Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Features
Speaker Deck
PRO
Sign in
Sign up for free
Search
Search
容器遷移與生成式 AI 應用的趨勢剖析
Search
Browny Lin
May 12, 2024
Technology
0
86
容器遷移與生成式 AI 應用的趨勢剖析
Browny Lin
May 12, 2024
Tweet
Share
More Decks by Browny Lin
See All by Browny Lin
Testing with Concurrency in Go
brownylin
2
350
Dependency Injection in Go
brownylin
3
2.3k
How to build habit-forming products
brownylin
0
180
Introduction to Docker
brownylin
0
1k
Introduction to Hidden Markov Model
brownylin
1
270
Other Decks in Technology
See All in Technology
TerraformをSaaSで使うとAzureの運用がこんなに楽ちん!HCP Terraformって何?
mnakabayashi
0
100
Devin(Deep) Wiki/Searchの活用で変わる開発の世界観/devin-wiki-search-impact
tomoki10
0
290
CIでのgolangci-lintの実行を約90%削減した話
kazukihayase
0
140
New Cache Hierarchy for Container Images and OCI Artifacts in Kubernetes Clusters using Containerd / KubeCon + CloudNativeCon Japan
pfn
PRO
0
150
SFTPコンテナからファイルをダウンロードする
dip
0
110
新卒3年目の後悔〜機械学習モデルジョブの運用を頑張った話〜
kameitomohiro
0
150
Claude Code どこまでも/ Claude Code Everywhere
nwiizo
40
22k
Amplifyとゼロからはじめた AIコーディング 成果と展望
mkdev10
1
150
Contract One Engineering Unit 紹介資料
sansan33
PRO
0
6.4k
Text-to-SQLの評価データセットを作って最新LLMモデルの性能評価をしてみた
gotalab555
3
770
TODAY 看世界(?) 是我們在看扣啦!
line_developers_tw
PRO
0
120
ゆるSRE #11 LT
okaru
1
590
Featured
See All Featured
Principles of Awesome APIs and How to Build Them.
keavy
126
17k
A Modern Web Designer's Workflow
chriscoyier
693
190k
Understanding Cognitive Biases in Performance Measurement
bluesmoon
29
1.8k
Dealing with People You Can't Stand - Big Design 2015
cassininazir
367
26k
The Success of Rails: Ensuring Growth for the Next 100 Years
eileencodes
45
7.3k
Build The Right Thing And Hit Your Dates
maggiecrowley
36
2.7k
BBQ
matthewcrist
89
9.7k
VelocityConf: Rendering Performance Case Studies
addyosmani
329
24k
Practical Tips for Bootstrapping Information Extraction Pipelines
honnibal
PRO
20
1.3k
The Language of Interfaces
destraynor
158
25k
Bootstrapping a Software Product
garrettdimon
PRO
307
110k
[RailsConf 2023] Rails as a piece of cake
palkan
55
5.6k
Transcript
Taiwan · Singapore · Thailand · Japan · Hong Kong
· Malaysia iKala Confidential and Proprietary 容器遷移與生成式 AI 應用的趨勢剖析 Browny Lin (林士翔) Technical Director, iKala Cloud
Agenda • Trend on Container • Trend on AI, especially
Gen AI • Takeaway
3 Trend on Container
Why use Container? Source: Why use Containers? 1. Runs anywhere
2. Light-weight
Why use Kubernetes? Source: Why use Containers? 1. Container orchestration
2. Declarative
Source: 10 insights on real-world container use
Source: 10 insights on real-world container use
None
Source: 10 insights on real-world container use
Source: 10 insights on real-world container use
Source: 10 insights on real-world container use
None
Source: 10 insights on real-world container use
Source: 10 insights on real-world container use
Recap 1. Serverless 2. GPU/AI workload 3. Utilization 4. Stability
16 Trend on AI, especially Gen AI
Source: IOT ANALYTICS Q2 2023 全世界的企業領導者現在眼中只有生成式 AI
We tend to overestimate the effect of a technology in
the short run and underestimate the effect in the long run - Roy Amara Amara's Law: 短期高估;長期低估 “
Source: https://chat.lmsys.org/
Source: LMSYS Chatbot Arena Leaderboard
Source: HAI AI Index Report 2024 💡AI 在部分任務上的 表現,已經超越人類 無論你最擅長什麼(通常是你最喜歡的),你可能比人
工智慧更好,但無論你不擅長什麼,人工智慧都可以 幫助填補空白。
Source: HAI AI Index Report 2024 💡產業界 > 學術界 的研究量能
Source: HAI AI Index Report 2024 💡基礎模型 訓練成本巨大
Source: HAI AI Index Report 2024 💡AI 不會取代你, 但會用 AI
的人會
Source: Put AI to work for application modernization 💡如何落地: 從
+AI 到 AI+
💡價值創造: ROI 難以計算, 錯過的 風險更大 Source: Now decides next: Getting
real about Generative AI - Deloitte
💡規模化導入: GenAI 工具導入仍低, 風險(資料)和利益仍需權衡 Source: Now decides next: Getting real
about Generative AI - Deloitte
💡AI 賦能: 重新設計工作流程, 改變人才策略 Source: Now decides next: Getting real
about Generative AI - Deloitte
1. 要有好的 AI 策略 必先有好的 數據策略 2. 找出 AI 對自身既有商業模式
能帶來的加值 3. 將 AI 當成電力來規劃事業策 略 企業應如何打造 +AI 策略
30 Takeaway
Source: How to Make AI UX Your Moat 💡AI 不是護城河,
UX (Data) 才是
ChatGPT 震驚了世界不是因為模型而是體驗 • Textbox in, streaming out 模擬了閱讀與聆聽 • Abstracting
state 不需要用戶將先前的問題和答案作為附加狀態傳遞給 新查詢。ChatGPT 提取過去的狀態,以便將來的查詢 可以自然地參照狀態。真正的「聊天」 Source: How to Make AI UX Your Moat
好的體驗需要在互動的每個環節下功夫 💡UX+ • 命令:讓用戶創建任務預設,然後能夠搜尋它們 • 約束:在應用程序中創建自動拉入的用 戶設置 • 背景:簡化數據收集
Customer Assistance: Apply 3P (Present -> Practical -> Powerful) Source:
Zalando to launch a fashion assistant powered by ChatGPT
None
https://ikala.ai +886-2-8768-1110
[email protected]