星期二, 2月 25, 2025

EX#2

 如何繳交作業 How to turn in your homework exercise.


課堂練習的工具

使用 Claude Sonnet 3.7

使用 ChatGPT 4o, o1

使用 Gemini 2.0 Flash

使用 Grok 3


課堂練習繳交期限 

Deadline: 

1. end of session 

2. This Saturday at 23:59


課堂練習繳交方式 

Send all the links to TA victorhuang111013@gmail.com and me chang212@gmail.com by email with subject EX#2  [your ID, your name]



1. 使用 Google Scholar 檢查前次作業你所產生報告當中文獻的真實性,將每一篇文獻,逐一輸入Google Scholar,確認是否為真實論文。查核結果建議放在Google Drive 然後把檔案分享出來,提供連結即可






3. Illustrations in SVG 

文字說明製作插圖





例子



4. 依據你上週的雷射武器報告,製作至少兩個插圖。

限使用 SVG 向量圖,不可使用仍有版權疑慮的 ChatGPT Dalle, 或是 Microsoft Image Creator






班上同學作品(張銘峻)


5. 使用科技英文描述以下圖片並產生動畫(看圖說故事,還記得嗎)




Animate a simple illustration with a light blue background depicting a small white robot (with green accents) at the base of a brown tree trunk or branch. The robot appears to be sawing the branch it's sitting on. sample


6. 計算面積




有些模型(ChatGPT 4o, Claude Sonnet 3.7)以內容比對見長,有些(ChatGPT o1, o3-mini, Grok 3 Think, Claude Sonnet 3.7 Extended)以推理見長,前者類似右腦,後者類似左腦。 計算面積這題並沒有辦法靠內容比對這種直觀法就能得到答案,推理模型會比較適合

建議使用各種AI(特別是有推理能力的)

Claude Sonnet 3.7 Extend
Grok 3 Think
ChatGPT o1
ChatGPT o3-mini


最好的程式語言

 



星期一, 2月 24, 2025

AI 作為一種實驗科學

 



AI 求解在數學上是如上圖在一個複雜非線性函數尋找最小誤差的數學過程,
以這張圖為例
每次答案之所以不一樣是因為AI每次啟動都會給予一個新的起始點
以至於每次最後收斂的誤差最低點都不會一樣


直覺式程式設計







 

星期六, 2月 22, 2025

River crossing puzzle 3

  

 problem statement



17 step solution 黃品勛







(fails to solve. Helps illustrate.)



Why LLM performs poorly with reasoning and planning





Code gen (Python)

code gen python A* search for ::River Crossing Puzzle ...




change to js



run js



use React to visualize the steps





Monologue is best.


reasoning mode in o1



Talking to itself



Running into difficulties? Cute



hours? Ha Ha Ha












River Crossing Puzzle 2

 我們解決了較簡單的  渡河問題 1,接下來我們來看

渡河問題2 (statement),Problem via 許元銘


如何驗證結果


Claude solution non-reasoning (incorrect)


ChatGPT o1 solution 1 

  • backtracking observed
  • cannot work independently
  • some human corrections required
  • overall, it is helpful


ChatGPT fails (4o or o1)

(use Claude to visualize the issue or debug)












Code Gen

Use ChatGPT o1 to generate A* search

4o fails to generate A* search that works


Use Claude second time to generate A*

 it takes 15 steps. (thanks to the python code generated by Claude)




another 15-step solution



yet another 15-step

one more 15-step

last 15-step


additional condition - Puzzle 3


心得

打開Sonnet 3.7推理模式
我目前還沒有辦法完全正確渡河
最好的一次用了2:57,錯在第12步,不過是個小失誤,不花腦筋很容易改

至於跑到耗盡token那次4:47,居然沒有給出答案


可以這麼說,這題渡河用 python 寫 A*
可以找到最佳解 15 步

我目前使用所有 no code 方式, 無論 ChatGPT, grok, Claude 各種推理模式,沒有一次能得到完美無瑕的解



to study later

Use Claude to generate A* which is unfortunately incorrectly generated as follows.



wrap as html


text display


visualize scenario step-by-step (default demo)



visualize scenario step-by-step (real case)





completion




if Python stays, we can use Google colab.




Got the result in text. Use Claude to visualize solution.