WFU

2024/05/02

STORM+Claude 產文章範例: 泡泡尿與腎臟健康


此篇文章,由 STORM_longform_writing_Claude_Contest (改寫中文版) 生成,原始Github連結

  • 文章為 Claude api 輸出
  • 使用 api model: claude-opus
  • token 使用量: input 19155 , output 8456。預估花費: US$0.92
  • 文章的文字沒有修改,只有去除額外產生的標題,另重新依照輸出的markdown格式排版。
  • 搜尋 Tavily API 使用次數: 3
  • 總字數: 3406字 (MS word計算)

以下正文開始 ☝

2024/04/20

RT&EBM - 投影片PDF檔

2024/03/19

向專家學習:ihower 「LLM 應用開發工作坊」課後心得

【初次】


第一次的課程開在11月,深深覺得講師 ihower 對於 LLM 真是是有「愛」。對於 LLM 各種不同的技巧,從初級入門的串接 API,到 prompt 技巧、CoT,然後 RAG 等,通通都有涉獵。非資訊本科生的我,如同劉姥姥進大觀園般,看著 ihower 快速講解每個 colab 程式。

說起來慚愧,11月後我自己在 LLM 的接觸面向減少許多,追了大半年的 AI 潮流,東碰一點、西碰一小塊,到最後真的累了,只想要躺平。

離上次的課程才隔四個月,謝謝主辦單位主動在課程群組歡迎大家「回訓」。

回訓,或許是重新讓自己再檢視看看,對於 LLM 這塊領域,哪些層面是自己有興趣的。


【回訓】


跟第一次上課的感想一樣,短短三堂加起來8小時的課,知識量真的很大。

三堂課,知識架構是一層層推上去的,也代表難度也是推疊上去的。

第一堂的重點放在 API的串接、prompt 基本知識(zero-shot, few-shot, CoT)。
第二堂的重點有 Chaining prompt , Embedding 和 RAG(檢索增強生成)
第三堂的重點是 Agent 及 functional calling。

非本科生的我,其實在第二堂的 RAG 開始,腦袋運轉就開始打結了,哈。

2023/11/05

[實證初階工作坊] 情緒分析


Part 1: 
Paul Ekman 於 1972年提出的基本情感
喜悅(Joy)
憂傷(Sadness)
憤怒(Anger)
恐懼(Fear)
驚訝(Surprise)
厭惡(Disgust)




Part 2:
分析大堂課情境中「黃醫師」的情緒狀態。要思考「prompt」怎麼寫、怎麼輸出…

###
「333,二樓門診治療室,333,二樓門診治療室…」 在一個平凡的下午,眾多病人在院內等待看診,醫療團隊在各自的崗位上工作時,突然間,一個警報聲響起,院內緊急急救小組隨即馬上啟動。 緊急的CPR流程在第一時間就展開,相關小組成員很快到達了現場,現場的 leader 王醫師很熟練的下著指令:壓胸、Ambu給氣、準備給予 bosmin 1 amp。門診醫師黃醫師被通知,也趕緊過來現場,目前在急救病人平常是黃醫師在追蹤的,這次因為病人的血色素只有 5.5 ,所以他讓病人來治療室輸血。 輸血的流程才在準備時,病人開始覺得胸口緊緊的不舒服,過了十分鐘後,意識開始不清楚,很快的人就叫不醒。治療室的護理師發現異況後隨即量病人的脈搏,但量不到,便緊急通知總機廣播333。 黃醫師到急救現場時,詢問現場是否有皮質類固醇的針劑藥物,黃醫師表示若現場有相關藥物的針劑,馬上給病人打一劑類固醇。緊救現場的leader王醫師雖然自己心中沒有很贊同緊救當下給予類固醇,因為目前使用類固醇在心跳停止 (cardiac arrest) 時並沒有共識,便問黃醫師為何想用此藥物。 黃醫師表示,雖然類固醇對預後似乎有爭議,但有些報告有做出急救後恢復自發性循環 (return of spontaneous circulation, ROSC) 的比率比較高,即使可能只有一點點幫助,他還是想試看看。
###

2023/11/04

[亞大實證AI] CEBM Q2-Q5 原始內文 與 題目拆解

主題拆解示範 prompt for Q2 part 1

這是 CEBM checklist for SR Q2 的內容
我想要拆解 Q2 為更細項的小題目,目標是 3~7 題。拆解的小題目要符合原始 Q2 的內容/評估順序/原則。
###
CEBM for SR: Q2
Is it unlikely that important, relevant studies were missed?

What is best? 
The starting point for a comprehensive search for all relevant studies is the major bibliographic databases (eg Medline, Cochrane, EMBASE, etc) but should also include a search of reference lists from relevant studies and contact with experts, particularly to inquire about unpublished studies. The search should not be limited to English language only. The search strategy should include both MESH terms and text words.

Where do I find the information?
The Methods section should describe the search strategy, including the terms used, in some detail. The Results section will outline the number of titles and abstracts reviewed, the number of fulltext studies retrieved, and the number of studies excluded together with the reasons for exclusion. This information may be presented in a figure or flow chart.

In this paper:
[ ] Yes
[ ] No
[ ] Unclear

Comment: ###


主題拆解示範 prompt for Q2 part 2

請將這些拆解的內容與目標,寫成適合用來評讀文獻的 prompt。各小題要各自評估(回答Yes/No/Unclear),回答時要引用原始文獻字句,最後總結要對原始問題做最後評估。最後的輸出要用 table。

請將此 prompt 輸出於 code block 內


CEBM checklist for SR: Q2

CEBM for SR: Q2
Is it unlikely that important, relevant studies were missed?

What is best? 
The starting point for a comprehensive search for all relevant studies is the major bibliographic databases (eg Medline, Cochrane, EMBASE, etc) but should also include a search of reference lists from relevant studies and contact with experts, particularly to inquire about unpublished studies. The search should not be limited to English language only. The search strategy should include both MESH terms and text words.

Where do I find the information?
The Methods section should describe the search strategy, including the terms used, in some detail. The Results section will outline the number of titles and abstracts reviewed, the number of fulltext studies retrieved, and the number of studies excluded together with the reasons for exclusion. This information may be presented in a figure or flow chart.

In this paper:
[ ] Yes
[ ] No
[ ] Unclear

Comment: 


CEBM checklist for SR: Q3

CEBM for SR: Q3  
Were the criteria used to select articles for inclusion appropriate?

What is best?  
The inclusion or exclusion of studies in a systematic review should be clearly defined a priori. The eligibility criteria should specify the patients, interventions or exposures, and outcomes of interest. In many cases, the type of study design will also be a key component of the eligibility criteria.

Where do I find the information?  
The Methods section should describe in detail the inclusion and exclusion criteria, which typically includes the study design.

In this paper:  
[ ] Yes  
[ ] No  
[ ] Unclear  

Comment:


CEBM checklist for SR: Q4

CEBM for SR: Q4  
Were the included studies sufficiently valid for the type of question asked?

What is best?  
The article should describe how the quality of each study was assessed using predetermined quality criteria appropriate to the type of clinical question (e.g., randomization, blinding, and completeness of follow-up).

Where do I find the information?  
The Methods section should describe the assessment of quality and the criteria used. The Results section should provide information on the quality of the individual studies.

In this paper:  
[ ] Yes  
[ ] No  
[ ] Unclear  

Comment:


CEBM checklist for SR: Q5

CEBM for SR: Q5
Were the results similar from study to study?

What is best?  
Ideally, the results of the different studies should be similar or homogeneous. If heterogeneity exists the authors may estimate whether the differences are significant (chi-square test). Possible reasons for the heterogeneity should be explored.

Where do I find the information?  
The Results section should state whether the results are heterogeneous and discuss possible reasons. The forest plot should show the results of the chi-square test for heterogeneity and discuss reasons for heterogeneity, if present.

In this paper:  
[ ] Yes  
[ ] No  
[ ] Unclear  

Comment: