原文標(biāo)題
The economics of AI
A stochastic parrot in every pot?
What a leaked memo from Google reveals about the future of artificial intelligence
人工智能經(jīng)濟(jì)學(xué)
每個(gè)鍋里都有一只隨機(jī)鸚鵡？
谷歌泄露的一份備忘錄揭示了人工智能的未來(lái)

Open-source AI is booming. That makes it less likely that a handful of firms will control the technology
開(kāi)源人工智能正在蓬勃發(fā)展，這使得少數(shù)公司壟斷該技術(shù)的可能性降低

[Paragraph 1]
THEY HAVE changed the world by writing software. But techy types are also known for composing lengthy memos in prose, the most famous of which have marked turning points in computing.
他們通過(guò)編寫軟件改變了世界。但技術(shù)人員也以因?qū)戦L(zhǎng)篇備忘錄而聞名，其中最著名的是計(jì)算領(lǐng)域的標(biāo)志性轉(zhuǎn)折點(diǎn)備忘錄。

Think of Bill Gates’s “Internet tidal wave” memo of 1995, which reoriented Microsoft towards the web; or Jeff Bezos’s “API mandate” memo of 2002, which opened up Amazon’s digital infrastructure, paving the way for modern cloud computing.
例如1995年比爾.蓋茨寫的 "互聯(lián)網(wǎng)浪潮 "備忘錄，它將微軟重新定位于網(wǎng)絡(luò)領(lǐng)域；又如2002年杰夫.貝索斯寫的 "API授權(quán) "備忘錄，它開(kāi)放了亞馬遜的數(shù)字基礎(chǔ)設(shè)施，為現(xiàn)代云計(jì)算鋪平了道路。

Now techies are abuzz about another memo, this time leaked from within Google, titled “We have no moat”.
現(xiàn)在技術(shù)人員們對(duì)另一份從谷歌內(nèi)部泄露的備忘錄議論紛紛，標(biāo)題是 "我們沒(méi)有護(hù)城河"。

Its unknown author details the astonishing progress being made in artificial intelligence (AI)—and challenges some long-held assumptions about the balance of power in this fast-moving industry.
該備忘錄的匿名作者詳細(xì)介紹了人工智能領(lǐng)域正取得的驚人進(jìn)展--并對(duì)這個(gè)快速發(fā)展的行業(yè)中一些長(zhǎng)期存在的均衡勢(shì)力假設(shè)提出了挑戰(zhàn)。

[Paragraph 2]
AI burst into the public consciousness with the launch in late 2022 of ChatGPT, a chatbot powered by a “l(fā)arge language model” (LLM) made by OpenAI, a startup closely linked to Microsoft.
隨著2022年底ChatGPT的推出，人工智能進(jìn)入公眾視野，這是一個(gè)由OpenAI開(kāi)發(fā)的 "大型語(yǔ)言模型"驅(qū)動(dòng)的聊天機(jī)器人，OpenAI是一家與微軟密切相關(guān)的創(chuàng)業(yè)公司。

Its success prompted Google and other tech firms to release their own LLM-powered chatbots.
它的成功促使谷歌和其他科技公司發(fā)布他們自己的大型語(yǔ)言模型驅(qū)動(dòng)的聊天機(jī)器人。

Such systems can generate text and hold realistic conversations because they have been trained using trillions of words taken from the internet.
這類系統(tǒng)可以生成文本并進(jìn)行逼真的對(duì)話，因?yàn)樗鼈兪苓^(guò)了互聯(lián)網(wǎng)上數(shù)萬(wàn)億詞語(yǔ)的訓(xùn)練。

Training a large LLM takes months and costs tens of millions of dollars. This led to concerns that AI would be dominated by a few deep-pocketed firms.
訓(xùn)練一個(gè)大型LLM需要數(shù)月時(shí)間，花費(fèi)數(shù)千萬(wàn)美元。這導(dǎo)致人們擔(dān)心人工智能將被少數(shù)財(cái)力雄厚的大公司所主宰。

[Paragraph 3]
But that assumption is wrong, says the Google memo. It notes that researchers in the open-source community, using free, online resources, are now achieving results comparable to the biggest proprietary models.
但谷歌的備忘錄稱這種假設(shè)是錯(cuò)誤的。它指出，開(kāi)源社區(qū)的研究人員使用免費(fèi)的在線資源，他們?nèi)〉玫某晒F(xiàn)在可與最大的專有模型相媲美。

It turns out that LLMs can be “fine-tuned” using a technique called low-rank adaptation, or LoRa. This allows an existing LLM to be optimised for a particular task far more quickly and cheaply than training an LLM from scratch.
事實(shí)證明，LLM可以使用一種叫做“低秩適應(yīng)（即LoRa）”的技術(shù)進(jìn)行 "微調(diào)"。這使得現(xiàn)有的LLM能夠?yàn)槟骋惶囟ㄈ蝿?wù)進(jìn)行優(yōu)化，比從頭開(kāi)始訓(xùn)練 LLM 更快、成本更低。

[Paragraph 4]
Activity in open-source AI exploded in March, when LLaMA, a model created by Meta, Facebook’s parent, was leaked online.
開(kāi)源人工智能的活動(dòng)在3月爆發(fā)，由于當(dāng)時(shí)臉書(shū)母公司Meta創(chuàng)建的LLaMA模型遭泄露。

Although it is smaller than the largest LLMs (its smallest version has 7bn parameters, compared with 540bn for Google’s PaLM) it was quickly fine-tuned to produce results comparable to the original version of ChatGPT on some tasks.
雖然它比最大的LLM模型要?。ㄋ淖钚“姹居?0億個(gè)參數(shù)，而谷歌的PaLM模型有5400億個(gè)參數(shù)），但它很快就被微調(diào)了，在一些任務(wù)上產(chǎn)生的結(jié)果可與ChatGPT原始版本相媲美。

As open-source researchers built on each other’s work with LLaMA, “a tremendous outpouring of innovation followed,” the memo’s author writes.
備忘錄的作者寫道：隨著開(kāi)源研究人員在LLaMA的工作基礎(chǔ)上的相互合作迭代，"巨大的創(chuàng)新浪潮將接踵而至”。

[Paragraph 5]
This could have seismic implications for the industry’s future.
這可能對(duì)該行業(yè)的未來(lái)產(chǎn)生巨大影響。

“The barrier to entry for training and experimentation has dropped from the total output of a major research organisation to one person, an evening, and a beefy laptop,” the Google memo claims.
谷歌的備忘錄聲稱："訓(xùn)練和實(shí)驗(yàn)的門檻已經(jīng)從一個(gè)主要研究機(jī)構(gòu)的總產(chǎn)出下降到一個(gè)人、一個(gè)晚上和一臺(tái)強(qiáng)大的筆記本電腦上。"

An LLM can now be fine-tuned for $100 in a few hours. With its fast-moving, collaborative and low-cost model, “open-source has some significant advantages that we cannot replicate.”
LLM模型現(xiàn)在可以在幾個(gè)小時(shí)內(nèi)以100美元的價(jià)格進(jìn)行微調(diào)。由于其快速、協(xié)作和低成本的模式，"開(kāi)源有一些我們無(wú)法復(fù)制的巨大優(yōu)勢(shì)"。

Hence the memo’s title: this may mean Google has no defensive “moat” against open-source competitors. Nor, for that matter, does OpenAI.
因此，備忘錄的標(biāo)題是：這可能意味著谷歌沒(méi)有針對(duì)開(kāi)源競(jìng)爭(zhēng)對(duì)手的防御性“護(hù)城河”。就此而言，OpenAI 也沒(méi)有。

[Paragraph 6]
Not everyone agrees with this thesis. It is true that the internet runs on open-source software. But people use paid-for, proprietary software, from Adobe Photoshop to Microsoft Windows, as well.
并非所有人都同意這一論點(diǎn)?；ヂ?lián)網(wǎng)確實(shí)是在開(kāi)源軟件上運(yùn)行。但人們也使用付費(fèi)的專有軟件，如Adobe Photoshop，微軟Windows。

AI may find a similar balance. Moreover, benchmarking AI systems is notoriously hard.
人工智能可能會(huì)找到一個(gè)類似的平衡點(diǎn)。此外，對(duì)人工智能系統(tǒng)進(jìn)行基準(zhǔn)測(cè)試極其困難。

Yet even if the memo is partly right, the implication is that access to AI technology will be far more democratised than seemed possible even a year ago.
然而，即使該備忘錄部分正確，這也意味著人工智能技術(shù)的獲取將比一年前更加容易。

Powerful LLMs can be run on a laptop; anyone who wants to can now fine-tune their own AI.
強(qiáng)大的LLM模型可以在筆記本電腦上運(yùn)行；任何有需求的人現(xiàn)在都可以微調(diào)自己的AI。

[Paragraph 7]
This has both positive and negative implications.
這既有積極的影響，也有消極的影響。

On the plus side, it makes monopolistic control of AI by a handful of companies far less likely.
從積極的方面來(lái)說(shuō)，它使少數(shù)公司壟斷控制人工智能的可能性大大降低。

It will make access to AI much cheaper, accelerate innovation across the field and make it easier for researchers to analyse the behaviour of AI systems (their access to proprietary models was limited), boosting transparency and safety.
它將使獲得人工智能的成本大大降低，加速整個(gè)領(lǐng)域的創(chuàng)新，并使研究人員更容易分析人工智能系統(tǒng)的行為（他們對(duì)專有模型的訪問(wèn)是有限的），從而提高透明度和安全性。

But easier access to AI also means bad actors will be able to fine-tune systems for nefarious purposes, such as generating disinformation.
但是，更容易獲得人工智能也意味著壞蛋能夠微調(diào)系統(tǒng)以用于邪惡目的，例如生成虛假信息。

It means Western attempts to prevent hostile regimes from gaining access to powerful AI technology will fail.
這意味著西方將無(wú)法阻止敵對(duì)政權(quán)獲得強(qiáng)大的人工智能技術(shù)。

And it makes AI harder to regulate, because the genie is out of the bottle.
這讓 AI 更難監(jiān)管，因?yàn)榕硕嗬Ш写蜷_(kāi)會(huì)導(dǎo)致一發(fā)而不可收。

[Paragraph 8]
Whether Google and its ilk really have lost their moat in AI will soon become apparent.
谷歌及其同類公司是否真的在人工智能領(lǐng)域失去了護(hù)城河，答案即將揭曉。

But as with those previous memos, this feels like another turning point for computing.
但與之前的備忘錄一樣，這似乎是計(jì)算領(lǐng)域的另一個(gè)轉(zhuǎn)折點(diǎn)。

（恭喜讀完，本篇英語(yǔ)詞匯量758左右）
原文出自：2023年5月13日《The Economist》Leaders版塊

精讀筆記來(lái)源于：自由英語(yǔ)之路

本文翻譯整理: Irene

本文編輯校對(duì): Irene
僅供個(gè)人英語(yǔ)學(xué)習(xí)交流使用。

【補(bǔ)充資料】（來(lái)自于網(wǎng)絡(luò)）
隨機(jī)鸚鵡Stochastic Parrot 是一個(gè)由計(jì)算機(jī)科學(xué)家、AI研究者Douglas Eck在2019年創(chuàng)造的術(shù)語(yǔ)。該術(shù)語(yǔ)指的是一種基于隨機(jī)化技術(shù)生成音樂(lè)的算法，這種算法使用神經(jīng)網(wǎng)絡(luò)和概率模型，通過(guò)對(duì)音符、樂(lè)器、和聲等進(jìn)行多次隨機(jī)采樣來(lái)生成新的音樂(lè)作品。名字源自于一只鸚鵡，因?yàn)樗軌蚰７氯祟愓Z(yǔ)言中的聲音，就像這個(gè)算法可以模仿音樂(lè)家的風(fēng)格，從而生成新的音樂(lè)作品。

低秩適應(yīng)Low-rank adaptation是機(jī)器學(xué)習(xí)和信號(hào)處理領(lǐng)域中的一個(gè)術(shù)語(yǔ)，指的是一種將低秩矩陣適應(yīng)應(yīng)用于數(shù)據(jù)分析和處理的技術(shù)。在這個(gè)技術(shù)中，通過(guò)尋找數(shù)據(jù)或信號(hào)中的低秩結(jié)構(gòu)（即包含相似模式的結(jié)構(gòu)）來(lái)對(duì)其進(jìn)行建模和適應(yīng)。這種技術(shù)通常用于降噪、壓縮和特征提取等任務(wù)中。例如，在圖像處理中，低秩適應(yīng)可以幫助我們從大量的圖像數(shù)據(jù)中提取出共同的特征，并將其映射到低維空間，以實(shí)現(xiàn)更高效的數(shù)據(jù)存儲(chǔ)和處理。

人工智能系統(tǒng)基準(zhǔn)測(cè)試Benchmarking AI systems指的是一種對(duì)人工智能系統(tǒng)進(jìn)行性能評(píng)估和比較的過(guò)程。這個(gè)過(guò)程通常涉及到設(shè)計(jì)實(shí)驗(yàn)來(lái)測(cè)試不同AI模型在相同數(shù)據(jù)集上的表現(xiàn)，以便對(duì)它們的性能、準(zhǔn)確性、效率等進(jìn)行量化并進(jìn)行比較分析。通過(guò)人工智能系統(tǒng)基準(zhǔn)測(cè)試，我們可以更好地了解哪些AI算法最適合特定的任務(wù)/應(yīng)用，并根據(jù)這些結(jié)果對(duì)AI系統(tǒng)進(jìn)行改進(jìn)和優(yōu)化，可以幫助研究者和開(kāi)發(fā)人員更好地了解人工智能技術(shù)的潛力和限制。

【重點(diǎn)句子】（3個(gè)）
Training a large LLM takes months and costs tens of millions of dollars. This led to concerns that AI would be dominated by a few deep-pocketed firms.
訓(xùn)練一個(gè)大型LLM需要數(shù)月時(shí)間，花費(fèi)數(shù)千萬(wàn)美元。這導(dǎo)致人們擔(dān)心人工智能將被少數(shù)財(cái)力雄厚的大公司所主宰。

On the plus side, it makes monopolistic control of AI by a handful of companies far less likely.

從積極的方面來(lái)說(shuō)，它使少數(shù)公司壟斷控制人工智能的可能性大大降低。

But easier access to AI also means bad actors will be able to fine-tune systems for nefarious purposes, such as generating disinformation.
但是，更容易獲得人工智能也意味著壞蛋能夠微調(diào)系統(tǒng)以用于邪惡目的，例如生成虛假信息。