2024年12月31日 星期二

Explore Our Coverage of Artificial Intelligence 紐約時報

 

Explore Our Coverage of Artificial Intelligence


News and Analysis

  • OpenAI revealed details about its plans to adopt a new corporate structure that will remove the company from control by a nonprofit.

  • The A.I. company founded by Elon Musk, xAI, said that it had raised $6 billion, giving the start-up a major lift as it competes with rivals.

  • OpenAI unveiled a new system, OpenAI o3, which is designed to “reason” through problems involving math, science and computer programming.


The Age of A.I.

A.I. “slop” can become a threat to A.I. itself

 

A.I. “slop” can become a threat to A.I. itself

The rapid expansion of artificial intelligence has posed a conundrum: If A.I. learns from the internet, and the internet starts containing more artificially generated content, will A.I. start ingesting its own output? Researchers I spoke to described an unintentional feedback loop that can reduce the quality and diversity of A.I.-generated content in the long run.

While reporting for this article, I trained a generative A.I. to mimic handwritten numbers. At first, these A.I.-generated reproductions weren’t bad. But as I repeatedly fed this A.I. its own output, I was amazed to watch as its drawings became increasingly incoherent and blurry. This was a genuine “aha” moment that illustrated the central problem that researchers had described — one that they are working on ways to solve. — Aatish Bhatia



2024年12月26日 星期四

Gödel, Escher, Bach, and AI. A dazzlingly fast chatbot cannot replace the authentic and reflective voice of a thinking, living human being,

 Douglas Hofstadter

Gödel, Escher, Bach: an Eternal Golden Braid, also known as GEB, is a 1979 book by Douglas Hofstadter.
A dazzlingly fast chatbot cannot replace the authentic and reflective voice of a thinking, living human being, Douglas Hofstadter wrote in 2023.


Gödel, Escher, Bach, and AI

----
Alan Turing: The Enigma.Gödel, Escher, Bach: an Eternal Golden Braid, also known as GEB, is a 1979 book by Douglas Hofstadter.

---
神經質的數學天才Alan Turing, 遭殘忍迫害同志。1870年代到1950年代長達70年,英國政府認為同志是犯了【猥褻罪】,適用刑法,非常不人道,而且法院用【化學去勢】殘害四萬七千多位同志,造成本片主人翁,真人真事破戒希特勒的Enigma機械的Alan Turing,在41歲自殺身亡。
bookjacket

Alan Turing: The Enigma:
The Book That Inspired the FilmThe Imitation Game
Updated Edition
Andrew Hodges
With a foreword by Douglas Hofstadter and a new preface by the author

New York Times Bestseller
The Imitation Game, Winner of the 2015 Academy Award for Best Adapted Screenplay
Winner of the 2015 (27th) USC Libraries Scripter Award, University of Southern California Libraries
One of The Guardian’s Best Popular Physical Science Books of 2014, chosen by GrrlScientist
Paperback | 2014 | $16.95 / £11.95 | ISBN: 9780691164724
768 pp. | 5 x 7 3/4
SHOPPING CART
Google full text of this book:
 
Andrew Hodges' Home Page
The Imitation Game movie trailer
from the major motion picture, starring
Benedict Cumberbatch and Keira Knightley
NEW YORK TIMES BESTSELLER
The official book behind the Academy Award-winning film The Imitation Game, starring Benedict Cumberbatch and Keira Knightley
It is only a slight exaggeration to say that the British mathematician Alan Turing (1912-1954) saved the Allies from the Nazis, invented the computer and artificial intelligence, and anticipated gay liberation by decades--all before his suicide at age forty-one. This New York Times–bestselling biography of the founder of computer science, with a new preface by the author that addresses Turing’s royal pardon in 2013, is the definitive account of an extraordinary mind and life.
Capturing both the inner and outer drama of Turing’s life, Andrew Hodges tells how Turing’s revolutionary idea of 1936--the concept of a universal machine--laid the foundation for the modern computer and how Turing brought the idea to practical realization in 1945 with his electronic design. The book also tells how this work was directly related to Turing’s leading role in breaking the German Enigma ciphers during World War II, a scientific triumph that was critical to Allied victory in the Atlantic. At the same time, this is the tragic account of a man who, despite his wartime service, was eventually arrested, stripped of his security clearance, and forced to undergo a humiliating treatment program--all for trying to live honestly in a society that defined homosexuality as a crime.
The inspiration for a major motion picture starring Benedict Cumberbatch and Keira Knightley, Alan Turing: The Enigma is a gripping story of mathematics, computers, cryptography, and homosexual persecution.
Andrew Hodges teaches mathematics at the University of Oxford.
Review:
"Scrupulous and enthralling."--A. O. Scott, New York Times
"One of the finest scientific biographies ever written."--Jim Holt, New Yorker
"Andrew Hodges' 1983 book Alan Turing: The Enigma, is the indispensable guide to Turing's life and work and one of the finest biographies of a scientific genius ever written."--Michael Hiltzik, Los Angeles Times
"Turing's rehabilitation from over a quarter-century's embarrassed silence was largely the result of Andrew Hodges's superb biography, Alan Turing: The Enigma (1983; reissued with a new introduction in 2012). Hodges examined available primary sources and interviewed surviving witnesses to elucidate Turing's multiple dimensions. A mathematician, Hodges ably explained Turing's intellectual accomplishments with insight, and situated them within their wider historical contexts. He also empathetically explored the centrality of Turing's sexual identity to his thought and life in a persuasive rather than reductive way . . ."--Michael Saler, Times Literary Supplement
"On the face of it, a richly detailed 500-page biography of a mathematical genius and analysis of his ideas, might seem a daunting proposition. But fellow mathematician and author Hodges has acutely clear and often extremely moving insight into the humanity behind the leaping genius that helped to crack the Germans' Enigma codes during World War II and bring about the dawn of the computer age. . . . This melancholy story is transfigured into something else: an exploration of the relationship between machines and the soul and a full-throated celebration of Turing's brilliance, unselfconscious quirkiness and bravery in a hostile age."--Sinclair McKay, Wall Street Journal
"A first-class contribution to history and an exemplary work of biography."--I. J. Good, Nature
"An almost perfect match of biographer and subject. . . . [A] great book."--Ray Monk, Guardian
Table of Contents:
List of Plates ix
Foreword by Douglas Hofstadter xi
Preface xv
PART ONE: THE LOGICAL
1 Esprit de Corps to 13 February 1930 3
2 The Spirit of Truth to 14 April 1936 60
3 New Men to 3 September 1939 141
4 The Relay Race to 10 November 1942 202
BRIDGE PASSAGE to 1 April 1943 305
PART TWO: THE PHYSICAL
5 Running Up to 2 September 1945 325
6 Mercury Delayed to 2 October 1948 394
7 The Greenwood Tree to 7 February 1952 491
8 On the Beach to 7 June 1954 574
Postscript 665
Author's Note 666
Notes 680
Acknowledgements 714
Index 715

ChatGPT LLM 不能解魔術方塊 (ChungChih Li) CS computer science. ChatGPT可成功地指出,偽造文書罪,洗錢罪,受賄,貪污或挪用公款,但無法具體分析與串聯事實,對案情內容的關聯也理不出頭緒

 

ChungChih Li

我是對柯的犯罪事實完全沒興趣,柯文哲做為一個政治人物對台灣的危害遠遠超過他做為一個貪汙犯。當然,台灣有幸,因他的犯罪被揭穿而消弭了他的政治力。但這是第三好的結果,第二好的結果應該是因他的政治主張而被唾棄,這部發酵得並不部全面,只影響了深綠的人。最好情況是因他的文化水準而拒絕此人於始,這才能算是台灣文化最根本的覺醒,只有這樣,這種政治詐術得逞的機會才會變小。
這是我對他的起訴書沒意願看的原因,但我還真想測試一下ChatGPT的能耐,結果頗差,它可成功地指出,偽造文書罪,洗錢罪,受賄,貪污或挪用公款,但無法具體分析與串聯事實,對案情內容的關聯也理不出頭緒。簡單講,律師暫時還不會失業。但我猜這是中文部分的訓練不足,因為我知道美國的事務所已用了許多AI在分析案件了 (不只是搜尋案例),例如 Westlaw Edge 和 Lexis+,是為法律設計的 LLM。聽說很厲害。


你如果去問 ChatGPT LLM 能不能解魔術方塊,它會洋洋得意說可以,還秀出程序,我追問了三次,它才承認 LLM 沒辦法解,不是呈現問題困難,是 LLM 本來就不可能「想出」解決的辦法。CS 的學生比較不會被唬,但 IS 的學生比較抓不到重點。 所以,CS 的訓練還是重要的。我對畢業生一直強調不要忽略 AI 對 IT 產業的衝擊,除了要會用,還要能很快告訴你老闆什麼是 AI 能做的,什麼是 AI 不能做的,AI 產出的東西,哪邊要注意。

2024年12月25日 星期三

國內的軟體人才 (洪士灝)

 

前天這篇長文,內容主要評論的是國內的軟體發展,雖然提及與硬體產業的相關性,但並非批判代工產業,也不大好簡化歸因「代工思維」。
我的基本立場是,代工是台灣產業的重要基礎,不宜藐視,但也必須與時俱進,不能故步自封。
現代完整的資訊系統解決方案的軟硬體供應鏈很複雜,其中很多委外研發、代工製造的項目,價值高低有差,提昇價值的關鍵在於技術門檻、智財權、以及是否有辦法做出差異化,如果只是靠價廉物美的人力資源和薄利多銷的戰術,特別是在一窩蜂(內卷)的情況下,會賺得很辛苦。
因此我是支持「有門檻、有特色的代工」,例如台積電和聯發科能夠做得頂尖,固然其成功的因素仍有很大部分來自於員工的血汗,但這些公司也重視技術研發,否則如何在國內脫穎而出?如何在國外與Intel、Samsung爭鋒?
然而,固守在硬體供應鏈的思維,長期而言會讓多數的硬體業者的價值難以成長,是個隱憂;由硬體思維主導軟體開發的現況,也是國內軟體技術層次成長的障礙。
至於走品牌或新創路線,那是說得容易、做得難,尤其是軟體,不應該一昧鼓吹,要有配套和戰略,實情有點複雜,改天再談。
國內的軟體人才
昨天的演講邀請到大學時代曾經跟我做過專題研究的許秉倫,講他在美國留學就業後創立的Liger Kernel開源軟體專案。有別於一般的演講,我希望這場演講著重於技術面的分享,進而呈現開源軟體的世界給聽眾。主要的演講約一小時,但演講會上的問答以及會後聽眾找講者的交流,超過一個小時,可見得言之有物、意猶未盡。這場演講的錄影,如果臉友們回應熱烈的話,在徵得講者和贊助單位同意之後,我們會考慮上網公開分享。
秉倫做的這個Liger Kernel專案,技術門檻不低,因為要懂得如何駕馭含有GPU或硬體加速器的異質計算架構,不僅要能運用CUDA寫GPU程式,還在系統上使用效能工具找尋效能瓶頸加以優化,同時與其他開源專案(Triton)密切接軌。我將這類的開源專案稱之為開源系統與中介軟體,一直是我指導學生鑽研的方向,也是我希望國內產業界多加以重視並支持的領域,只可惜國內積極支持的業者太少。
業者不支持就算了,還學國外軟體業者考程式題篩選面試者,搞得學生在學校的時候競相拼刷題,價值觀錯亂。原本LeetCode這些網站是為了幫助學生們學習寫程式,卻因為美國軟體業求職者太多,被用來作為面試前的篩選工具,而越演越烈變成惟刷題強者才錄用。秉倫在演講一開始就說(參考附圖),美國軟體公司已經不大靠刷題了,也不見得在乎你的成績,而是看你實際上做過什麼專案、能否進入一些未曾碰觸過的新領域。秉倫在知名的求職平台LinkedIn工作,看到美國最新的趨勢,但國內還停留在過去。
很多學生從小為了考試而一路刷題到現在,還要繼續花時間刷題,我也管不著,只能建議各位趁著在校時多充實素養、學一些有用的東西。在科舉時代,熟背經史詩文的人可以上位當官,但不見得能做大事、解決問題。要知道,現在的AI比絕大多數人都要會刷題、會作刻板文章、寫程式,於是有人說學生們不用再學寫程式了。
的確,很可能不用再寫一些別人寫過的程式,但如果不學寫程式的話,如何懂得電腦科學中所蘊藏的道理(運算思維)呢?如果遇到未曾碰觸過的新領域,光靠嘴砲指揮AI就能研發程式來解決問題嗎?AI最可能淘汰的是只會照本宣科、依樣畫葫蘆的人,奉勸各位不要一昧從眾,而喪失個人的特色。
在這場演講中,秉倫不只一次被問到,LinkedIn為什麼願意讓他花大把時間做開源軟體專案,而且將這種專業的好東西分享給包含競爭對手的其他公司呢?其實這些問題正好曝露出國內業界的生態距離國外太過遙遠,聽眾不只難以置信,也會質疑在國內是否有這樣的機會。
首先,放眼國內開源社群所關注的專案,很多都是基於個人的興趣和熱情,只能利用個人閒暇時間參與,往往跟任職公司的業務脫節。理論上,如果參與的開源軟體專案與公司業務相關性頗高,公司或許會支持,但實際上國內硬體製造的思維中還有極濃厚的保護意識,與開源文化背道而馳,深怕公司機密洩漏,所以要這些公司支持開源軟體社群的發展是困難重重。
我認為,追求軟體的價值要有方法,心態與硬體製造截然不同,然而賺大錢的晶片系統廠網羅了許多頂尖大學培育的軟體人才,然後讓他們在硬體製造的思維框架中只做對公司有利的研發工作,這樣的大環境是相當不利於國內軟體人才培育和發展的。我這些年的個人經驗告訴我,在這個極度重視硬體製造的國度,願意積極探索和開創軟體價值的晶片系統廠少得可憐。
曾經,在中國製造的壓力下,有些晶片系統廠開始思考如何轉型,除了在原本擅長的硬體製造技術上精益求精之外,發展較為貼近硬體的系統軟體、培育相關人才。但意外幸運地,近年受惠於美中經濟戰,晶片和系統產業的成功,讓國內許多人引以為傲的同時,並沒有很積極投入資源培育系統軟體人才、致力於提升系統軟體技術,我認爲非常可惜。
曾經是晶片系統產業龍頭的Intel,為何這些年逐漸失去競爭力,是一個值得參考的案例。不過美國就算Intel不行了,也有很多資訊業的翹楚,除了炙手可熱的NVIDIA之外,更重要的是軟體業的霸主們,因為那是美國業者想要的價值。反觀國內,硬體製造之外,能夠追求哪些價值?
因為國內的系統產業如果真的想提昇價值,除了在原本擅長的硬體製造技術上精益求精之外,發展較為貼近硬體的系統軟體、培育相關人才,是最能借重原有的晶片、系統產業的利基的作為。但很可惜地,大多數廠商只想撿現成人才,沒有很積極投入資源培育人才,在系統軟體的技術上難以躍升,實在可惜。
話說曾經是晶片系統產業龍頭的Intel,為何這些年逐漸失去競爭力,是一個值得參考的案例。不過美國就算Intel不行了,也有很多資訊業的翹楚,除了炙手可熱的NVIDIA之外,更重要的是軟體業的霸主們,因為那是美國業者想要的價值。反觀國內,硬體製造之外,能夠追求哪些價值?
當我上個月在超級電腦年會(SC24)的展場上看到一票台灣系統廠都秀出長相差不多的GB200/NVL72的機櫃之時,我心想,能夠把超級電腦變成幾乎每家系統廠都會代工生產的盛況,真正厲害和賺大錢的是NVIDIA!而台灣廠商在NVIDIA獨大,各家做不出明顯的差異化,又彼此競爭的情況下,只能像小弟般跟著NVIDIA大哥做事分紅,遲早利潤會越來越薄的。
我絕對不是嫌棄代工,只是指出國內大量優秀的軟體人才集中在硬體製造產業,只限做一些有助於長官提升業績的研發工作,難以發揮其能力。我敢說,這並不是國內大學加強培養軟體人才就能解決的問題;相對地,不少國內培育的人才到了國外,卻能夠獲得重用,還能海闊天空地發展。
對了,有位學生問秉倫,你因為看到某個效能問題而開創出Liger Kernel,為什麼其他人沒看到?沒有被其他人解掉?
秉倫說,他在LinkedIn一開始用的都是配備80GB記憶體的GPU,因緣際會遇到朋友提及GPU只有40GB的記憶體,不夠訓練大模型的時候,才開始思考如何節省記憶體 -- 或許很厲害的人有用不完GPU,沒有碰到這個問題?
我在會後對學生說,這類問題很常發生,只要多碰觸實作問題就會看到。或許有人認為不夠學術性、不是SOTA,但絕對不要忽視這類工作對學術的貢獻。你看今年的諾貝爾獎頒給AI技術的引航者、應用工具的開發者,就有這個意涵。對我來說,只要學習做對人類有幫助的技術,就是學術。