第 398 期·中文精讀

馬克·扎克伯格:元宇宙中的首次訪談

Mark Zuckerberg: First Interview in the Metaverse

扎克伯格以虛擬形象接受採訪,談元宇宙的技術構想、VR與社交的未來,以及Meta的長期投入方向。

官方音訊00:00 / --:--
中文精讀由 AI 基於原始訪談改寫,非官方翻譯,可能存在偏差或錯誤;表述以原音訊與官方文稿為準。說明

Lex Fridman 坐在奧斯汀一間漆黑的房間裡,戴著 Quest 頭顯。馬克·扎克伯格(Mark Zuckerberg)在幾百英里外的另一間會議室,頭上也戴著一個。

房間裡再沒有第二個人。

但 Lex 能看清對方眼角那道細紋,能看見說話時眉毛抬起的幅度,能看見笑起來時嘴角先動的是不對稱的那一側。兩個人近到有點越界,Lex 為此專門道了個歉。扎克伯格回了一句:柔術都練過,之前互相鎖喉的距離比這遠多了。

這場對話沒有畫面傳輸。螢幕兩端傳的都不是影片畫面。對面坐著的是一個掃描出來的人,被實時驅動著,坐在你面前。

Lex 在開場說了一句很實在的話:這是我見過最不可思議的東西。他說這話的時候聲音有點抖,說自己的心跳在加速,說這幾乎讓他有點情緒化。

隔了半個美國,近得能看清皺紋

這套東西叫 Codec Avatar(編解碼器虛擬形象)。名字聽著像壓縮演算法,事實上也確實是。

流程分兩段。第一段是掃描:你走進一個佈滿相機陣列的棚裡,做出大量不同的表情,笑起來、皺眉、抬眉、說話、閉嘴。系統把這些表情全部記錄下來,為你的臉和身體建一個計算機模型,壓成一個編碼。

第二段是實時驅動。你戴上頭顯,頭顯上的相機盯著你的臉,捕捉此刻的表情,然後把這組引數編碼發過去。對面那頭收到引數,渲染出你現在的樣子。

「它傳輸的是一份編碼,說明你應該長什麼樣。所以除了照片級真實,它的頻寬效率其實比傳一整段影片高得多,更別說傳一個完整的三維沉浸式場景了。」(馬克·扎克伯格)

這句話裡藏著一個反直覺的事實:畫面越真實,反而越省流量。

影片傳的是畫素,幾百萬個畫素逐幀重新整理,一寸不落全部要發過去。Codec Avatar 傳的是「表情引數」,本質上是「眉毛上抬 2 毫米、左嘴角上揚 3 度」這種極簡指令,對面用本地算力把這張臉重新渲染出來。頻寬從畫素級掉到了引數級。

這也解釋了它為什麼能做到傳統視訊會議做不到的事。視訊會議裡,你看到的是一個被壓縮演算法糊過的、延遲半秒的、眼神不對焦的畫素堆。而這裡,你看到的是一個有立體感、有空間音訊、能和你保持眼神接觸的人。

恐怖谷是被皺紋填平的

Lex 在對話裡反覆提到一個詞:恐怖谷(uncanny valley)。那個「越像人越嚇人」的區間。

他的判斷是,這次跨過去了。

「完美不是帶來沉浸感的東西。是那些微小的瑕疵,雀斑、顏色的細微變化、皺紋、不對稱,還有眼角,你笑的時候眼睛會怎麼動。」(Lex Fridman)

扎克伯格接過話頭補了一個詞:不對稱(asymmetry)。

過去十年,做虛擬人的團隊都在追求「更準」,把臉做得更對稱、更光滑、更符合某種理想化的平均臉。結果是越做越假。

真實的活人臉從來不對稱。一邊眉毛比另一邊高,笑起來只有一側有酒窩,眼角的紋路左右深淺不同。人腦識別一張臉是不是真的,靠的恰恰是這些偏差。把偏差抹掉,等於把「人」的訊號抹掉了。

扎克伯格說,眼睛佔了很大比重。人在說話的時候,絕大部分資訊不在字詞裡,而在表情裡。你笑的時候眼角會不會擠出褶子,真笑和假笑的分界線就在這裡,這套機制在人腦裡跑了幾十萬年,騙不過去。

一個意外收穫:讓虛擬形象比本人更會表達

對話裡出現了一個岔路。

Lex 說,這期播客大概會被做成梗圖,標題就叫「兩個最面癱的人在元宇宙裡對談」。他承認別人常說他面無表情、像機器人,但他並非沒有情緒,只是表達幅度小,主要靠眼睛。

扎克伯格順著這個話頭講了一件很少有人公開聊的事。

「我經常因為表情比較僵硬被批評。我可能心裡挺高興的,但只是笑得很小。所以對我來說,我可能希望我的虛擬形象比我本人更能表達出我的感受。」(馬克·扎克伯格)

這句話的重量被低估了。

它的潛臺詞是:虛擬形象的第一版預設值會忠實還原你真實的表情幅度,但接下來一定會出現一個滑塊,用來調節「你想讓虛擬形象表達得多強烈」。有人會調到 100%,有人會調到 130%。

這就開啟了另一扇門:當一個人的數字形象比他本人更外向、更愛笑、更有感染力,那個形象還是他嗎?這個問題的答案,會在未來十年裡被幾億人各自給出不同的版本。

從坐幾小時掃描,到舉手機兩分鐘

現在這套東西沒法普及,卡點很具體:掃描太慢。

Lex 專程飛去匹茲堡做了完整掃描,軟硬體都極其複雜,但整個過程耗時很長。扎克伯格的說法是「如果你有幾個小時可以坐著」。這種產能,註定只能服務少數人。

Meta 正在做的專案,目標是把整個流程壓縮到兩三分鐘:你舉著手機在臉前晃一晃,說幾句話,做一串表情,結束。產出的質量和現在這個幾小時的版本一樣。

「這個難題還沒解決。現在只要有時間坐著,我們就能給你做掃描。但用非常高效的方式生產這些掃描,是我們還需要真正攻下來的最後幾塊拼圖之一。」(馬克·扎克伯格)

這裡的因果關係很值得看。Codec Avatar 卡住的地方既不在渲染,也不在算力,而在「資料採集的人口門檻」。限制一項技術變成產品的,常常是它最笨重的那部分能不能被砍掉;最驚豔的部分反倒容易。

扎克伯格強調「漸進式鋪開」,也是同一個邏輯:先讓更多人被掃描進系統,再把它接進各個 App,先用在會議、遠端協作這類剛需場景,再慢慢滲透到別的地方。

499 美元的頭顯,晶片比 1500 美元的還強

對話中途從元宇宙切到了硬體。Quest 3 是這期播客錄製的背景。

扎克伯格給了一組數字:Quest Pro 去年上市,售價 1500 美元。Quest 3 售價 500 美元,薄了 40%,混合現實的相機和感測器解析度提升了好幾倍,某些方面體驗甚至比 Pro 還好。

更值得看的是晶片。高通這一代面向 VR 和 MR 的晶片,Meta 參與了定製,但這款晶片今年才量產。Quest Pro 上市時用不上,Quest 3 用上了。

「所以說,Quest 3 雖然不是 Pro 產品線,但它裡面的晶片比 Pro 那一代更強,價格只有三分之一。」(馬克·扎克伯格)

這個錯位有產品節奏上的必然。誰先上市,誰就要用過時的晶片。 一個品類在快速迭代期,晚一年釋出的「低端款」經常打敗早一年的「高階款」。手機行業早年也這樣,現在不這樣了,因為手機已經沒什麼可迭代的了。

扎克伯格專門點了這一層:手機每年發新機,你幾乎看不出和上一代的區別。而頭顯每發一代,都帶著一個全新的能力。這一代是混合現實。

混合現實最先解決的問題:別把檯燈打碎

Lex 描述了他體驗 Quest 3 的場景:周圍坐著朋友,房間裡擺滿傢俱。你能看見沙發,能看見朋友,能看見朋友看著你揮胳膊時憋笑的表情。

扎克伯格講了一個更樸素的價值點。

「今天有很多 VR 體驗並不需要完全沉浸。你在玩射擊遊戲或者做健身,有時候人會擔心揮胳膊會不會打到檯燈、會不會撞到什麼。把這些放進混合現實裡,人就會舒服很多。」(馬克·扎克伯格)

行業宣傳裡,MR 的關鍵詞都是「把數字內容疊加到現實」,聽著很炫。但使用者第一次戴上之後真正在意的,是我能不能看見我家客廳。

Lex 補了一個畫面:殭屍從牆裡衝出來,你躲到自家真沙發後面去。房間裡沒有窗戶,你可以在牆上開一扇窗,窗外風景不錯,殭屍正朝你跑過來。

這種體驗的荒誕感裡藏著一個轉變:物理世界成了遊戲的一部分,不再是遊戲的敵人。

技術實現上並不輕鬆。頭顯外面有多顆相機,要從不同角度捕捉三維空間,然後把這些畫面重對映到你的視角。這裡有個細節:相機的位置不等於你眼睛的位置,每個人的瞳距、眼高都不同,所以需要實時做一次視角變換,還得做得看不出破綻。

手部追蹤是另一個被低估的進展。扎克伯格自己說,手部識別推進的速度超出了他的預期。Lex 提到了那個彈鋼琴的 demo:一張真實的桌子上,憑空出現琴鍵,手指落下去,聲音就出來,沒有手柄,沒有延遲。同時系統還追蹤著房間裡其他人的位置。

十年後,房間裡的全息物體和實物一樣多

扎克伯格做了一個思想實驗。

「可能到這個十年結束的時候,我們走進一個房間,看到的全息物體數量會和物理物體一樣多。」(馬克·扎克伯格)

他接著列了一份清單,哪些東西適合變成全息:

  • 所有的螢幕。任何一塊顯示屏都可以是全息投影
  • 所有的媒體。書、畫、影片、藝術品
  • 所有的遊戲。桌遊、卡牌、乒乓球

乒乓球的例子最能說明問題。想要一張球桌,打個響指它就出現在客廳;不想玩了,再打個響指它消失。物理世界裡,你需要一個房間、一筆錢、一次搬運、一次組裝,然後它會一直佔著那塊地。

全息版本的優勢,除了免費,更在可撤銷。物理世界的一切都帶著沉沒成本,你買了就得留著,擺了就得佔地方。數字物體沒有這個負擔。

再把這兩種能力疊起來:全息物體加上 Codec Avatar,會議就變成了這樣一幅畫面:你戴著眼鏡走進會議室,幾個同事物理在場,另外幾個以照片級真實的形態坐在旁邊,看起來就是他們本人,坐在那把椅子上。

會議室裡會坐著一個 AI 同事

順著這個邏輯,扎克伯格往前推了一步,這一步是很多人沒想過的。

「未來你開會的時候,可能你本人坐在那裡,另外幾個同事是全息形態,然後還有 Bob,一個 AI,他是你團隊裡的工程師,幫著處理一些事情,現在他也能以一個真實的虛擬形象被具身化,就這樣加入會議。」(馬克·扎克伯格)

把「AI 助手」和「虛擬形象」這兩件事接在一起,性質就變了。

現在的 AI 助手,互動介面是一個輸入框。你要打一段話,它回一段話,中間的往返有摩擦。當它有了臉、有了聲音、有了空間位置,它就從工具變成了在場者。你的大腦會自動把它當成一個人來對待,會記得它坐在哪,會在說話時下意識看向它。

這個變化的效率收益不小,問題也不少。當 AI 有了人形,人對它的信任閾值會大幅下調,很多本來會保持警惕的判斷,會因為一張臉而放鬆。

哪個版本才是你

對話在這裡拐進了一個很個人化的細節。

扎克伯格說,他那天早上戴上頭顯,發現虛擬形象裡的頭髮比現實中的短一點,該理髮了。他還說,他那天早上正好颳了鬍子,但就算沒刮,虛擬形象也依然是刮乾淨的樣子。

「所以會開始出現這些微妙的問題:這個虛擬形象是真實的,它大體上是你在掃描那個時點的樣子,但它不一定在時間上精確對應你此刻的樣子。」(馬克·扎克伯格)

這個細節看著小,往下想很麻煩。

Lex 把它推到了更狠的地方:很多人在掃描之前會減肥、弄頭髮、拿出最好看的狀態。那麼這個「數字的你」是一個被精心修飾過的版本,它會長期存在於別人的記憶裡,甚至比真實的你更常被看到。

扎克伯格的反應挺坦誠,他說人的肉體本來也在變,體重會波動,所以「哪個版本的你」這個問題在物理世界裡也存在。他補了一句觀察:很多人心裡認同的自己,可能是最好版本的那個,而日常狀態和最好版本、最差版本都對不上。

這段討論沒有結論,但它可能比整場對話裡任何一個技術引數都更值得琢磨。當一個人擁有了一個可控的、恆定的、可修飾的數字形象,他與自己的關係會被重新校準。

逝者的數字分身,和一條還沒劃完的線

Lex 提出了整場最沉重的一個問題。他說自己戴上頭顯之後,第一個念頭是想和愛的人聊天。第二個念頭是:那些已經不在的人呢。

父母、祖父母,如果他們已經去世,但還能以這個形態坐在你面前,和你說話。

扎克伯格的回答不算漂亮,但很實際。

「這中間大概有個平衡。如果一個人失去了親人、正在哀悼,能互動或者重溫某些記憶,可能是有幫助的。但它也可能走到一個不健康的地步。我不是這方面的專家,我們需要研究它、更細緻地理解它。」(馬克·扎克伯格)

他接著講了 Meta 已經在做的事。每天都有使用者去世,家屬往往想要回賬號,公司有一整套流程:一部分內容做成紀念狀態,賬號不會立刻消失;但另一部分內容堅決不給,比如私信。

這條線其實已經有先例:公開的表達可以保留,私密的交流不能移交。

麻煩的地方在於,Codec Avatar 會把這條線攪亂。一個能說會笑的數字分身,不只是一堆靜態內容,它會生成新的對話、新的表情、新的回應。那些回應從來不屬於逝者本人,但它們帶著逝者的臉和聲音。這已經超出了紀念的範疇,進入了續寫的範疇。

給創作者造一個賽博分身,比造新角色難十倍

Meta 最近放出了一批 AI 人格,可以在 WhatsApp、Messenger、Instagram 裡對話。扎克伯格把它們分了類。

第一類是純工具型,叫 Meta AI,沒有臉,比較中性、事實導向,負責查東西、答問題。

第二類是商業型。你想聯絡一家小店,店裡也不想專門僱人接電話,你也不想在電話裡等著。有個能自然對話的物件,幫你處理退換貨、訂位、下單,兩邊都省事。

第三類是娛樂型。扎克伯格提到內部測試裡最受歡迎的一個,是一個文字冒險類的「地下城主」(dungeon master)。他們還找了真人來扮演,Snoop Dogg 演地下城主。另外還有簡·奧斯汀(Jane Austen)和馬可·奧勒留(Marcus Aurelius)的版本。

「我們做了一個簡·奧斯汀,一個馬可·奧勒留,我很好奇會怎麼樣。」(馬克·扎克伯格)

這裡的技術分層,扎克伯格講得很清楚。

做一個全新角色,比做一個真人分身容易得多。因為新角色不需要對某個具體的人負責,Snoop Dogg 在這件事裡就是個演員,他演的是地下城主,不是他自己。

但要讓一個 AI 真正代表一個活著的創作者,難度完全是另一個量級。你要保證它不說創作者不想說的話,要保證它知道創作者知道的事,要保證它在超出邊界時能自己收住。

「你不是像寫普通程式那樣去程式設計它,你是在訓練它。它們是 AI 模型,不是普通計算機程式。但你要把它訓練得足夠可預測,這樣你才能給它設定一些邊界。」(馬克·扎克伯格)

這句話點出了大模型走向產品時最真實的困境。傳統軟體的行為是確定的,你可以審計每一行程式碼。模型的行為是機率的,你能做的只有「大機率不會越界」。所以創作者分身的釋出時間被推到了明年,而那些原創角色、助手、小商家客服,今年就會放出來。

扎克伯格還提到一個平臺,叫 AI Studio,目標是讓任何人都能像發一條普通 UGC 內容一樣,造一個自己的 AI。他的原話是,社群的創造力全開之前,這件事的潛力看不全。

Llama 2 開源:贏面大於風險

聊到開源,扎克伯格給了一個明確的覆盤。

「上次我們做播客的時候,正在爭論要不要開源 Llama 2。我很高興我們做了。在我看來,開源一個像 Llama 2 這樣的基礎模型,價值顯著大於風險。」(馬克·扎克伯格)

他提到團隊做了很嚴格的評估和紅隊測試。結果超出預期:下載量和使用量比他原本就樂觀的估計還要高。

關於 Llama 3,他的說法很剋制,沒有釋出時間。他的重心在於,Llama 2 本身不是消費品,它是基礎設施。現在的優先事項是把它和它的分支做成幾十億人能用的產品。

「就像我們當初圍繞 Llama 2 開源有過一場辯論,下一個版本我們也會需要類似的辯論和紅隊流程,確保它是安全的。我希望到時候也能開源。但這件事離做成還早,我們還在早期階段。」(馬克·扎克伯格)

這段話的資訊量在於「同樣的辯論會再來一次」。開源是一次次決策疊加出來的姿態,並非一次性動作,每個版本都要重新過一遍程式。這個程式本身,比任何一次具體結論都重要。

匿名,是數字世界最大的漏洞

Lex 問了一個社會層面的問題:物理世界裡違法的事,在元宇宙裡怎麼劃線。

扎克伯格的回答從一個小場景切入。

「我和朋友進到某個世界裡,我們乾的第一件事就是互相開槍。這在物理世界裡顯然不會做,因為會傷到對方。但在遊戲裡就是好玩。哪怕在遊戲大廳裡,這和遊戲本身都沒關係,就是一種搞笑的舉動。這有問題嗎?我不覺得,因為你在那個世界裡沒有造成傷害。」(馬克·扎克伯格)

他由此推出一個判斷框架:先看傷害。物理世界裡很多限制存在,是因為會造成實質傷害,數字世界裡這些限制可以放開。數字世界也有物理世界沒有的新傷害來源。

他挑了一個最要命的:匿名。

「你去餐廳吃飯,所有規範都建立在最後買單這件事上,因為你只有一個身份。你要是逃單,生活會懲罰你,這是一場重複博弈。但在數字世界裡,你可以匿名,可以換著身份出現,做一個好公民的動機就會弱很多,這會引發大量有害行為。這件事必須解決。」(馬克·扎克伯格)

這個類比很精準。現實社會的秩序,很大程度上不靠警察,靠的是「你還會再遇到這些人」。重複博弈讓守規矩變成理性選擇。匿名把這個前提抽走了。

元宇宙把這個問題放大了。在這裡,身份不只是一個使用者名稱,而是一張臉、一個聲音、一整套身體語言。偽造一張臉的難度,會決定這個空間的秩序成本。

扎克伯格沒有給出解法,只說了這件事必須被解決。

物理世界很重要,但「真實世界」是兩者之和

整場對話裡,扎克伯格講得最有哲學味的一段,是他對「真實」的重新定義。

「很多人把物理世界叫作真實世界。我越來越覺得,物理世界當然極其重要,但真實世界其實是物理世界和數字世界合在一起。在這項技術之前,這兩者是分開的。你透過一塊螢幕進入數字世界,螢幕可能是兜裡的小屏,也可能是桌上更大的屏。但它們本質上是割裂的。」(馬克·扎克伯格)

他給出的替代方案是:真實世界是兩者的合併。物理不能丟,因為人是物理的存在。數字內容不該被鎖在一塊螢幕裡。

他還補了一句關於身體的話,這句可能是他少見的袒露:

「人和人不該是坐著不動的。你和我都喜歡運動、武術這類東西,我們一直在動。人是物理的存在,不是一顆腦子加一個罐子。人的經驗是物理的。」(馬克·扎克伯格)

這段放在整場對話的語境裡很有意思。他在論證一個用頭顯才能進入的世界,理由卻是「人應該動起來」。邏輯上說得通:現在的計算裝置把人釘在椅子上,頭顯至少讓人站著、揮胳膊、在房間裡走動。

他也承認這套東西的當前形態很笨重。但他給了一個判斷:這會是迄今為止最像人、最社交的計算平臺。這個判斷對不對,要看未來五年 Quest 能不能瘦到一副眼鏡的樣子。他自己給的時間線是五年左右。

關於遠端工作,他談的是地理自由

對話裡有一段關於遠端辦公的推論,被大多數報道略過了。

「現在的爭論是遠端工作還是大家待在一起。這個東西能把我們帶到一種更接近的狀態:人在物理上分散各地,但感覺像是在一起。夢想是有一天人們可以想住哪就住哪,同時還能獲得所有機會,因為你能感覺自己和同事物理上在一起。」(馬克·扎克伯格)

他給「在一起」下了一個具體的功能定義:建立文化、建立關係、建立信任。這三件事在純文字的遠端協作裡損耗最大。視訊會議補了一部分,但補不了眼神、微表情、身體距離這些東西。

如果 Codec Avatar 真能把這幾項補回來,受影響的就不只是辦公方式,還有房價、城市人口分佈、家庭和工作的地理取捨。技術問題的下游,往往是一連串社會學問題。

Lex 在結尾處說了一段挺動人的話。他經歷過網際網路出現之前的年代,經歷過語音通訊、影片通訊,一直以為這條路上有個天花板。這次體驗讓他覺得,天花板可能不存在,物理和數字的融合才是終點。

他還說,那些抱怨「網際網路不是現實、你應該關掉手機去親近自然」的人,可能會喜歡這個東西。因為它感覺是真的,連瑕疵都是真的。

給讀者的幾件可操作的事

這場對話裡的技術細節,對大多數人不構成當下去做什麼的理由。但有幾件事,現在就能做,而且做了會有回報。

  • 做產品的人,去研究一下「頻寬逆向最佳化」這個模式。 Codec Avatar 的核心洞察是:傳引數比傳畫素省得多。任何需要實時傳輸的場景,先問一遍「我是不是在傳原始資料,而實際上傳指令就夠了」。
  • 做數字人、虛擬形象或任何擬人化介面的人,優先保留不對稱和不完美。 完美化處理是恐怖谷的成因之一。皺紋、色差、左右不對稱,這些是大腦判定「真」的訊號,不需要被修掉。
  • 創作者現在就該想清楚,你的 AI 分身邊界在哪裡。 扎克伯格明確說,訓練一個能代表真人的 AI,比造一個新角色難得多,卡點是可預測性。在技術成熟之前,先想清楚哪些話它絕對不能替你說。
  • 家裡有長輩或親人的,趁現在錄一段高質量的影片。 未來這類技術的第一批真實需求,會來自失去親人的人。今天的普通影片素材,可能就是明天的建模原料。
  • 關心開源模型走向的,把 Llama 系列當成一個「每版都要重新辯論一次」的過程來看。 一次開源決策代表不了長期承諾,下一個版本依然會走一遍紅隊和評估。關注這個流程,比關注某一次結論更有價值。
  • 做社群或平臺的人,把匿名機制當成一級問題。 扎克伯格自己承認,數字世界最大的漏洞是換身份的成本太低,導致重複博弈失效。任何社群治理方案,繞不開這一條。

訪談快結束時,Lex 說了一句很實在的話:我大概會一直用這個,用了就不想回去了。

扎克伯格回得更實在:我確實在這裡,我們只是不在同一個物理位置。你也不是在和一個 AI 說話。

那間會議室裡只有他一個人,那個漆黑的虛擬空間裡只有兩個人。但 Lex 說,他有好幾段時間完全忘了對面的人不在同一個房間。

技術成熟的標誌,往往是使用者已經忘了它的存在。

文稿來自 Lex Fridman 官網公開發布的官方 Transcript(含說話人與時間戳)。
Lex Fridman (00:00:00): The following is a conversation with Mark Zuckerberg, Inside the Metaverse. Mark and I are hundreds of miles apart from each other in physical space, but it feels like we’re in the same room because we appear to each other as photorealistic Kodak Avatars in 3D with spatial audio. This technology is incredible and I think it’s the future of how human beings connect to each other in a deeply meaningful way on the internet. These avatars can capture many of the nuances of facial expressions that we humans use to communicate and motion to each other. Now, I just need to work on upgrading my emotion expressing capabilities of the underlying human. This is the Lex Fridman Podcast. And now, dear friends, here’s Mark Zuckerberg. This is so great. Lighting change? Wow. Mark Zuckerberg (00:01:16): Yeah, we can put the light anywhere. Lex Fridman (00:01:19): And it doesn’t feel awkward to be really close to you. Mark Zuckerberg (00:01:22): No, it does. I actually moved you back a few feet before you got into the headset. You were right here. Lex Fridman (00:01:27): I don’t know if people can see this, but this is incredible. The realism here is just incredible. Where am I? Where are you, Mark? Where are we? Mark Zuckerberg (00:01:39): You’re in Austin, right? Lex Fridman (00:01:41): No. I mean this place. We’re shrouded by darkness with ultra realistic face, and it just feels like we’re in the same room. This is really the most incredible thing I’ve ever seen. And sorry to be in your personal space. We have done jujitsu before. Mark Zuckerberg (00:01:58): Yeah. I was commenting to the team before that I feel like we’ve choked each other from further distances than it feels like we are right now. Lex Fridman (00:02:08): This is just really incredible. I don’t know how to describe it with words. It really feels like we’re in the same room. Mark Zuckerberg (00:02:17): Yeah. Lex Fridman (00:02:17): It feels like the future. This is truly, truly incredible. I just wanted to take it in. I’m still getting used to it. It’s you, it’s really you, but you’re not here with me. You’re there wearing a headset and I’m wearing a headset. It’s really, really incredible. Can you describe what it takes currently for us to appear so photo realistic to each other? Mark Zuckerberg (00:02:44): Yeah. So, for background, we both did these scans for this research project that we have at Meta called Kodak Avatars. And the idea is that instead of our avatars being cartoony and instead of actually transmitting a video, what it does is we’ve scanned ourselves and a lot of different expressions, and we’ve built a computer model of each of our faces and bodies and the different expressions that we make and collapsed that into a Kodak that then when you have the headset on your head, it sees your face, it sees your expression, and it can basically send an encoded version of what you’re supposed to look like over the wire. So, in addition to being photorealistic, it’s also actually much more bandwidth efficient than transmitting a full video or especially a 3D immersive video of a whole scene like this. Lex Fridman (00:03:47): And it captures everything. To me, the subtleties of the human face, even the flaws, that’s all amazing. It makes it so much more immersive. It makes you realize that perfection isn’t the thing that leads to immersion. It’s the little subtle flaws like freckles and variations in color and just… Mark Zuckerberg (00:04:11): Wrinkles. Lex Fridman (00:04:12): … all stuff about noses. Mark Zuckerberg (00:04:12): Asymmetry. Lex Fridman (00:04:14): Yeah, asymmetry, and just the corners of the eyes, what your eyes do when you smile, all that kind of stuff. Mark Zuckerberg (00:04:20): Eyes are a huge part of it. Lex Fridman (00:04:22): It’s just incredible. Mark Zuckerberg (00:04:23): There’s all these studies that most of communication, even when people are speaking, is not actually the words that they’re saying. It’s the expression and all that. And we try to capture that with the classical expressive avatar system that we have. That’s the more cartoon designed one. You can put those expressions on those faces as well. But there’s obviously a certain realism that comes with delivering this photo realistic experience that, I don’t know, I just think it’s really magical. This gets to the core of what the vision around virtual and augmented reality is, of delivering a sense of presence as if you’re there together no matter where you actually are in the world. This experience I think is a good embodiment of that, where we’re in two completely different states halfway across the country, and it looks like you’re just sitting right in front of me. It’s pretty wild. Lex Fridman (00:05:17): Yeah. I’m almost getting emotional. It feels like a totally fundamentally new experience. For me to have these kinds of conversations with loved ones, it would just change everything. Maybe just to elaborate, I went to Pittsburgh and went through the whole scanning procedure, which has so much incredible technology, software and hardware, going on, but it is a lengthy process. So what’s your vision for the future of this in terms of making this more accessible to people? Mark Zuckerberg (00:05:50): It starts off with a small number of people doing these very detailed scans. That’s the version that you did and that I did, and before there were a lot of people who we’ve done this a scan for, we probably need to over collect expressions when we’re doing the scanning because we haven’t figured out how much we can reduce that down to a really streamlined process and extrapolate from the scans that have already been done. But the goal, and we have a project that’s working on this already, is just to do a very quick scan with your cell phone where you just take your phone, wave it in front of your face for a couple of minutes, say a few sentences, make a bunch of expressions, but, overall, have the whole process just be two to three minutes and then produce something that’s of the quality of what we have right now. Mark Zuckerberg (00:06:44): So I think that that’s one of the big challenges that remains, and right now we have the ability to do the scans if you have hours to sit for one. And with today’s technology, you’re using a Meta headset that exists. It’s a product that’s for sale now. You can drive these with that, but the production of these scans in a very efficient way is one of the last pieces that we still need to really nail. And then, obviously, there’s all the experiences around it. Right now we’re sitting in a dark room, which is familiar for your podcast, but I think part of the vision for this over time is not just having this be a video call. That’s fine, it’s cool, it feels like it’s immersive, but you can do a video call on your phone. Mark Zuckerberg (00:07:35): The thing that you can do in the Metaverse that is different from what you can do on a phone is doing stuff where you’re physically there together and participating in things together. And we could play games like this. We could have meetings like this in the future. Once you get mixed reality and augmented reality, we could have Kodak Avatars like this and go into a meeting and have some people physically there and have some people show up in this photorealistic form superimposed on the physical environment. Stuff like that is going to be super powerful. So we’ve got to still build out all those applications and the use cases around it. But I don’t know, I think it’s going to be a pretty wild next few years around this. Lex Fridman (00:08:17): I’m actually almost at a loss for words. This is just so incredible. This is truly incredible. I hope that people watching this can get a glimpse of how incredible it is. It really feels like we’re in the same room. I guess there’s an uncanny valley that seems to have been crossed here. It looks like you. Mark Zuckerberg (00:08:38): There’s still a bunch of tuning that I think we’ll want to do where different people emote to different extents, so I think one of the big questions is, when you smile, how wide is your smile? And how wide do you want your smile to be? And I think getting that to be tuned on a per person basis is going to be one of the things that we’re going to need to figure out. It’s like to, what extent do you want to give people control over that? Some people might prefer a version of themselves that’s more emotive in their avatar than their actual faces. So, for example, I always get a lot of critique and shit for having a relatively stiff expression. I might feel pretty happy, but just make a pretty small smile. Mark Zuckerberg (00:09:31): So maybe, for me, it’s like I’d want to have my avatar really be able to better express how I’m feeling than how I can do physically. So I think that there’s a question about how you want to tune that, but, overall, yeah, we want to start from the baseline of capturing how people actually emote and express themselves. And I think the initial version of this has been pretty impressive. And like you said, I do think we’re beyond the uncanny valley here where it does feel like you. It doesn’t feel weird or anything like that. Lex Fridman (00:10:05): That’s going to be the meme that the two most monotone people are in the Metaverse together, but I think that actually makes it more difficult. The amazing thing here is that the subtleties of the expression of the eyes, people say I’m monotone and emotionless, but I’m not. It’s just maybe my expression of emotion is more subtle, usually, with the eyes. And that’s one of the things I’ve noticed is just how expressive the subtle movement of the corners of the eyes are in terms of displaying happiness or boredom or all that stuff. Mark Zuckerberg (00:10:39): I am curious to see, just because I’ve never done one of these before, I’ve never done a podcast as one of these Kodak Avatars, and I’m curious to see what people think of it. Because one of the issues that we’ve had in some of the VR and mixed reality work is it tends to feel a lot more profound when you’re in it than the 2D videos capturing the experience. So I think that this one, because it’s photorealistic, may look as amazing in 2D for people watching it as it feels, I think, to be in it. But we’ve certainly had this issue where a lot of the other things, it’s like you feel the sense of immersion when you’re in it that, that doesn’t quite translate to a 2D screen. But I don’t know, I’m curious to see what people think. Lex Fridman (00:11:21): Yeah, I’m curious to see if people could see that my heart is actually beating fast now. This is super interesting that such intimacy of conversation could be achieved remotely. I don’t do remote podcasts for this reason, and this breaks all of that. This feels like just an incredible transition to something else, the different communication. It breaks all barriers, like geographic physical barriers. You mentioned, do you have a sense of timeline in terms of how many difficult things have to be solved to make this more accessible to like scanning with a smartphone? Mark Zuckerberg (00:12:02): Yeah. I think we’ll probably roll this out progressively over time. So it’s not going to be like we roll it out and one day everyone has a Kodak Avatar. We want to get more people scanned and into the system, and then we want to start integrating it into each one of our apps, making it so that I think that for a lot of the work style things, productivity, I think that this is going to make a ton of sense. In a lot of game environments, this could be fine, but games tend to have their own style where you almost want to fit more with the aesthetic style of the game. But I think for doing meetings, one of the things that we get a lot of feedback on Workrooms where people are pretty blown away by the experience and this feeling that you can be remote but feel like you’re physically there around a table with people, but then we get some feedback that people have a hard time with the fact that the avatars are so expressive and don’t feel as realistic in that environment. Mark Zuckerberg (00:12:58): So I think something like this could make a very big difference for those remote meetings. And especially with Quest Three coming out, which is going to be the first mainstream mixed reality product where you’re really taking digital expressions of either a person or objects and overlaying them on the physical world, I think the ability to do remote meetings and things like that where you’re just remote hang sessions with friends, I think that that’s going to be very exciting. So rolling it out over the next few years, it’s not ready to be a mainstream product yet, but we we’ll keep tuning it and keep getting more scans in there and rolling it out and into more of the features. But, yeah, definitely in the next few years you’ll be seeing a bunch more experiences like this. Lex Fridman (00:13:44): Yeah, I would love to see some celebrities scanned and some non-celebrities and just more people to experience this. I would love to see that. My mind blown. I’m literally at a loss for words because it’s very difficult to just convey how incredible this is, how I feel the emotion, how I feel the presence, how I feel the subtleties of the emotion in terms of work meetings or in terms of podcasts. This is awesome. I don’t even need your arms or legs. Mark Zuckerberg (00:14:17): Well, we got to get that. That’s its own challenge. Lex Fridman (00:14:22): Okay. Mark Zuckerberg (00:14:22): And part of the question is also, so you have the scan, then it takes a certain amount of compute to go drive that, both for the sensors on the headset and then rendering it. So one of the things that we’re working through is, what is the level of fidelity that is optimal? You could do the full body in a Kodak and that can be quite intensive, but one of the things that we’re thinking about is, all right, maybe you can stitch a somewhat lower fidelity version of your body and have the major movements, but your face is really the thing that we have the most resolution on in terms of being able to read and express emotions. Like you said, if you move your eyebrows a millimeter, that really changes the expression and what you’re emoting whereas moving your arm like an inch probably doesn’t matter quite as much. So, yes, I think that we do want to get all of that into here, and that’ll be some of the work over the next period as well. Lex Fridman (00:15:27): So you mentioned Quest Three. That’s coming out. I’ve gotten a chance to try that too. That’s awesome. How’d you pull off the mix? So it’s not just virtual reality, it’s mixed reality. Mark Zuckerberg (00:15:37): I think it’s going to be the first mainstream mixed reality device. Obviously, we shipped Quest Pro last year, but it was $1,500. And part of what I’m super proud of is we try to innovate not just on pushing the state-of-the-art and delivering new capabilities, but making it so it can be available to everyone. And we have this, and it’s coming out, it’s $500, and in some ways, I think the mixed reality is actually better in Quest Three than what we’re using right now in Quest Pro. And I’m really proud of the team for being able to deliver that kind of an innovation and get it out. But some of this is just software you tune over time and get to be better. Part of it is you put together a product and you figure out, what are the bottlenecks in terms of making it a good experience? Mark Zuckerberg (00:16:26): So we got the resolution for the mixed reality cameras and sensors to be multiple times better in Quest Three, and we just figured that, that made a very big difference when we saw the experience that we were able to put together for Quest Pro. And part of it is also that Qualcomm just came out with their next generation chip set for VR and MR that we worked with them on, on a custom version of it. But that was available this year for Quest Three and it wasn’t available in Quest Pro. So in a way, Quest Three, even though it’s not the Pro product, actually has a stronger chip set in it than the Pro line at a third of the cost. So I’m… Mark Zuckerberg (00:17:00): …line at a third of the cost. So I’m really excited to get this in people’s hands. It does all the VR stuff that Quest 2 and the others have done too. It does it better because the display is better and the chip is better. So you’ll get better graphics. It’s 40% thinner, so it’s more comfortable as well. But the MR is really the big capability shift. And part of what’s exciting about the whole space right now is this isn’t like smartphones, where companies put out a new smartphone every year, and you can almost barely tell the difference between that and the one the year before it. Mark Zuckerberg (00:17:36): For this, each time we put out a new headset, it has a major new capability. And the big one now is mixed reality. The ability to basically take digital representations of people or objects and superimpose them on the world. And basically, there’s a, one version of this is you’re going to have these augments or holograms and experiences that you can bring into your living room or a meeting space or office. Mark Zuckerberg (00:18:06): Another thing that I just think is going to be a much simpler innovation, is that there are a lot of VR experiences today that don’t need to be fully immersive. And if you’re playing a shooter game or you’re doing a fitness experience, sometimes people get worried about swinging their arms around, like, am I going to hit a lamp or something and am I going to run into something? So having that in mixed reality, actually, it’s just a lot more comfortable for people. You kind of still get the immersion and the 3D experience and you can have an experience that just wouldn’t be possible in the physical world alone. But by being anchored to and being able to see the physical world around you, it just feels so much safer and more secure. And I think a lot of people are really going to enjoy that too. So yeah, I’m really excited to see how people use it. But yeah, Quest 3 coming out later this fall. Lex Fridman (00:18:53): And I got to experience it with other people sitting around and there’s a lot of furniture. And so you get to see that furniture, you get to see those people, and you get to see those people enjoy the ridiculousness of you swinging your arms. I mean, presumably their friends of yours, even if they make fun of you, there’s a lot of love behind that and I got to experience that. So that’s a really fundamentally different experience than just pure VR with zombies coming out of walls and- Mark Zuckerberg (00:19:20): Yeah, it’s like someone shooting at you and you hide behind your real couch in order to duck the fire. Yeah. Lex Fridman (00:19:26): It’s incredible how it’s all integrated, but also subtle stuff, like in a room with no windows, you can add windows to it and you can look outside as the zombies run towards you, but it’s still a nice view outside. And so that’s pulled off by having cameras on the outside of the headset that do the pass through. That technology is incredible to do that on a small headset. Mark Zuckerberg (00:19:50): Yeah, and it’s not just the cameras. You basically, you need multiple cameras to capture the different angles and sort of the three-dimensional space, and then it’s a pretty complex compute problem, an AI problem to map that to your perspective because the cameras aren’t exactly where your eyes are because no two people’s eyes are, you’re not going to be in exactly the same place. You need to get that to line up and then do that basically in real time and then generate something that kind of feels natural and then superimpose whatever digital objects you want to put there. Mark Zuckerberg (00:20:24): So yeah, it’s a very interesting technical challenge and I think we’ll continue tuning this for the years to come as well. But I’m pretty excited to get this out because I think Quest 3 is going to be the first device like this that millions of people are going to get that’s mixed reality. And it’s only when you have millions of people using something that you start getting the whole developer community really starting to experiment and build stuff because now there are going to be people who actually use it. So I think we got some of that flywheel going with Quest Pro, but I think it’ll really get accelerated once Quest 3 gets out there. So yeah, I’m pretty excited about this one. Lex Fridman (00:21:01): Plus there’s hand tracking, so you don’t need to have a control, so the cameras aren’t just in the pass through of the entire physical reality around you. It’s also tracking the details of your hands in order to use that for gesture recognition, this kind of stuff. Mark Zuckerberg (00:21:17): Yeah, we’ve been able to get way further on hand recognition in a shorter period of time than I expected, so that’s been pretty cool. I don’t know, did you see the demo experience that we built around? Lex Fridman (00:21:29): Piano? Mark Zuckerberg (00:21:30): Yeah, the piano. Learning to play piano. Lex Fridman (00:21:32): Yeah, it’s incredible. You’re basically playing piano on a table, and that’s without any controller. And how well it matches physical reality with no latency, and it’s tracking your hands with no latency and it’s tracking all the people around you with no latency. Integrating physical reality and digital reality, obviously that connects exactly to this Codec Avatar, which is in parallel allows us to have ultra realistic copies of ourselves in this mixed reality. Lex Fridman (00:22:06): So it’s all converging towards an incredible digital experience in the Metaverse. To me, obviously I love the intimacy of conversation, so even this is awesome, but do you have other ideas of what this unlocks, of something like Codec Avatar unlocks in terms of applications, in terms of things we’re able to do? Mark Zuckerberg (00:22:28): Well, there’s what you can do with avatars overall, in terms of superimposing digital objects on the physical world, and then there’s psychologically, what does having photorealistic do? So I think we’re moving towards a world where we’re going to have something that looks like normal glasses, where you can see the physical world, but you’ll also see holograms. And in that world, I think that there are going to be, not too far off, maybe by the end of this decade, we’ll be living in a world where there are as many holograms when you walk into a room as there are physical objects. And it really raises this interesting question about what are… A lot of people have this phrase where they call the physical world the real world. Mark Zuckerberg (00:23:19): And I kind of think increasingly, the physical world is super important, but I actually think the real world is the combination of the physical world and the digital worlds coming together. But until this technology, they were sort of separate. It’s like you access the digital world through a screen and maybe it’s a small screen that you carry around or it’s a bigger screen when you sit down at your desk and strap in for a long session, but they’re kind of fundamentally divorced and disconnected. And I think part of what this technology is going to do is bring those together into a single coherent experience of what the modern real world is, which is, it’s got to be physical because we’re physical beings. So the physical world is always going to be super important. Mark Zuckerberg (00:24:01): But increasingly, I think a lot of the things that we kind of think of can be digital holograms. I mean, any screen that you have can be a hologram, any media, in any book, art. It can basically be just as effective as a hologram, as a physical object. Any game that you’re playing, a board game or any kind of physical game, cards, ping pong, things like that, they’re often a lot better as holograms. Because you could just snap your fingers and instantiate them and have them show up. It’s like you have a ping pong table show up in your living room, but then you can snap your fingers and have it be gone. So that’s super powerful. So I think that it’s actually an amazing thought experiment of like how many physical things we have today that could actually be better as interactive holograms. Mark Zuckerberg (00:24:52): But then beyond that, I think the most important thing obviously is people. So the ability to have these mixed hangouts, whether they’re social or meetings where you show up to a conference room, you’re wearing glasses or a headset in the very near term, but hopefully by over the next five years, glasses or so. And you’re there physically. Some people are there physically, but other people are just there as holograms and it feels like it’s them who are right there. Mark Zuckerberg (00:25:23): And also by the way, another thing that I think is going to be fascinating about being able to blend together the digital and physical worlds in this way, is we’re also going to be able to embody AIs as well. So I think you’ll also have meetings in the future where you’re basically, maybe you’re sitting there physically and then you have a couple of other people who are there as holograms, and then you have Bob, the AI, who’s an engineer on your team who’s helping with things, and he can now be embodied as a realistic avatar as well, and just join the meeting in that way. So I think that that’s going to be pretty compelling as well. Mark Zuckerberg (00:26:03): Okay, so what can you do with photorealistic avatars compared to the more expressive ones that we have today? Well, I think a lot of this actually comes down to acceptance of the technology. And because all of the stuff that we’re doing, I mean, the motion of your eyebrows, the motion of your eyes, the cheeks and all of that, there’s actually no reason why you couldn’t do that on an expressive avatar too. I mean, it wouldn’t look exactly like you, but you can make a cartoon version of yourself and still have it be almost as expressive. Mark Zuckerberg (00:26:38): But I do think that there’s this bridge between the current state of most of our interactions in the physical world and where we’re getting in the future with this kind of hybrid, physical and digital world, where I think it’s going to be a lot easier for people to take some of these experiences seriously with the photorealistic avatars to start. And then I’m actually really curious to see where it goes longer term. I could see a world where people stick to the photorealistic and maybe they modify them to make them a little bit more interesting, but maybe fundamentally, we like photorealistic things. Mark Zuckerberg (00:27:14): But I can also see a world that once people get used to the photorealistic avatars and they get used to these experiences, that I actually think that there could be a world where people actually prefer being able to express themselves in kind of non, ways that aren’t so tied to their physical reality. And so that’s one of the things that I’m really curious about. And I don’t know, in a bunch of our internal experiments on this, one of the things that I thought was psychologically pretty interesting is, people have no issues blending photorealistic stuff and not. Mark Zuckerberg (00:27:50): So for this specific scene that we’re in now, we happen to sort of be in a dark room. I think part of that aesthetic decision I think was based on the way you like to do your podcast, but we’ve done experiences like this, where you have a cartoony background, but photorealistic people who you’re talking to, and people just seem to just think that that is completely normal. It doesn’t bother you, it doesn’t feel like it’s weird. Mark Zuckerberg (00:28:21): Another thing that we have experienced with, is basically you have a photorealistic avatar that you’re talking to, and then right next to them you have an expressive kind of cartoon avatar. And that actually is pretty normal too. It’s not that weird to basically being interacting with different people in different modes like that. So I’m not sure, I think it’ll be an interesting question, to what extent these photorealistic avatars are a key part of just transitioning from being comfortable in the physical world to this kind of new, modern, real world that includes both the digital and physical, or if this is the long-term way that it stays?I think that there are going to be uses for both the expressive and the photorealistic over time. I just don’t know what the balance is going to be. Lex Fridman (00:29:08): Yeah. It’s a really good, interesting philosophical question, but to me, in the short term, the photorealistic is amazing. To where I would prefer, you said the workroom, but on a beach with a beer, just to see a buddy of mine remotely on a chair next to me, drinking a beer. I mean that, as realistic as possible, is an incredible experience. So I don’t want any fake hats on him. I don’t want any, just chilling with a friend, drinking beer, looking at the ocean, while not being in the same place together. I mean, that experience is just, it’s a fundamentally, it’s just a high quality experience of friendship. Whatever we seek in friendship, it seems to be present there in the same kind of realism I’m seeing right now. This is totally a game changer. So to me, this is, I can see myself sticking with this for a long time. Mark Zuckerberg (00:30:01): Yeah, and I mean it’s also, it’s novel. And it’s also a technological feat, right? It’s like being able to pull this off, it’s a pretty impressive and I think to some degree, it’s just this kind of awesome experience. Lex Fridman (00:30:15): But I’m already, sorry to interrupt, I’m already forgetting that you’re not real. This really- Mark Zuckerberg (00:30:22): Well, I am real. Mark Zuckerberg (00:30:23): It’s novel. Mark Zuckerberg (00:30:24): This is just an avatar version of me. Lex Fridman (00:30:26): That’s a deep philosophical question. Yes. Mark Zuckerberg (00:30:29): But here’s some of the… So I put this on this morning and I was like, “All right.” It’s like, okay, my hair is a little shorter in this than my physical hair is right now. I probably need to go get a haircut. And I actually, I did happen to shave this morning, but if I hadn’t, I could still have this photorealistic avatar that is more cleanly shaven, even if I’m a few days in, physically. So I do think that there are going to start to be these subtle questions that seep in where the avatar is realistic in the sense of, this is kind of what you looked like at the time of capture, but it’s not necessarily temporarily accurate to exactly what you look like in this moment. And I think that there are going to end up being a bunch of questions that come from that over time, that I think are going to be fascinating too. Lex Fridman (00:31:22): You mean just the nature of identity of who we are? You know how people do summer beach body? Where people will be, for the scan, they’ll try to lose some weight and look their best and sexiest with the nice hair and everything like that. It does raise the question of if a lot of people interacting with the digital version of ourselves, who are we really? Are we the entity driving the avatar or are we the avatar? Mark Zuckerberg (00:31:52): Well, I mean, I think our physical bodies also fluctuate and change over time too. So I think there’s a similar question of which version of that are we? And it’s an interesting identity question because all right, it’s like, I don’t know, it’s like weight fluctuates or things like that. I think most people don’t tend to think of themselves as the… Well, I don’t know. It’s an interesting psychological question. Maybe a lot of people do think about themselves as the kind of worst version, but I think a lot of people probably think about themselves as the best version. Mark Zuckerberg (00:32:26): And then it’s like what you are on a day-to-day basis doesn’t necessarily map to either of those. Yeah, there will definitely be a bunch of social scientists and folks will have to, and psychologists, really, there’s going to be a lot to understand about how our perception of ourselves and others has shifted from this. Lex Fridman (00:32:51): Well, this might be a bit of a complicated and a dark question, but one of the first feelings I had experiencing this is I would love to talk to loved ones. And the next question I have is I would love to talk to people who are no longer here that are loved ones. So if you look into the future, is that something you think about? Who people who pass away, but they can still exist in the metaverse, you could still have, talk to your father, talk to your grandfather and grandmother and a mother once they pass away. The power of that experience is one of the first things my mind jumped because it’s like, this is so real. Mark Zuckerberg (00:33:30): Yeah, I think that there are a lot of norms and things that people have to figure out around that. There’s probably some balance, where if someone has lost a loved one and is grieving, there may be ways in which being able to interact or relive certain memories could be helpful. But then there’s also probably an extent to which it could become unhealthy. And I mean, I’m not an expert in that, so I think we’d have to study that and understand it in more detail. We have a fair amount of experience- Mark Zuckerberg (00:34:00): … understand it in more detail. We have a fair amount of experience with how to handle death and identity, and people’s digital content through social media already, unfortunately. Unfortunately people who use our services die every day and their families often want to have access to their profiles and we have whole protocols that we go through where there are certain parts of it that we try to memorialize so that way the family can get access to it so that the account doesn’t just go away immediately. But then there are other things that are important, private things that person has. We’re not going to give the family access to someone’s messages, for example. Mark Zuckerberg (00:34:42): So I think that there’s some best practices, I think from the current digital world that will carry over. But I think that this will enable some different things. Another version of this is how this intersects with AIs because one of the things that we’re really focused on is we want the world to evolve in a way where there isn’t a single AI super intelligence, but where a lot of people are empowered by having AI tools to do their jobs and make their lives better. Mark Zuckerberg (00:35:19): And if you’re a creator and if you run a podcast like you do, then you have a big community of people who are super interested to talk to you. I know you’d love to cultivate that community and you interact with them online outside of the podcast as well. But I mean, there’s way more demand both to interact with you, and I’m sure you’d love to interact with the community more, but you just are limited by the number of hours in the day. Mark Zuckerberg (00:35:46): So at some point, I think making it so that you could build an AI version of yourself that could interact with people not after you die, but while you’re here to help people fulfill this desire to interact with you and your desire to build a community. And there’s a lot of interesting questions around that, and obviously, it’s not just in the metaverse. I think we’d want to make that work across all the messaging platforms, WhatsApp, and Messenger, and Instagram Direct. But there’s certainly a version of that where if you could have an avatar version of yourself in the metaverse that people can interact with, and you could define that sort of an AI version where people know that they’re interacting with an AI, that it’s not the physical version of you, but maybe that AI, even if they know it’s an AI, is the next best thing because they’re probably not going to necessarily all get to interact with you directly. Mark Zuckerberg (00:36:45): I think that could be a really compelling experience. There’s a lot of things that we need to get right about it that we’re not ready to release the version that a creator can build a version of themselves yet, but we’re starting to experiment with it in terms of releasing a number of AIs that people can interact with in different ways. I think that that is also just going to be a very powerful set of capabilities that people have over time. Lex Fridman (00:37:13): So you’ve made major strides in developing these early AI personalities with the idea where you can talk to them across the Meta apps and have interesting, unique kind of conversations. Can you describe your vision there and these early strides and what are some technical challenges there? Mark Zuckerberg (00:37:34): Yeah. So a lot of the vision comes from this idea that… I don’t think we necessarily want there to be one big super intelligence. We want to empower everyone to both have more fun, accomplish their business goals, just everything that they’re trying to do. We don’t tend to have one person that we work with on everything, and I don’t think in the future we’re going to have one AI that we work with. I think you’re going to want a variety of these. So there are a bunch of different uses. Some will be more assistant oriented. There’s a sort of the plain and simple one that we are building is called just Meta AI. It’s simple. You can chat with it in any of your Threads. It doesn’t have a face. Mark Zuckerberg (00:38:22): It’s just more vanilla and neutral and factual, but it can help you with a bunch of stuff. Then there are a bunch of cases that are more business oriented. So let’s say you want to contact a small business. Similarly, that business probably doesn’t want to have to staff someone to man the phones, and you probably don’t want to wait on the phone to talk to someone. But having someone who you can just talk to in a natural way who can help you if you’re having an issue with a product or if you want to make a reservation or if you want to buy something online, having the ability to do that and have a natural conversation rather than navigate some website or have to call someone and wait on hold think is going to be really good both for the businesses and for normal people who want to interact with businesses. Mark Zuckerberg (00:39:11): So I think stuff like that makes sense. Then there are going to be a bunch of use cases that I think are just fun. So I think people are going to… I think there will be AIs that I can tell jokes, so you can put them into chat thread with friends. I think a lot of this, because we’re like a social company. I mean we’re fundamentally around helping people connect in different ways. Part of what I’m excited about is how do you enable these kind of AIs to facilitate connection between two people or more, put them in a group chat, make the group chat more interesting around whatever your interests are, sports, fashion, trivia. Lex Fridman (00:39:51): Video games. I love the idea of playing. I think you mentioned Baldur’s Gate, an incredible game. Just having an AI that you play together with. I mean, that seems like a small thing, but it could deeply enrich the gaming experience. Mark Zuckerberg (00:40:08): I do think that AI will make the NPCs a lot better in games too. So that’s a separate thing that I’m pretty excited about. I mean, one of the AIs that we’ve built that just in our internal testing people have loved the most is an adventure text-based like a dungeon master. Lex Fridman (00:40:30): Nice. Mark Zuckerberg (00:40:31): I think, part of what has been fun, and we talked about this a bit, but we’ve gotten some real cultural figures to play a bunch of these folks and be the embodiment in the avatar of them. So Snoop Dogg is the dungeon master, which I think is just hilarious. Lex Fridman (00:40:48): Yes. In terms of the next steps of, you mentioned Snoop, to create a Snoop AI, so basically AI personality replica a copy… Or not a copy, maybe inspired by Snoop, what are some of the technical challenges of that? What does that experience look like for Snoop to be able to create that AI? Mark Zuckerberg (00:41:11): So starting off, creating new personas is easier because it doesn’t need to stick exactly to what that physical person would want, how they’d want to be represented. It’s like it’s just a new character that we created. So Snoop in that case, he’s basically an actor. He’s playing the Dungeon Master, but it’s not Snoop Dogg, it’s whoever the dungeon master is. If you want to actually make it so that you have an AI embodying a real creator, there’s a whole set of things that you need to do to make sure that that AI is not going to say things that the creator doesn’t want and that the AI is going to know things and be able to represent things in the way that the creator would want, the way that the creator would know. Mark Zuckerberg (00:42:06): So I think that it’s less of a question around having the avatar express them. I mean that I think where it’s like, well, we have our V1 of that that we’ll release soon after Connect. But that’ll get better over time. But a lot of this is really just about continuing to make the models for these AIs that they’re just more and more, I don’t know, you could say reliable or predictable in terms of what they’ll communicate so that way when you want to create the Lex assistant AI that your community can talk to. You don’t program them like normal computers, you’re training them. They’re AI models, not normal computer programs, but you want to get it to be predictable enough so that way you can set some parameters for it. Mark Zuckerberg (00:42:59): And even if it isn’t perfect all the time, you want it to generally be able to stay within those bounds. So that’s a lot of what I think we need to nail for the creators, and that’s why that one is actually a much harder problem, I think, than starting with new characters that you’re creating from scratch. So that one I think will probably start releasing sometime next year. Not this year, but experimenting with existing characters and the assistant, and games, and a bunch of different personalities and experimenting with some small businesses. I think that that stuff we’ll be ready to do this year. And we’re rolling it out basically right after Connect. Lex Fridman (00:43:42): Yeah. I’m deeply entertained by the possibility of me sitting down with myself and saying, “Hey, man, you need to stop the dad jokes or whatever.” Mark Zuckerberg (00:43:52): I think the idea of a podcast between you and AI assistant Lex podcast. Lex Fridman (00:43:59): I mean, there is just even the experience of an avatar, being able to freeze yourself like basically first mimic yourself, so everything you do, you get to see yourself do it. That’s a surreal experience. That feels like if I was an ape looking in a mirror for the first time, realizing, “Oh, that’s you.” But then freezing that and being able to look around like I’m looking at you, I don’t know how to put it into words, but it just feels like a fundamentally new experience. I’m seeing maybe color for the first time. I’m experiencing a new way of seeing the world for the first time because it’s physical reality, but it’s digital. And realizing that that’s possible, it’s blowing my mind. It’s just really exciting. Lex Fridman (00:44:50): I lived most of my life before the internet and experiencing the internet and experiencing voice communication, video communication. You think like, “Well, there’s a ceiling to this, but this is making me feel like there might not be, there might be that blend of physical reality and digital reality. That’s actually what the future is.” Mark Zuckerberg (00:45:12): Yeah, I think so. Lex Fridman (00:45:13): It’s a weird experience. It feels like the early days of a totally new way of living, and there’s a lot of people that kind of complain, “Well, the internet, that’s not reality. You need to turn all that off and go in nature.” But this feels like this will make those people happy. I feel like, because it feels real, the flaws in everything. Mark Zuckerberg (00:45:37): Yeah. Well, I mean, a big part of how we’re trying to design these new computing products is that they should be physical, but I think that’s a big part of the issue with computers and TVs and even phones is like, “Yeah, maybe you can interact with them in different places.” But they’re fundamentally like you’re sitting, you’re still. I mean, people are just not meant to be that way. I mean, I think you and I have this shared passion for sports and martial arts and doing stuff like that. We’re just moving around. It’s so much of what makes us people is like, you move around. You’re not just like a brain and a tank. It’s where the human experience is a physical one. Mark Zuckerberg (00:46:17): So it’s not just about having the immersive expression of the digital world, it’s about being able to really natively bring that together. I do really think that the real world is this mix of the physical and the digital. There’s too much digital at this point for it to just be siloed to a small screen, but the physical is too important. So you don’t want to just sit down all day long at a desk. I do think that this is the future. This is, I think the kind of philosophical way that I would want the world to work in the future is a much more coherently, blended, physical and digital world. Lex Fridman (00:46:56): There might be some difficult philosophical and ethical questions we have to figure out as a society. Maybe you can comment on this. So the metaverse seems to enable, sort of unlock a lot of experiences that we don’t have in the physical world. And the question is what is and isn’t allowed in the metaverse? In video games, we allow all kinds of crazy stuff. And in physical reality, a lot of that is illegal. So where’s that line? Where’s that gray area between video game and physical reality? Do you have a sense of that? Mark Zuckerberg (00:47:37): I mean, there are content policies and things like that in terms of what people are allowed to create, but I mean, a lot of the rules around physical… I think we try to have a society that is as free as possible, meaning that people can do as much of what they want unless you’re going to do damage to other people and infringe on their rights. And the idea of damage is somewhat different in a digital environment. Mark Zuckerberg (00:48:02): I mean, when I get into some world with my friends, the first thing we start doing is shooting each other, which obviously we would not do in the physical world because you’d hurt each other. But in a game, it’s just fun. And even in the lobby of a game, it’s not even bearing on the game, it’s just kind of a funny sort of humorous thing to do. So it’s like, is that problematic? I don’t think so because fundamentally you’re not causing harm in that world. So I think that part of the question that I think we need to figure out is what are the ways where things could have been harmful in the physical world that we’ll now be freed from that? And therefore there should be fewer restrictions in the digital world. Mark Zuckerberg (00:48:48): And then there might be new ways in which there could be harm in the digital world that there weren’t the case before. So there’s more anonymity. It’s when you show up to a restaurant or something, it’s like all the norms where you pay the bill at the end. It’s because you have one identity. And if you stiff them, then life is a repeat game and that’s not going to work out well for you. But in a digital world where you can be anonymous and show up in different ways, I think the incentive to act like a good citizen can be a lot less, and that causes a lot of issues and toxic behavior. So that needs to get sorted out. Mark Zuckerberg (00:49:28): So I think in terms of what is allowed, I think you want to just look at what are the damages, but then there’s also other things that are not related to harm, less about what should be allowed and more about what will be possible that are more about the laws of physics. It’s like if you wanted to travel to see me in person, you’d have to get on a plane, and that would take a few hours to get here. Whereas we could just jump in a conference room and put on these headsets and we’re basically teleported into a space where it feels like we’re together. Mark Zuckerberg (00:50:04): So that’s a very novel experience that it breaks down some things that previously would’ve defied the laws of physics for what it would take to get together. And I think that that will create a lot of new opportunities. One of the things that I’m curious about is there are all these debates right now about remote work or people being together. I think this gets us a lot closer to being able to work physically in different places, but actually have it feel like we’re together. So I think that the dream is that people will one day be able to just work wherever they want, but we’ll have all the same opportunities because you’ll be able to feel like you’re physically together. I think we’re not there today with just video conferencing and the basic technologies that we have, but I think part of the idea is that with something like this, over time, you could get closer to that and that would open up a lot of opportunities, right? Because then people could live physically where they want while still being able to get the benefits of being physically or feeling like you’re together. Mark Zuckerberg (00:51:00): … to get the benefits of being physically or feeling like you’re together with people at work, all the ways that that helps to build more culture and build better relationships and build trust, which I think are real issues that if you’re not seeing people in person ever. So yeah, I don’t know. I think it’s very hard from first principles to think about all the implications of a technology like this and all the good and the things that you need to mitigate. So you try to do your best to envision what things are going to be like and accentuate the things that they’re going to be awesome and hopefully mitigate some of the downside things. But the reality is that we’re going to be building this out one year at a time. It’s going to take a while, so we’re going to just get to see how it evolves and what developers and different folks do with it. Lex Fridman (00:51:52): If you could comment, this might be a bit of a very specific technical question, but Llama 2 is incredible. You’ve released it recently. There’s already been a lot of exciting developments around it. What’s your sense about its release and is there a Llama 3 in the future? Mark Zuckerberg (00:52:15): Yeah, I mean, I think on the last podcast that we did together, we were talking about the debate that we were having around open sourcing Llama 2. And I’m glad that we did. I think at this point there’s the value of open sourcing, a foundation model like Llama 2. It’s significantly greater than the risks in my view. I mean, we spent a lot of time, took a very rigorous assessment of that and red teaming it. But I’m very glad that we released Llama 2. I think the reception has been… It’s just been really exciting to see how excited people have been about it. It’s gotten way more downloads and usage than I would’ve even expected, and I was pretty optimistic about it. So that’s been great. Mark Zuckerberg (00:53:05): Llama 3, I mean, there’s always another model that we’re training. So for right now, we train Llama 2 and we released it as an open source model. And right now the priority is building that into a bunch of the consumer products, all the different AIs and a bunch of different products that we’re basically building as consumer products. Because Llama 2 by itself, it’s not a consumer product, right? It’s more of a piece of infrastructure that people could build things with. Mark Zuckerberg (00:53:36): So that’s been the big priority, is continuing to fine tune and just get Llama 2 and the branches that we’ve built off of it ready for consumer products that hopefully hundreds of millions of people will enjoy using those products in billions one day. But yeah, I mean we’re also working on the future foundation models. I don’t have anything new or news on that. I don’t know exactly when it’s going to be ready. I think just like we had a debate around Llama 2 and open sourcing it, I think we’ll need to have a similar debate and process to red team this and make sure that this is safe. And my hope is that we’ll be able to open source this next version when it’s ready too. But we’re not close to doing that this month. I mean, that’s a thing that we’re still somewhat early and working on. Lex Fridman (00:54:37): Well, in general, thank you so much for open sourcing Llama 2 and for being transparent about all the exciting developments around AI. I feel like that’s contributing to a really awesome conversation about where we go with AI. And obviously, it’s really interesting to see all the same kind of technology integrated into these personalized AI systems with the AI personas, which I think when you put in people’s hands and they get to have conversations with these AI personas, you get to see interesting failure cases where the things are dumb or they go into weird directions. And we get to learn as a society together what’s too far, what’s interesting, what’s fun, how much personalization is good, how much generic is good. And we get to learn all of this. And you probably don’t know this yourself. We have to all figure it out by using it, right? Mark Zuckerberg (00:55:31): Yeah, I mean part of what we’re trying to do with the initial AI’s launch is having a diversity of different use cases just so that people can try different things because I don’t know what’s going to work. I mean, are people going to like playing in the tech-based adventure games or are they going to like having a comedian who can add jokes to threads or they can want to interact with historical figures? We made one of Jane Austin and one of Marcus Aurelius, and I’m curious to see how that goes. Lex Fridman (00:56:03): I’m excited for both. Aa a big fan I’m excited for both. I have conversations with them. And I am also excited to see, the internet, I don’t know if you heard, can get kind of weird and I applaud them for it. So- Mark Zuckerberg (00:56:18): I’ve heard that, yeah. Lex Fridman (00:56:19): Yeah. So it’d be nice to see how weird they take it, what kind of memes are generated from this. And I think all of it is, especially in these early stages of development as we progress towards AGI, it’s good to learn by playing with those systems and interacting with them at a large scale like you said. Mark Zuckerberg (00:56:38): Yeah, totally. I mean, that’s why we’re starting out with a set. And then we’re also working on this platform that we call AI Studio that’s going to make it so that over time anyone will be able to create one of these AI almost like they create any other UGC content across the platform. So I’m excited about that. I think that to some degree we’re not going to see the full potential of this until you just have the full creativity of the whole community being able to build stuff, but there’s a lot of stuff that we need to get. So I’m excited to take this in stages. I don’t think anyone out there is really doing what we’re doing here. I think that there are people who are doing fictional or consumer-oriented character type stuff, but the extent to which we’re building it out with the avatars and expressiveness and making it so that they can interact across all of the different apps and they’ll have profiles and we’ll be able to engage people on Instagram and Facebook, I think it’s going to be really fun. Lex Fridman (00:57:49): Well, we’re talking about AI, but I’m still blown away this entire time that I’m talking to Mark Zuckerberg. And you’re not here, but you feel like you’re here. I’ve done quite a few intimate conversations with people alone in a room, and this feels like that. So I keep forgetting for long stretches of time that we’re not in the same room. And for me to imagine a future where I can with a snap of a finger do that with anyone in my life, the way we can just call right now and have this kind of shallow 2D experience, to have this experience like we’re sitting next to each other is like… I don’t think we can even imagine how that changes things where you can immediately have intimate one-on-one conversations with anyone. In a way, we might not even predict change civilization. Mark Zuckerberg (00:58:44): Well, I mean this is a lot of the thesis behind the whole Metaverse, is giving people the ability to feel like you’re present with someone. I mean, this is the main thing I talk about all the time, but I do think that there’s a lot to process about it. I mean, from my perspective, I’m definitely here. We’re just not physically in the same place. You’re not talking to an AI. So I think the thing that’s novel is the ability to convey through technology a sense of almost physical presence. So the thing that is not physically real is us being in the same physical place, but everything else is. And I think that that gets to this somewhat philosophical question about what is the nature of the modern real world? And I just think that it really is this combination of physical world and the presence that we feel, but also being able to combine that with this increasingly rich and powerful and capable digital world that we have and all of the innovation that’s getting created there. Mark Zuckerberg (00:59:52): So I think it’s super exciting because I mean, the digital world is just increasing in its capability and our ability to do awesome things, but the physical world is so profound, and that’s a lot of what makes us human is that we’re physical beings. So I don’t think we want to run away from that and just spend all day on a screen. It’s one of the reasons why I care so much about helping to shape and accelerate these future computing platforms. I just think this is so powerful. And even though the current version of this is like you’re wearing a headset, I just think this is going to be by far the most human and social computing platform that has ever existed. And that’s what makes me excited. Lex Fridman (01:00:36): Yeah, I think just to linger on this kind of changing nature of reality of what is real, maybe shifting it towards the sort of consciousness. So what is real is the subjective experience of a thing that makes it feel real versus necessarily being in the same physical space, because It feels like we’re in the same physical space. And that the conscious experience of it, that’s probably what is real. Not like that the space time, the physics of it. You’re basically breaking physics and focusing on the consciousness. That’s what’s real. Just whatever’s going on inside my head. Mark Zuckerberg (01:01:17): But there are a lot of social and psychological things that go along with that experience that was previously only physical presence, right? I think that there’s an intimacy, a trust. There’s a level of communication because so much of communication is nonverbal and is based on expressions that you’re sharing with someone when you’re in this kind of environment. And before, those things would’ve only been possible had I gotten on a plane and flown to Austin and sat physically with you in the same place. So I think we’re basically short cutting those laws of physics and delivering the social and psychological benefits of being able to be present and feel like you’re there with another person, which I are real benefits to anyone in the world. Mark Zuckerberg (01:02:10): Like you said, I think that is going to be a very profound thing. A lot of that is that’s the promise of the Metaverse and why I think that that’s the next frontier for what we’re working on. I started working on social networks when they were primarily text, where the first version of Facebook, your profile, you had one photo and the rest of it was lists of things that you were interested in. And then we kind of went through the period where we were doing photos. And now we’re kind of in the period where most of the content is video, but there’s a clear trend where over time the way that we want to express ourselves and kind of get insight and content about the world around us gets increasingly just richer and more vivid. Mark Zuckerberg (01:02:57): And I think the ability to be immersed and feel present with the people around you or the people who you care about is, from my perspective, clearly the next frontier. It just so happens that it’s incredibly technologically difficult. It requires building up these new computing platforms and completely new software stacks to deliver that, but I kind of feel like that’s what we’re here to do as a company. Lex Fridman (01:03:21): Well, I really love the connection you have through conversation. And so for me, this photo realism is really, really exciting. I’m really excited for this future and thank you for building it. Thanks to you and thanks to the amazing Meta teams that I’ve met, the engineers and just everybody I’ve met here. Thank you for helping to build this future. And thank you, Mark, for talking to me inside the Metaverse. This is blowing my mind. I can’t quite express. I would love to measure my heart rate this whole time. It would be hilarious if you’re actually sitting in a beach right now. Mark Zuckerberg (01:04:00): I’m not. I’m in a conference room. Lex Fridman (01:04:02): Okay. Well, I’m at a beach and not wearing any pants. I’m really sorry about that for anyone else who’s watching me in physical space. Anyway, thank you so much for talking today. This really blew my mind. It’s one of the most incredible experiences in my life, so thank you for giving that to me. Mark Zuckerberg (01:04:17): Awesome. Awesome. Glad you got to check it out. And it’s always fun to talk. All right, I’ll catch you soon. See you. Lex Fridman (01:04:23): See you later. This is so, so amazing, man. This is so-

本集資源與出處