人工智能大戰(zhàn)如火如荼,大大小小的科技公司都在爭先恐后地推出人工智能驅(qū)動的產(chǎn)品。但這種新興技術(shù)所犯的錯誤越來越多,而谷歌即將推出的人工智能驅(qū)動聊天機器人,已經(jīng)使公司市值縮水了1,000億美元。
OpenAI在去年11月率先推出的ChatGPT聊天機器人,可使用人工智能生成類似于人類對話的文本,在競爭中領(lǐng)先一步。ChatGPT被廣泛應(yīng)用,被譽為史上增長最快的消費者應(yīng)用,并且有望顛覆互聯(lián)網(wǎng)搜索領(lǐng)域,這迫使競爭對手不得不放棄了人工智能研究領(lǐng)域多年來執(zhí)行的緩慢謹(jǐn)慎的策略。
微軟(Microsoft)、谷歌(Google)和中國搜索巨頭百度(Baidu)都在快速推出各自的人工智能產(chǎn)品。谷歌無疑是老牌公司中進(jìn)展最快的一家,周一,其人工智能驅(qū)動聊天機器人首次亮相。谷歌聊天機器人Bard由公司自己的人工智能驅(qū)動語言學(xué)習(xí)模型LaMDA驅(qū)動,該模型與ChatGPT采用的技術(shù)類似。
谷歌母公司Alphabet的CEO桑達(dá)爾·皮查伊寫道,LaMDA等技術(shù)最終將被整合到谷歌的搜索引擎當(dāng)中。
皮查伊上周一寫道,Bard將在幾周后對大眾公布,在此之前谷歌應(yīng)該投入更多時間持續(xù)完善該產(chǎn)品。皮查伊的博客文章中有一條推廣視頻,介紹了Bard的功能。視頻中,用戶要求Bard幫助處理日常活動,例如提供食譜建議和如何策劃迎接新生嬰兒的派對等,還有更多技術(shù)性問題,例如如何向一名9歲兒童解釋詹姆斯·韋伯空間望遠(yuǎn)鏡的發(fā)現(xiàn)。
但Bard給出的回答并不準(zhǔn)確,令觀眾感到失望,包括對詹姆斯·韋伯望遠(yuǎn)鏡問題的回答。它回答稱,韋伯望遠(yuǎn)鏡“首次拍攝了一顆太陽系外行星的照片”。
但實際上,據(jù)美國宇航局(NASA)記錄,這個系外行星的照片(該行星于2005年被確認(rèn)為系外行星),最早是在2004年,由位于智利的地面望遠(yuǎn)鏡陣列甚大望遠(yuǎn)鏡拍攝,遠(yuǎn)遠(yuǎn)早于2021年詹姆斯·韋伯望遠(yuǎn)鏡的發(fā)射時間。詹姆斯·韋伯望遠(yuǎn)鏡被用于系外行星的識別和分類。
路透社最先報道了Bard的錯誤,隨后谷歌股價暴跌。上周三下午,谷歌股價在盤中交易中下跌近8%,從周二收盤價每股108美元下跌到約99美元。谷歌上周三的市值為1.27萬億美元,較此前的1.35萬億美元大幅縮水。Bard的錯誤被發(fā)現(xiàn)后,谷歌隨即便在巴黎舉辦活動介紹Bard的更多功能。
一位谷歌發(fā)言人對《財富》雜志表示,Bard的錯誤凸顯出“嚴(yán)格測試過程的重要性,我們將通過可信測試者(Trusted Tester)計劃開始對其進(jìn)行測試。”發(fā)言人還表示,谷歌將采納各方面的反饋,保證在公開發(fā)布之前,Bard的回答符合“對現(xiàn)實世界信息在質(zhì)量、安全性和有理有據(jù)方面的高標(biāo)準(zhǔn)”。
Bard其實并不是唯一一個準(zhǔn)確性不足的聊天機器人。ChatGPT在多種情境下表現(xiàn)出種族和性別偏見,并且在被問及小眾問題時,給出了錯誤或令人迷惑的信息。(財富中文網(wǎng))
翻譯:劉進(jìn)龍
審校:汪皓
人工智能大戰(zhàn)如火如荼,大大小小的科技公司都在爭先恐后地推出人工智能驅(qū)動的產(chǎn)品。但這種新興技術(shù)所犯的錯誤越來越多,而谷歌即將推出的人工智能驅(qū)動聊天機器人,已經(jīng)使公司市值縮水了1,000億美元。
OpenAI在去年11月率先推出的ChatGPT聊天機器人,可使用人工智能生成類似于人類對話的文本,在競爭中領(lǐng)先一步。ChatGPT被廣泛應(yīng)用,被譽為史上增長最快的消費者應(yīng)用,并且有望顛覆互聯(lián)網(wǎng)搜索領(lǐng)域,這迫使競爭對手不得不放棄了人工智能研究領(lǐng)域多年來執(zhí)行的緩慢謹(jǐn)慎的策略。
微軟(Microsoft)、谷歌(Google)和中國搜索巨頭百度(Baidu)都在快速推出各自的人工智能產(chǎn)品。谷歌無疑是老牌公司中進(jìn)展最快的一家,周一,其人工智能驅(qū)動聊天機器人首次亮相。谷歌聊天機器人Bard由公司自己的人工智能驅(qū)動語言學(xué)習(xí)模型LaMDA驅(qū)動,該模型與ChatGPT采用的技術(shù)類似。
谷歌母公司Alphabet的CEO桑達(dá)爾·皮查伊寫道,LaMDA等技術(shù)最終將被整合到谷歌的搜索引擎當(dāng)中。
皮查伊上周一寫道,Bard將在幾周后對大眾公布,在此之前谷歌應(yīng)該投入更多時間持續(xù)完善該產(chǎn)品。皮查伊的博客文章中有一條推廣視頻,介紹了Bard的功能。視頻中,用戶要求Bard幫助處理日常活動,例如提供食譜建議和如何策劃迎接新生嬰兒的派對等,還有更多技術(shù)性問題,例如如何向一名9歲兒童解釋詹姆斯·韋伯空間望遠(yuǎn)鏡的發(fā)現(xiàn)。
但Bard給出的回答并不準(zhǔn)確,令觀眾感到失望,包括對詹姆斯·韋伯望遠(yuǎn)鏡問題的回答。它回答稱,韋伯望遠(yuǎn)鏡“首次拍攝了一顆太陽系外行星的照片”。
但實際上,據(jù)美國宇航局(NASA)記錄,這個系外行星的照片(該行星于2005年被確認(rèn)為系外行星),最早是在2004年,由位于智利的地面望遠(yuǎn)鏡陣列甚大望遠(yuǎn)鏡拍攝,遠(yuǎn)遠(yuǎn)早于2021年詹姆斯·韋伯望遠(yuǎn)鏡的發(fā)射時間。詹姆斯·韋伯望遠(yuǎn)鏡被用于系外行星的識別和分類。
路透社最先報道了Bard的錯誤,隨后谷歌股價暴跌。上周三下午,谷歌股價在盤中交易中下跌近8%,從周二收盤價每股108美元下跌到約99美元。谷歌上周三的市值為1.27萬億美元,較此前的1.35萬億美元大幅縮水。Bard的錯誤被發(fā)現(xiàn)后,谷歌隨即便在巴黎舉辦活動介紹Bard的更多功能。
一位谷歌發(fā)言人對《財富》雜志表示,Bard的錯誤凸顯出“嚴(yán)格測試過程的重要性,我們將通過可信測試者(Trusted Tester)計劃開始對其進(jìn)行測試。”發(fā)言人還表示,谷歌將采納各方面的反饋,保證在公開發(fā)布之前,Bard的回答符合“對現(xiàn)實世界信息在質(zhì)量、安全性和有理有據(jù)方面的高標(biāo)準(zhǔn)”。
Bard其實并不是唯一一個準(zhǔn)確性不足的聊天機器人。ChatGPT在多種情境下表現(xiàn)出種族和性別偏見,并且在被問及小眾問題時,給出了錯誤或令人迷惑的信息。(財富中文網(wǎng))
翻譯:劉進(jìn)龍
審校:汪皓
The artificial intelligence wars are heating up, with tech companies large and small racing to outdo one another and get their A.I.-powered product to market. But the mistakes by the nascent technology are starting to pile up, and Google’s up-and-coming A.I.-powered chatbot has already cost the company $100 billion in market value.
OpenAI has a head start in the fight with its ChatGPT chatbot launched in November that uses A.I. to generate humanlike text. ChatGPT’s wide range of uses, status as the fastest-growing consumer app in history, and potential to disrupt internet searches has pushed rivals to throw out the slow and cautious strategy that has dominated A.I. research for years.
Microsoft, Google, and Chinese search giant Baidu are all moving fast on their A.I. products. Google is arguably the furthest along among the established companies, premiering its own A.I.-powered chatbot for testers on Monday. Google’s chatbot, Bard, is powered by LaMDA, the company’s own A.I.-fueled language learning model that is similar to the technology behind ChatGPT.
Sundar Pichai, CEO of Google’s parent company, Alphabet, wrote that technologies like LaMDA will eventually be integrated with Google’s search engine.
Bard will be released to the general public in a few weeks, Pichai wrote Monday, and it may be a good idea for Google to spend more time perfecting it before that happens. Pichai’s blog post featured a promotional video to exhibit Bard’s capabilities. In it, users asked Bard for help with day-to-day activities, including recipe ideas and how to plan a baby shower, as well as more technical questions like how to explain to a 9-year-old what the James Webb Space Telescope has discovered.
But Bard underwhelmed audiences with inaccurate responses, including to the James Webb query. In that case, it responded that the telescope took the “very first pictures of a planet outside of our own solar system.”
But the very first image of an exoplanet was captured by the Very Large Telescope, a ground-based array in Chile, in 2004 and confirmed as an exoplanet in 2005, according to NASA—long before James Webb’s 2021 launch. James Webb, however, is being used to identify and catalog exoplanets.
Reuters was first to report on Bard’s mistake, and Google’s shares took a hit as a result. Google shares fell nearly 8% on Wednesday afternoon in midday trading, to around $99 a share, down from $108 at market close Tuesday. Its market cap Wednesday was $1.27 trillion, down from $1.35 trillion last week. The error was discovered just before Google hosted an event in Paris to display more of Bard’s capabilities.
Bard’s error underscores “the importance of a rigorous testing process, something that we’re kicking off this week with our Trusted Tester program,” a Google spokesperson told Fortune, adding that the company is incorporating feedback to ensure Bard’s responses meet “a high bar for quality, safety, and groundedness in real-world information” before its wider public release.
To be sure, Bard is far from the only chatbot to be inaccurate. ChatGPT has succumbed to racial and gender biases in certain scenarios while also providing incorrect or confused information when asked about niche topics.