makaleden bakalım:

11.1 Design Philosophy (sayfa 27)
The values and design principles introduced in Section 2 predict an architecture that invests in operational infrastructure rather than decision scaffolding. The implementation confirms this: the architecture documented in Sections 3 to 9 is overwhelmingly deterministic infrastructure (permission gates, tool routing, context management, recovery logic), with the LLM invoked as a stateless completion endpoint. An estimated 1.6%
of the codebase constitutes decision logic, the remaining 98.4% is the operational harness. This ratio is not accidental.

The design principles documented in Section 2.2 underpin this approach: the harness creates conditions under which the model can decide well, rather than constraining its choices. This design runs counter to the dominant pattern in agent engineering, where frameworks such as LangGraph route model outputs through explicit graph nodes with typed edges, and systems like Devin pair multi-step planners with heavy operational infrastructure. Claude Code instead gives the model maximum decision latitude within a rich operational harness. The engineering complexity exists not to constrain the model’s decisions but to enable them. This layered architecture, where the model reasons and the harness enforces, raises the question of whether agentic coding tools are converging toward operating-system-like abstractions in which the core loop serves as the kernel and everything else constitutes the OS.

Siz nasıl anlarsınız bilmem, makine öğrenmesinin kasik programlamadan temel farkı, modelin mümkün mertebe kuralları verilerden kendisinin çıkarmasıdır. Kodun sadece %1.6'sının decision logic geri kalan %98.4'ünün de operasyonel altyapıyla alakalı olması, yukarıdaki "Claude Code gives the model maximum decision latitude within a rich operational harness" cümlesiyle uyumlu. Yani dil modelinin (LLM'in) yönlendirmelerine kodla yazılmış kurallarla müdahaleyi minimize etmeye çalışmışlar anladığım kadarıyla. -Gürsu


--
Gruba davet etmek istediklerinize bu linki ulaştırabilirsiniz:
https://groups.google.com/g/yz-ve-insan/about
---
Bu iletiyi Google Grupları'ndaki "Yapay Zeka ve İnsan" grubuna abone olduğunuz için aldınız.
Bu grubun aboneliğinden çıkmak ve bu gruptan artık e-posta almamak için yz-ve-insan...@googlegroups.com adresine e-posta gönderin.
Bu tartışmayı görüntülemek için https://groups.google.com/d/msgid/yz-ve-insan/CAC5h89zuohkCpPDpEO6U2Y-0%2BzgeJO7hHauLnZprr1_%3DZQ4eGg%40mail.gmail.com adresini ziyaret edin.
Reply all
Reply to author
Forward
0 new messages