Sam Altman在斯坦福大学的全面演讲解析:详细探讨GPT4设定的AI新标准和对GPT5的期待,以及人工智能的未来社会影响和商业模式变革
Summary
TLDRIn a recent YouTube channel feature, the host discusses a closed-door meeting at Stanford University where Sam Altman, a prominent Silicon Valley entrepreneur and the figurehead behind OpenAI, shared his insights. Altman, who famously dropped out of Stanford to pursue entrepreneurship, touched on several topics including the rapid development of artificial intelligence, the role of GPT-4 in setting industry standards, and the potential of GPT-5 to redefine those benchmarks. He also addressed the debate around open-source versus proprietary models, advocating for a free, ad-free GPT model to democratize AI services. Altman's vision includes using AI to eliminate inequalities and prepare society for the swift pace of technological change. He emphasized that while AI can augment human capabilities, it's crucial for society to evolve alongside these advancements to prevent chaos and ensure that the benefits of AI are accessible to all, albeit with the caveat that the actual participants in this innovation might remain a select group.
Takeaways
- 🚀 **Innovation in AI**: The rapid development of AI, particularly with GPT4 setting a new standard for the industry to follow and surpass.
- 🎉 **Celebrating Achievements**: Sam Altman, a prominent Silicon Valley entrepreneur, was celebrated for his birthday, highlighting his contributions to the tech industry.
- 📚 **Education to Entrepreneurship**: Altman's story parallels other successful entrepreneurs like Bill Gates and Mark Zuckerberg, who also left their studies at Stanford to pursue business ventures.
- 📈 **Scaling AI Models**: The effectiveness of the scaling law, where increasing model size and data lead to better performance, was emphasized with examples like the Llama and Tongyi models.
- 💡 **Open Source vs. Free Access**: OpenAI's stance on open sourcing their models was questioned, with the argument that providing free, ad-free access to AI might be more beneficial.
- 🌐 **Global AI Services**: Altman's vision for AI services to be cheap and widely available globally, potentially disrupting the current business models of AI companies.
- 💰 **Funding and Burning**: A discussion on the financial model of sustaining AI development, with the concept of 'burning money' to create societal value and finding funds to support this.
- 🛠️ **Innovation and Entrepreneurship**: Advice against targeting current AI deficiencies for innovation, as newer models like GPT5 are expected to obsolete current solutions.
- 🎮 **Convergence of Media**: Mention of Sora, a platform that aims to blend movies and games, suggesting new directions for创业 (entrepreneurship/innovation).
- 👑 **AI and Human Interaction**: The exploration of the relationship between humans and AI, and the societal roles they will each play, drawing examples from chess and mental health counseling.
- 🏛️ **Organizational Structure**: The importance of organizational structure being subservient to the company's vision and mission, with adaptability being key to achieving goals.
Q & A
What was the context of the meeting at Stanford University where Sam Altman spoke?
-Sam Altman spoke at a closed-door meeting at Stanford University, where he discussed various topics including AI development, the future of OpenAI, and the societal implications of rapid technological advancements.
What did Sam Altman mention about the development of AI and GPT models?
-Sam Altman mentioned that AI is developing rapidly, with GPT4 setting a new standard for the industry. He also hinted at the upcoming GPT5, suggesting it would be very powerful and set new benchmarks.
What was the significance of the DGX H200 gift to Sam Altman?
-The DGX H200 was likely a birthday gift to Sam Altman, who was celebrating his birthday around the time of the event. It's a high-performance AI computing system, indicating the respect and recognition for his work in the AI field.
What is Sam Altman's view on the open-source model for AI development?
-Sam Altman expressed that he does not believe open-source is the best path for AI development. Instead, he advocates for a model where AI services are freely available without ads, suggesting that this approach could be more effective in achieving widespread adoption and use.
How does Sam Altman envision the role of AI in reducing inequality?
-Altman believes that by making AI services widely available and inexpensive, it can help reduce inequality. However, the speaker of the transcript expresses skepticism about the ability of tools alone to eliminate social inequalities.
What did Sam Altman say about the future of AI and its impact on jobs?
-Altman suggested that while AI will surpass human capabilities in certain areas, such as playing chess, people will still enjoy human interactions and activities. He also mentioned that new AI tools will enable more significant innovations from humans, rather than stifling them.
What is the 'scaling law' that Sam Altman referred to?
-The 'scaling law' or 'Grading low' that Altman mentioned refers to the principle that as the scale of AI models increases, with more data and computational power, the performance and outcomes of these models continue to improve.
How did Sam Altman address the criticism regarding OpenAI's business model and funding?
-Altman indicated that the focus of OpenAI is on creating social value and achieving its vision of AGI (Artificial General Intelligence), rather than following traditional business logic. He suggested that they are willing to 'burn money' to achieve their goals and find funding to support their operations.
What was the birthday gift sent to Sam Altman by '老黄' (Huang)?
-The birthday gift sent to Sam Altman by '老黄' was a DGX H200, a high-end AI supercomputer, which is indicative of the close ties between the tech community and Altman.
What is the relationship between Sam Altman's educational background and his entrepreneurial journey?
-Sam Altman is a Stanford dropout, similar to other notable Silicon Valley entrepreneurs like Bill Gates and Mark Zuckerberg. His decision to leave Stanford to pursue entrepreneurship is part of a trend among tech founders who have chosen to forego formal education to start their companies.
How does Sam Altman view the future of AI products and services?
-Altman envisions a future where AI services are globally available, extremely cheap, and widely accessible. He also discusses the importance of society preparing for technological advancements to prevent chaos and ensure a smooth integration of AI into daily life.
Outlines
🎤 Introduction to Sam Altman's Stanford Talk
The video introduces Sam Altman's recent talk at Stanford University, discussing the content of a closed-door meeting where only snippets of information are available online. It mentions a gift sent to Altman, a DGX H200, and a group photo with Huang Renxun and Greg. The talk focused on AI development, with Altman emphasizing the rapid progress in the field, particularly with GPT-4 setting a new standard for the industry to follow. Altman also touched on his background as a Stanford dropout and the significance of the scaling laws in AI, suggesting that larger models with more data will continue to improve performance.
🤖 AI's Rapid Development and Open Source Debate
This paragraph delves into the advancements in AI, particularly the impact of models like Lama 3 and the upcoming GPT-5, which is expected to set new benchmarks. It contrasts the traditional business model of earning and reinvesting profits with Altman's approach of seeking funds to create social value. The paragraph also addresses the open-source debate, where Altman argues that providing free, ad-free AI services is more beneficial than open-source code, which is utilized by a smaller audience. Furthermore, it discusses the potential for AI to eliminate inequalities and the speaker's ambition to make AI services globally affordable and accessible.
🚀 Disrupting Traditional Business Models
The speaker challenges traditional business logic by focusing on finding funds to sustain operations rather than earning profits through conventional means. It reflects on the experience of dealing with Google's business practices and the realization that even a seemingly open company like Google has its rules and boundaries. The paragraph also warns against trying to fix current GPT models, as future iterations will naturally surpass them. It briefly mentions Sora, a platform that aims to integrate movies and games, and questions its viability based on OpenAI's unconventional commercial understanding.
🧲 AI and the Future of Entrepreneurship
The speaker advises against exploiting the weaknesses of current AI models, as they will soon be obsolete with the release of newer models like GPT-5. It emphasizes the importance of adhering to business principles despite OpenAI's disruptive approach. The paragraph also explores the relationship between humans and AI, using examples such as people's preference for watching human chess games and青少年 (youth) preferring AI therapists. It stresses the need to consider the roles and boundaries of humans and AI in product and service design.
🌐 Preparing Society for AI Advancements
The speaker discusses the social implications of AI, emphasizing that society must prepare for rapid technological progress. It acknowledges concerns about the speed of AI development and its potential to disrupt jobs, suggesting that society needs time to adapt. The paragraph also addresses the evolution of OpenAI's organizational structure to align with its vision and mission, and refutes criticisms by stating that the structure will continue to change as needed. Lastly, it asserts that new tools, like advanced AI, will not stifle human innovation but rather enable higher levels of creativity and innovation.
Mindmap
Keywords
💡Sam Altman
💡Artificial Intelligence (AI)
💡GPT-4
💡Scaling Laws (Grading low)
💡Open Source
💡AGI (Artificial General Intelligence)
💡
💡Innovation and Entrepreneurship
💡Sora
💡Commercialization
💡Social Impact of AI
💡Technological Unemployment
Highlights
Sam Altman, a prominent Silicon Valley entrepreneur, spoke at Stanford University, discussing various topics including AI development and its societal implications.
Altman mentioned that GPT-4 has set a new standard for AI, prompting other companies to catch up and potentially surpass it.
He touched on the Scalability Law, suggesting that as AI scales up with more data and computational power, performance will continue to improve.
Altman discussed the potential for GPT-5 to establish new benchmarks and standards in the field of AI.
He expressed skepticism about open-source as the best path for AI development, emphasizing the importance of creating value for society.
Altman suggested that OpenAI's goal is not to be open-source but to achieve AGI (Artificial General Intelligence).
He indicated that free and ad-free access to advanced AI models like GPT could be more beneficial than open-source code.
Altman shared his vision of making AI services extremely cheap and widely available globally, possibly leading to GPT-4 being offered for free.
He addressed the criticism regarding OpenAI's organizational structure and its mission, stating that the structure serves the vision and is subject to change.
Altman argued that society needs to prepare for rapid technological advancements to avoid chaos and ensure smooth integration.
He highlighted the importance of considering the relationship and boundaries between humans and AI in the design of future AI products and services.
Altman suggested that while AI has surpassed humans in games like chess, people still prefer to watch human players, indicating a continued desire for human interaction.
He pointed out that contrary to chess,青少年 (youth) prefer chatting with AI therapists, showing a shift in human-AI interaction preferences.
Altman warned against focusing on current GPT-4's shortcomings as future iterations like GPT-5 will render those efforts obsolete.
He emphasized that AI startups should adhere to established business principles, unlike OpenAI's disruptive approach.
Altman shared insights on the fusion of movies and games, hinting at new entrepreneurial directions with platforms like Sora.
He concluded by expressing optimism that new tools like advanced AI models will enable higher levels of human innovation, rather than stifling it.
Transcripts
大家好
歡迎收聽老範講故事的YouTube頻道
今天咱們來講一講
前幾天山貓奧特曼
到底在斯坦福大學裏說了些什麼
大家注意啊
這是一個閉門會議
我們現在所能夠看到的資訊
都是網上流傳的
各種隻言片語
有照片但是呢
並沒有看到全文
什麼視頻啊
我反而是沒看到了
以及根據這些隻言片語做的各種總結
4月25號上午呢
應該是黃教主考去送禮去了
送的是DGX H200
山貓奧特曼啊
黃仁勳以及Greg
他們三個人合了一張影
下午呢
就應該跑到斯坦福去做活動去了
斯坦福大學創業思想領袖講壇
這樣的一個活動
山姆奧特曼
號稱自己是這樣的一個領袖
應該沒有任何毛病了
上千人參加
而且呢還有給他的生日祝賀啊
說祝山姆奧特曼生日快樂
那你說
上午老黃是不是也去送生日禮物了呢
呃嚴格來說不算
但是稍微寬泛一點來說呢
問題不大
山姆奧特曼呢
是1985年4月22號生的
到4月25號呢
差個兩三天
給你補份生日禮物
補一份生日祝福
應該也算是應有之意吧
山毛特曼呢
算是一個非常根紅苗正的矽谷創業者
為什麼這麼講呢
他是在大二的時候
從斯坦福大學輟學的
比爾蓋茨輟學了啊
紮克伯格我印象裏好像也是輟學了
很多的創業者都輟學過
斯萊姆奧特曼斯坦福輟學啊
你輟學呢
就是聲明你能考得上斯坦福
否則你怎麼有機會從斯坦福輟學呢
讓了一半就突然創業
熱情熊熊燃燒
不能再念下去
了我要去創業
我原來也碰到過一位輟學創業的朋友
啊就是編程貓的創始人啊
他呢
是當時在法國一個學校裏面念碩士
也就是創新創業專業的一個碩士
到最後來
參加我們的這個創業比賽的時候說
哎呀再不創業就要畢業了啊
再不輟學就畢業了
當時講了這麼一句名言啊
我們記到現在
所以山伯奧特曼
是一個根紅苗正的斯坦福輟學創業者
那麼他講了幾件事呢
第一個
人工智慧的發展很快
說GPT4啊
已經指明了方向了
然後大家就去抄襲
抄襲是很容易的
他呢講的這個話並沒有說啊
穀歌抄襲了啊
還是蘋果抄襲了
而是什麼呢
GPT4啊他樹立了標準
樹立了標杆
然後呢各大廠商呢
就圍繞這些標準進行追趕
進行超越
這就是從GPT的發佈
以後的這一段時間裏頭
大家做的事情
你說我們現在做這麼多的模型測試
這個說
我今天在哪一個模型上超越GPT4了
明天在那個模型上說啊
我們在某一個特定場景裏超越GPT4了
那你說他們到底是怎麼幹的呢
其實都是有GPT4先出來了以後
然後再去在這個基礎上
設置克重的評測標準
PT4真正起到的作用呢
就是幫大家設立了一個標杆啊
你們按照GPT4的周邊
去設立評測標準啊
然後再去追趕
山奧特曼講了啊
說GPT4是一個很愚蠢的很尬的東西
你們現在不得不跟他去聊天
你說GPT5快出了嗎
GPT5會很強大啊
他已經講出這話了然
後講說這個縮放法則
也叫Grading low
依然沒有過時
依然有效
這個東西是什麼
就是隨著規模的擴大
數據的擴大
呃整個的性能或者結果
還是會繼續變好的
而人家會更嚴謹一些了啊
用我們的話講是什麼
就是當縮放法則依然成立的時候
就可以大力出奇跡
我可以堆更多的GPU
更多的數據進去
然後訓練出更好的模型
來跟大家講一個縮放法則起作用的
最現實的案例
比如說同1,000問
在拉瑪3出來以後
馬上又出了一個新的模型啊
原來統計千萬開放出來的模型是72B
但是拉瑪370B明顯的超越了
通1,000問72B
那麼通1,000問
應該就是在昨天還是前天
發了一個新模型
叫通1,000問110B
雖然我比拉瑪3的70B要大那麼一點點
但是呢
我的性能比他好啊
輸出的結果要比他強
各種的評測指標都比他好
再往後是什麼
再往後統一前面說了
我們準備出2.0
那麼在這個過程中
縮放法則到底是怎麼起作用的
你看拉瑪3出來了
70幣啊效果比我好了
那怎麼比我好的呢
拉瑪3
其實整個的訓練數據集是非常大的
就是他訓練70幣
雖然參數只有70幣
但是他數據集啊
要比現在咱們使用的正常模型都要大
非常多它的整個的訓練模型
數據品質都是很高的
它就可以得到70B
品質最好的這個開源模型
同1,000萬說不
我們即使不調整演算法
不調整數據品質
我只管增加模型參數
就可以達到拉瑪3
70幣的水準
他就出了一個110幣
而且拉瑪3說
我已經把該公開的公開了
指明方向了
你只要是用更多的數據去訓練
用更好品質的數據去訓練
你就可以訓練出像拉瑪這樣的模型來
同意千粉說
那我來了啊
我準備上通1,000問2.0
前面是我們在用的是通1,000問1.5啊
那麼2.0呢
大家可以想像
這一定會用更多的數據
更高質量的數據
就像拉瑪3那樣的數據
再去訓練
他就有可能在七十幾幣的這個模型下
可以超越拉瑪370幣
對吧
所以這個事情大家依然在同一個路上
再往前懟
或者你們管這個過程叫內卷都行
但是GPT5不一樣啊
GPT5上來以後會幹嘛
會樹立一大堆新的標杆
一大堆新的標準
在這些標準上
一開始就只有他一個人
然後再等著其他人慢慢往上追啊
這是人工智慧發展的很快啊
第二個講的是什麼呢
叫開源並不是一個好的途徑啊
因為很多人說哎
你open AI你怎麼不開源呢
特別是跟埃羅馬斯克吵了半天
埃羅馬斯克說
你應該把名字改成叫close AI
你不能叫open AI
這個肯定要回應一下了
那回應的方式是什麼呢
就是open AI的目標啊
不是開源
而是AGI叫通用人工智慧
讓機器可以像人一樣去思考問題
而我們並不認為
開源是更好的一個途徑
而是什麼呢
他認為免費
無廣告的GPT才是更好的途徑
因為什麼呢
開源了以後
真正能夠用開源代碼人
其實是很少的對吧
絕大部分人
是直接使用GPT去聊天的啊
而如果我能夠免費
無廣告的給大家用了
那麼這個不比開元效果更好嗎
他是這樣來去講這個事情
前面我去猜測過
如果GPT5出來以後
GPT4會不會免費
我覺得
從他現在講話的這個呃透的口風來說
這個可能性是存在的
而且很大
那麼GPT4一旦免費了
國內什麼文心藝言啊
這些付費的大模型
包括cloud 3這樣的模型
估計日子就不好過了
山姆奧特曼想要幹什麼呢
讓所有人都可以獲得人工智慧的服務
讓AI服務在全球範圍內
變得極其廉價和廣泛可用
我不知道這個全球範圍內
算不算中國啊
讓他立這個雄心壯志吧
那能咋辦呢
等哪天看山貓奧特曼到中國來
誰接見他吧
他應該沒來過吧
我印象裏是他沒來過啊
然後呢要消除不平等
現在很多做open AI
做AIGC的人
都在想去怎麼通過AI消除不平等啊
從這一點上來說呢
我個人表示不認可啊
為什麼我從來不認為工具的強大
可以消除不平等
工具越強大
不平等越厲害
搜索引擎出來的時候
大家就應該不會再上當受騙了
但實際上
搜索引擎對於各種的欺詐
一點碰觸都沒有
絕大部分人是不會去用的
哪怕搜索引擎是免費的啊
是每一個人都可用的
依然有很多人會去相信
各種街談巷議這樣的事情
所以我並不認為
AIGC就可以改變這個問題啊
那麼它如果免費的無廣告的
讓大家省差的GPT了
那這個錢的問題怎麼辦呢
啊他也講了
說燒錢啊
我就去燒啊
5億50億500億美金
我去燒去
無所謂啊
他覺得
我並不認為這是一個多大的事啊
那麼奧特曼也是很神奇啊
要只要我堅持下去
不斷的找錢
為社會創造價值就可以了
這個呢真的是顛覆很多人的認知
為什麼他不符合商業邏輯
他是一個反商業的過程
因為大家注意
這裏頭有一個很關鍵的一個字
叫什麼叫找啊
不斷找錢
穀歌是這麼幹的嗎
不是的微軟亞馬遜
任何一個互聯網巨頭或傳統巨頭
都不是這麼幹的
而現在他說我要去找錢
那麼找錢跟前面有什麼區別
前面的錢那叫賺錢
我賺了錢再去花
而且賺了錢花了以後呢
要量入而出
還要跟所有的股東去分紅
對吧這是傳統的商業模式啊
在互聯網之前
最早的商業模式是我造出一東西
你來買然後我去賺差價
或者我倒騰一東西
我從左手進右手出
我賺一差價
這是傳統的方式
互聯網呢
是已經在這個地方做了一個啊
進步了他是什麼呢
我分配流量啊
我來賺這個差價
賺這個流量費
對吧他找出一個新的到山茅
他班長說來哼
我燒錢創造社會價值
燒的錢我自己去找
最後怎麼能夠讓商業進行閉環
還是說整個商業的理論會再往前發展
讓整個的社會來為他去買單
這個事情啊
怎麼能夠讓他迴圈起來
我們還是要拭目以待這
個
確實是超出很多普通人的理解範疇了
關於創新和創業
他講了些什麼呢
啊他說
不要針對當前GPT4的缺陷
去做任何事情
在這塊去做專案的團隊
最後會被GPT5覆蓋掉的
甚至會被GPT6覆蓋掉
所以哪怕是GPT底層有什麼問題
你在這個基礎上做了些什麼東西修改
這事是不對的
跟大家講一小故事
就是我曾經有一次
跑去穀歌北京辦公室去吃早飯
當然目的不是吃早飯啊
目的是為了見穀歌大中華區
應該是香港那邊過來的
一位商業的負責人
當時在獵豹移動
給他去推薦獵豹移動相關的產品
希望能夠得到穀歌的一些扶持
啊他就問我啊
你們這個Cleanmaster是清理大師嘛
到底是清理記憶體啊
還是清理磁片啊
當然講的是英文啊
就是clean disc
oh clean memory
我們一堆人都去了
其他人沒聽懂啊
我的英文也很不好
但是我就突然反應過來了
這可能他們也在想
應該怎麼回答這個問題
其實clean master是兩個都清理的啊
但是我當時就靈機一動
一拍腦袋我就講了
我說我們只clean disc
絕對不碰memory
我們一起去的同事很詫異的看著我說
你為什麼會回答這樣的一個事情
我明明兩個都清理了
而且清理了記憶體才會有用戶體驗
才可以讓用戶覺得手機變快嘛
啊但是我就堅持說了
我們只清理硬碟
不清理記憶體
他們也不好意思來反駁我
畢竟是在外人面前嘛
哪怕是我說錯了
我在當時
級別也不高嘛
回去再收拾唄
但是呢這句話說對了
為什麼呢
穀歌的這位領導就跟我們講了
說你做的對
你就不應該去清理記憶體
那是安卓操作系統該做的事情
你不能鑒閱
如果你去清理記憶體
相當於是什麼
在安卓操作系統所控制的記憶體裏邊
去做了一些手腳
這是會對整個安卓操作系統
造成影響的
是會讓安卓系統的
整個的穩定性下降的
這事不該你管
安卓的硬碟管理是做的比較爛的
磁片管理比較爛
那麼你去把那那個清一清
這個是沒有問題的啊
所以呢當時不能叫龍顏大悅吧