From cc2b2257c03a04a018730e4638a3e6b0c1abee86 Mon Sep 17 00:00:00 2001 From: Godopu Date: Sun, 19 Jul 2026 11:27:51 +0900 Subject: [PATCH] docs: add Model eats the Scaffolding slide showing transition of Planning to models --- lib/make_slide_for_260724_seminar/SLIDE.md | 12 ++++++++++++ .../paper_draft/SLIDE.md | 12 ++++++++++++ 2 files changed, 24 insertions(+) diff --git a/lib/make_slide_for_260724_seminar/SLIDE.md b/lib/make_slide_for_260724_seminar/SLIDE.md index 6cd451a..35d3f07 100644 --- a/lib/make_slide_for_260724_seminar/SLIDE.md +++ b/lib/make_slide_for_260724_seminar/SLIDE.md @@ -33,6 +33,18 @@ marp: true --- +## πŸš€ νŒ¨λŸ¬λ‹€μž„ μ‰¬ν”„νŠΈ: Model eats the Scaffolding +μ—μ΄μ „νŠΈ ν•˜λ„€μŠ€(Scaffolding)의 핡심 κΈ°λŠ₯은 μž₯기적으둜 λͺ¨λΈ λ‚΄λΆ€λ‘œ νŽΈμž…(Upstream)λ˜λŠ” μΆ”μ„Έμž…λ‹ˆλ‹€. + +- **μΆ”λ‘  μ‹œκ°„ μ—°μ‚° (Test-time Compute)의 λ„μž…** + - Gemini Thinking, OpenAI o1/o3 λ“± μ΅œμ‹  λͺ¨λΈμ€ μ™ΈλΆ€ 루프 없이 λͺ¨λΈ λ‚΄λΆ€μ—μ„œ 슀슀둜 κ³„νš(Planning)ν•˜κ³  μžκ°€ μˆ˜μ •(Reflection)을 끝낸 ν›„ μ΅œμ’… 닡변을 λ„μΆœ +- **인지(Cognitive)와 집행(Execution)의 뢄리** + - **λͺ¨λΈ(Model)**: 고차원 μΆ”λ‘ κ³Ό κ³„νš 섀계(Planning)λ₯Ό λ‹΄λ‹Ή + - **μ—μ΄μ „νŠΈ 인프라(Harness)**: λ³΄μ•ˆ μƒŒλ“œλ°•μ‹±, μƒνƒœ 보쑴, κΆŒν•œ ν†΅μ œ 및 도ꡬ 싀행을 λ‹΄λ‹Ή +- **μ•ŒνŒŒ(Alpha)의 이동**: λ‹¨μˆœν•œ μ—μ΄μ „νŠΈ κ³„νš λ£¨ν”„μ˜ κ΅¬ν˜„μ„ λ„˜μ–΄, λ³΅μž‘ν•œ 인프라 μ œμ–΄μ™€ 이쒅 ν‘œμ€€ 연동(ACP/A2A)이 핡심 차별화 μš”μ†Œλ‘œ 진화 + +--- + # 1. AI μ—μ΄μ „νŠΈ κ΅¬ν˜„μ„ μœ„ν•œ 3λŒ€ 핡심 μ—”μ§€λ‹ˆμ–΄λ§ 기법 ## AI λͺ¨λΈ(LLM) vs AI μ—μ΄μ „νŠΈ diff --git a/lib/make_slide_for_260724_seminar/paper_draft/SLIDE.md b/lib/make_slide_for_260724_seminar/paper_draft/SLIDE.md index 6cd451a..35d3f07 100644 --- a/lib/make_slide_for_260724_seminar/paper_draft/SLIDE.md +++ b/lib/make_slide_for_260724_seminar/paper_draft/SLIDE.md @@ -33,6 +33,18 @@ marp: true --- +## πŸš€ νŒ¨λŸ¬λ‹€μž„ μ‰¬ν”„νŠΈ: Model eats the Scaffolding +μ—μ΄μ „νŠΈ ν•˜λ„€μŠ€(Scaffolding)의 핡심 κΈ°λŠ₯은 μž₯기적으둜 λͺ¨λΈ λ‚΄λΆ€λ‘œ νŽΈμž…(Upstream)λ˜λŠ” μΆ”μ„Έμž…λ‹ˆλ‹€. + +- **μΆ”λ‘  μ‹œκ°„ μ—°μ‚° (Test-time Compute)의 λ„μž…** + - Gemini Thinking, OpenAI o1/o3 λ“± μ΅œμ‹  λͺ¨λΈμ€ μ™ΈλΆ€ 루프 없이 λͺ¨λΈ λ‚΄λΆ€μ—μ„œ 슀슀둜 κ³„νš(Planning)ν•˜κ³  μžκ°€ μˆ˜μ •(Reflection)을 끝낸 ν›„ μ΅œμ’… 닡변을 λ„μΆœ +- **인지(Cognitive)와 집행(Execution)의 뢄리** + - **λͺ¨λΈ(Model)**: 고차원 μΆ”λ‘ κ³Ό κ³„νš 섀계(Planning)λ₯Ό λ‹΄λ‹Ή + - **μ—μ΄μ „νŠΈ 인프라(Harness)**: λ³΄μ•ˆ μƒŒλ“œλ°•μ‹±, μƒνƒœ 보쑴, κΆŒν•œ ν†΅μ œ 및 도ꡬ 싀행을 λ‹΄λ‹Ή +- **μ•ŒνŒŒ(Alpha)의 이동**: λ‹¨μˆœν•œ μ—μ΄μ „νŠΈ κ³„νš λ£¨ν”„μ˜ κ΅¬ν˜„μ„ λ„˜μ–΄, λ³΅μž‘ν•œ 인프라 μ œμ–΄μ™€ 이쒅 ν‘œμ€€ 연동(ACP/A2A)이 핡심 차별화 μš”μ†Œλ‘œ 진화 + +--- + # 1. AI μ—μ΄μ „νŠΈ κ΅¬ν˜„μ„ μœ„ν•œ 3λŒ€ 핡심 μ—”μ§€λ‹ˆμ–΄λ§ 기법 ## AI λͺ¨λΈ(LLM) vs AI μ—μ΄μ „νŠΈ