2024-10-02
한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina
focus on
altman now uses chatgpt as a general tool, including writing emails and assisting with programming, and hopes that it will be able to handle more daily tasks in a few years.
altman admits that he is not particularly talented at writing, but sees writing as a valuable tool for clarifying ideas and helping communicate within the company.
if he were to write a book, altman said he would most like to write a book for those who want to understand ai, to help them understand ai and its impact on people's lives this year and in the next five years.
according to reports on october 2, it is well known that sam altman is the co-founder and ceo of openai, but what is less known is that he is also a passionate literary creator.
in the latest interview, altman rarely shared how he cleverly uses chatgpt to assist work and record notes in his daily life. at the same time, he also reviewed why he set his personal career goal on artificial general intelligence (agi), and looked forward to how large language models can reshape writing.
the following is the highlight of ultraman’s latest interview:
q: let’s start with large language models (llm). how do you think llm will reshape the way we write? have they quietly changed your understanding of the nature of writing and communication?
altman: i think many of us will write differently in the future. i'm not predicting that people will completely rely on llm ghostwriting, because one of the strangest things i find is that people will input certain bullet points into llm and have it generate a complete email and then send it to someone else, who in turn will summarize the main points. we cannot communicate with bullet points alone. but someone will develop a great tool (that may already exist) to write in a completely new way. this type of tool is not just a simple expansion of input points, but can lead us to explore the unknown in the vast field of thinking and inspire innovation. as computers are good at, they are a tool that helps you accomplish tasks that would otherwise be difficult to accomplish.
question: although the concept of "thinking tools" has been mentioned frequently for a long time, it is puzzling that today, the main way most people write is still to open traditional software such as microsoft word. it seems that the computer it has not become a true writing assistant as we expected, its role is closer to that of a typewriter.
altman: indeed, as an art, the essential charm of writing cannot be ignored. of course, there is still huge room for improvement to optimize this experience.
q: i've noticed an interesting trend, that is, when typing, we increasingly value the power and precision of words, and when it comes to output, through platforms like sora and dolly, we seem to be moving away from traditional text forms. what do you think of this phenomenon?
altman: i firmly believe that text will continue to play a vital role in our interaction with computers, programming, and even as the natural language interface we expect. this trend has long been foreseen by science fiction works, and the emergence of chatgpt undoubtedly marks a major revolution in this field. it allows us to communicate with computers in an unprecedentedly straightforward way and complete a variety of tasks with ease. of course, this is only one of many ways of interaction, and we will also witness multi-modal input and output in the future, but language, as our best communication tool, has an irreplaceable status.
q: what makes text unique is not only its searchability, but also its malleability and rigor.
ultraman: language is an indispensable cornerstone of humankind and its culture. it is difficult for people to imagine a world without language, because it is language that shapes our way of thinking and promotes the development of society and the progress of civilization.
q: when we are talking, i may not be able to accurately recall every word you said, but if it is transcribed into text, i can easily find and modify the words that touched me. so, how will technology like chatgpt impact the way we teach children to write?
altman: we don’t know yet what the writing process will look like in the future. but i believe that the basic teaching framework may remain relatively stable and not change too much. however, i think we will have new tools that will allow people to write in different ways and get more inspiration from them. the idea that "no one will ever learn to write again" is too one-sided, because that is not the true original intention of people writing. using chatgpt to write english compositions is convenient but difficult to touch the true meaning of writing. if this technology can help people stimulate deeper thinking and promote higher-quality thinking output when performing similar writing tasks, it will be an extremely positive and exciting development.
q: if we believe that a large part of the value of writing lies in clarifying our own thoughts, and we have new tools to do that better than we did before, that would be a huge win. one way i like to use chatgpt is: i have a clear idea, and if chatgpt disagrees with my point of view, once i have this idea and can clarify it to a certain extent, chatgpt can help me find examples and stories that can amplify and helped develop my original ideas.
altman: i try to observe how people in different areas of life use chatgpt, which has been very enlightening. recently, i saw two students using it to assist with homework. to be honest, one of the students basically input his own stuff and then let chatgpt generate the entire essay. i was shocked because i knew people were doing something like this in theory in significant numbers or otherwise, but to see someone do it and get a bad but acceptable essay, it really surprised me. i'd never seen anyone do this before, so it felt intuitive.
then i saw another person using chatgpt in a very different and more interactive way, trying to do something more like what you're talking about, which is having an idea but not being able to articulate it, and getting a little stuck in my thinking to shake me off dilemma and generate more ideas. the end result was far better than anyone could have done on their own, and i reflect on it all the time. the first example is a bad question like "what if just typing something out would get a super interesting or super actionable answer" and i think we're getting people to do the wrong thing. and if something makes them think about things in a different way and uses that tool to help them get to places they couldn't get to before, that's really fun.
q: how do you use chatgpt every day?
ultraman: i used to only use it for a few things, but as chatgpt has gotten better and better, i've learned how to use it more. so what's great now is that i'm really using it as a general tool. i hope in a few years, when you ask me about this, i will say i use it for most everything. every few months, i always discover new use cases and integration strategies, and the potential is endless. while it's not yet well integrated into most people's workflows, it's only going to get better.
q: when communicating with friends, if chatgpt is mentioned, which fields would you usually recommend them to use?
altman: what i hear most from my friends is that they most enjoy using chatgpt to help with computer programming, and many people say it has changed their lives. it was really comforting and interesting to hear that. there are other improvements as well. for example, people say it changes the way children learn, and teachers say it changes the way they teach. however, the medical applications of chatgpt are equally impressive, and people use it for creative work. however, assisted programming is the closest thing to me because a lot of my friends are programmers, so i hear about it all the time.
q: do you often use chatgpt to write emails?
altman: the emails i write are very short, often only seven words.
q: how does chatgpt help you write this kind of email?
altman: it's great at summarizing long emails, because i usually just skip most long emails without even reading them. but if you must read one, chatgpt's ability to effectively summarize long content, such as very long threads, is really impressive.
question: there is a famous saying that the essence of a masterpiece is often difficult to sum up in one sentence. perhaps gpt also has such characteristics and is difficult to define simply.
ultraman: one of the really interesting things here is that, in a way, it took me a few years to really understand this. but ilya (OpenAIformer chief scientist ilya sutskev always said that the real point of these models is compression, that we figure out how to compress as much knowledge as possible, and that's how we build artificial intelligence. compression is like a key to wisdom, i've been meditating on it for a long time and i'm sure i still don't fully understand it, but there's something deeper there.
q: i talked with your assistant and she said that you think clearly and don't speak much, but everything you say is very clear. you know exactly what you want and the information you convey is very accurate.
altman: i guess the reason this part resonates is because i really try to capture the essence of the problem, and i also really don't like it when other people are unclear.
q: did you ever want to be a novelist? this surprised me.
altman: i did think about it, but i only dreamed of becoming a novelist when i was young. i just enjoyed the romance of it, not because i thought i would be a good writer. it seemed like a cool thing to do, you know, like sitting in a parisian cafe and smoking. i probably wouldn’t have taken the path i’m on in life. turns out, i really wasn’t a good writer and i wasn’t going to be a blogger, but that’s okay. it's just that i'm still really happy with this attempt because i found that i could write for myself to clarify my thoughts, which worked really well. even if it's writing a letter explaining what a plan is and why we're doing it, i think it's often more persuasive if it's expressed in words rather than in a meeting.
q: if you were to write a book, what would it be about?
altman: many times, people will ask me, are there any important books on artificial intelligence ideas that i would recommend? i would think about it and tell them, no. so, i wanted to write this book for people who want to understand artificial intelligence. i might start with the historical context of other technological revolutions and explore why this one is similar and how it is different. i explain how the technology works and what is currently possible. this will involve predictions about the impact on your life this year, as well as what may happen in the next 5 years and what it means for all of us.
q: why did you choose to use written documents for internal coordination?
altman: if there are a lot of teams that need to reach consensus, i think written documentation is necessary, and we have a culture that values documentation in this regard. i think this is helpful for teamwork.
q: what have you learned in the process of running yc that you believe will have an impact on the way you run openai?
altman: yc encourages founders to be more ambitious and pursue things they believe in. i think this trait is also very important in the company openai.
q: why are you excited about using chatgpt for writing? what are some things you can’t currently do?
altman: the reason i'm excited about using chatgpt for writing is because i'm thinking about how to use it to make writing more efficient and less risky. for example, if i need to write a 10-page article, it still seems to me to be a challenge that requires a lot of energy and requires being in the right mood and focus for hours on end. it would be great if i could make this task feel like waiting for an uber in 15 minutes via chatgpt (although i haven't quite figured it out yet).
q: how does gpt reflect different personalities? i like to use it to rewrite a passage in the style of someone like tyler cowen (phd in economics, harvard university). how will future versions of gpt continue to develop this?
altman: a fair question is, how do we treat tyler cowen in this situation? we are working hard to resolve this issue. but what everyone agrees on is that gpt can show many different personalities from real people, which is a cool feature. you can ask chatgpt to reorganize content with different personalities, which i think is helpful for the creative process. as for what i'm most looking forward to about chatgpt and its future versions? i hope it becomes a tool that allows us to do things we couldn't do before, generate ideas we didn't think of before, and be more creative. this is the way forward in technology development and i believe will become a great use case.
q: creativity is no longer limited by skill, but by the ability to come up with the first idea.
ultraman: this will be a new field, and people will gradually master this ability like any new tool, thereby expanding human potential.
q: one of the things i admire so much about you is how thoughtful you are about the work you do. i'm curious how you decided to focus on artificial general intelligence (agi) and what the process was like for you envisioning yourself focusing on that goal.
altman: this decision can indeed be described as a “process”. at first, it may seem absurd, like a joke, or a goal that seems unattainable. now, working on agi seems to be my only obvious choice. in general, ideas are very fragile, and the best ideas are extremely fragile, but the most important thing is to find methods, pathways, or organizational structures that don't stifle these fragile but potentially great ideas. it depends on how you think, your decision-making process, and who you choose to work with. one toxic way of thinking to avoid is that some people are too smart and always find reasons why every great idea is bad. in the early stages, the most important thing is not to accidentally kill good ideas.
q: please share some insights about vulnerability and how writing helps you deal with it.
altman: the most important thing about writing for me is that it externalizes thinking and organization and makes it more explicit. writing can help clarify real ideas and help things become clearer, which for many people is not possible by thinking carefully alone. it's like thinking in your head for a long time. it's hard to hide those jumbled thoughts when you have to write them down and look at it.
q: in the early stages of openai, how did you think about future plans, especially in the process of focusing on this goal, what was the ultimate conclusion, and why do you think this move was necessary.
altman: we have a lot of ideas, and when we write them down, some of them may seem like something we’re going to do, but immediately find out that it doesn’t work. or maybe we figure it out, but when we look at it more closely, like talking to friends about building an agi, it’s another story. and then we started thinking about what this would look like as a complete, coherent plan, which would overturn some of the illusions. writing down these ideas helps us think about different things we could do and how we would build an organization. we considered joining a research lab at a university, etc. this process helps us eliminate some unfeasible ideas. this all seems obvious now, but it wasn't so then. this is a testament to me. then, if you write something that seems credible, you share it with others. they will go through the same process, rewriting, editing, but when they read it in black and white, they will have a different feeling. i believe it's important to get feedback from a broad range of people, especially on tough questions like "what should we do?"
q: when developing annual plans, do you tend to focus on one year, three years, or a longer-term time frame? is this process consistent across different time frames, or does it have its own unique considerations?
altman: i don’t strictly follow an annual planning rhythm, maybe more like every two years. i wrote a detailed document for openai called "our plan," which was gradually refined from an initial 25 pages to 15 pages, then to 4 pages, and i believe we can now condense it down to about half a page. undoubtedly a testament to the significant progress we have made in planning and clarity.
q: approximately how much time do you devote to writing every day?
altman: i am used to writing every weekend. these words are often shared with ten members of the team. the content is mostly about the concepts and strategies i have been thinking about, as well as the direction we should take. i’m currently working on a planned article that, in a rare move for me, explores how the world will evolve if artificial intelligence leads to prosperity, and why this vision is crucial. of course, this article will take some time to complete.
q: what insights do you have about the art of writing itself, and how artificial intelligence will reshape the field?
ultraman: in a future world where artificial intelligence can undertake many tasks, which skills will become increasingly valuable, and which ones may become relatively devalued? obviously, having excellent ideas and knowing clearly the goals you want artificial intelligence to help achieve will become particularly critical in an era where artificial intelligence is omnipotent. as the creator's unique taste and rich experience, as shown by masters such as graham, will become an immeasurable asset. i personally like to use chatgpt to assist writing, especially when i encounter creative bottlenecks, it seems to be a treasure trove of vocabulary. if i have trouble finding the right expression or am stuck on a certain idea, chatgpt can always provide strong support. however, it will never quickly replace the source of creativity. for writers, chatgpt is an amazing tool, but it is by no means a partner to create with, communicate with, or share creative tasks with.
q: in my daily use of chatgpt, i find that when i'm stuck on word choice, a simple request: "give me 10 words that are suitable for this sentence" and i will quote the sentence, it can quickly provide diverse choices. this feature is extremely effective for flexibility in text expression. in your opinion, as chatgpt becomes more popular, how should we adjust the way we train writers? does current writing education need corresponding changes?
ultraman: i once heard an interesting story. although i have not personally verified its authenticity, it is thought-provoking. in the story, a creative writing teacher took the initiative and asked his students to write the beginning of a novel on the first day of the new semester. students come with fallacies typical of novices and college students, such as overuse of metaphors and rhetoric. afterwards, the teacher led the students through a series of unique exercises, such as deleting one metaphor from each page and eliminating one redundant word from each sentence, until the originally bloated ten pages were reduced to one page.
as the class listened to these distilled works, a thoughtful discussion ensued that revealed the underlying truth: excessive rhetoric fails to create a compelling story. students often pursue the beauty of words too much, but ignore the reader's deep desire for stories. at the same time, on the other hand, those works that have been highly successful in the market, such as "twilight", have won widespread praise for their gripping storylines despite their plain writing. this makes people think: can we find a way to make excellent narrative and superb writing more harmonious and symbiotic? can it guide people to learn to make good use of these auxiliary tools to achieve a perfect integration of the two?
q: do you have any unique insights on how to improve chatgpt’s storytelling capabilities? suppose we turn on the voice mode of chatgpt and let the stories told by it be conveyed to the children through sound. how do you think chatgpt will perform compared with the stories told by the mother herself?
ultraman: at present, i think chatgpt is still immature in terms of storytelling, but its potential cannot be underestimated. we are in a stage of continuous optimization and improvement of the model, and there are still many areas worth exploring in terms of improving storytelling capabilities. if chatgpt can become more intelligent in the future and we conduct specialized training for its storytelling capabilities, its performance will undoubtedly be even better.
q: how do you plan to improve chatgpt’s storytelling capabilities?
altman: the key is to show it lots of examples of good and bad stories, and through comparative analysis, let the model learn and understand what makes a compelling narrative. we already understand this concept, but we have not yet put it into practice.
q: when you set out to write and strive to achieve a state of focus, what strategies do you use to effectively create this atmosphere?
altman: i used to think that finding a perfect environment, setting a specific time period, such as going to a coffee shop, and wearing noise-canceling headphones were necessary to get into a writing state. now, however, i realize that any 11 consecutive minutes, whether in the backseat of a car or in bed, can become precious creative time for me. of course, if conditions permit, a cup of fragrant coffee and unfettered time on saturday morning is undoubtedly the most ideal state. but when faced with long-form writing tasks, i will try to arrange such a prime time. however, in most cases, i prefer to use fragmented time to create.
q: i personally often use the voice function of chatgpt to organize my thoughts. i find this function to be extremely efficient for me because i tend to be more creative when speaking than typing.
ultraman: that's an interesting observation indeed. but for me, it's the exact opposite. i find that some ideas come to me while typing that may never come to mind when talking to someone. this once again proves that everyone’s creative habits and sources of inspiration are unique. therefore, it is crucial to find a rhythm that works for you. be inspired when interacting with others, followed by deep thinking, writing, or deep work when alone. i think this alternating pattern is key to improving creativity for many people, myself included. my daily rhythm is roughly the same: during the weekdays, i am busy in the office and have almost no time to stop and think. the pace of life is crazy and fulfilling; while on the weekends, i have a longer period of quiet time with few people around me. disturb. this cycle is of great significance to me.
q: when discussing the impact of agi (artificial general intelligence) on creative media, i deeply feel that the expressive power of written words is experiencing unprecedented improvement. especially with the advent of tools like sora, where we can now create videos, generate music and even images through text input, this has undoubtedly had a profound impact on our world.
altman: for me, writing has always been a tool for thinking, and i don’t think this core function will change in the future. so i think people still need to learn to write. just as traditional programming jobs may decrease in the future, the value of programming as an effective method of learning to think cannot be ignored, and learning programming is still of great significance.
q: many people are deeply concerned and even outraged by predictions that artificial intelligence may end writing. what is your view on this?
altman: i haven’t seen any conclusive signs that artificial intelligence is killing the art of writing. it is true that the internet is full of content generated by artificial intelligence of varying quality, including some poor student work that may have been produced by artificial intelligence, but in the end it is difficult to shake the foundation of human writing. i firmly believe that masters like graham will never give up writing because of worries about the rise of artificial intelligence.
unless artificial intelligence evolves to a stage of unimaginable super intelligence, i don’t think it will fully replace human writing. however, even if that day comes, we will face challenges far more severe than literary creation. assuming that there is a system in the future that can create works that surpass human levels, i still firmly believe that in 2027, the most popular novels among readers will still be written by humans. because, when i finish reading a great book, the first thing i do is i want to know about the author, i want to know their life story. i don’t think i’ll ever feel this way about ai writing.
to me, when you read a good book, you feel a connection with the author across time and space. even if you don’t really understand them, you seem to be able to resonate with them and feel the common experiences and emotions that belong to humans. this unique reading experience is the essence of the beauty of reading and cannot be replicated by artificial intelligence writing. therefore, i firmly believe that mankind’s love and pursuit of writing will continue forever. (tencent news special editor/jinlu)
collect and organize the latest news information and make it free for everyone to check.
