At the Sequoia AI Summit, Altman discussed the future of AI: AI agents will become practical in 2025, scientific discoveries in 2026, and robots will enter the real world in 2027. He reviewed OpenAI's evolution from the laboratory to ChatGPT and proposed the vision of a "core AI subscription".
Just now, Altman's speech at the Sequoia AI Summit has finally been fully "broadcast"!
This soul figure of OpenAI performed with ease in this information-dense speech, seemingly unaffected by the recent OpenAI restructuring and negotiations with Microsoft Microsoft.
In this speech, Altman predicted that AI will truly enter the real world within three years and completely change our lives.
Future AI models will be very compact but capable of containing your entire life experience.
- <> 2025,>, AI agents will truly become practical, especially in coding. 2>,, driving the next of economic growth.
- In 027, years AI will the enter the physical world to and create value.
Moreover, he also shared key insights about OpenAI's past, future AI products and business logic, and ways to maintain organizational efficiency:
- How ChatGPT transformed from a laboratory toy in 2016 to the most popular product from 2022 to<>AI product and business logic: Creating a "core AI subscription" accommodate" accommodating's experiences
Collaboration with Academic Community
Regarding customized cooperation with academic research projects, Altman said they have tried many approaches.
But what's even cooler is that the work of these researchers drives OpenAI to make models as intelligent, affordable, and widely accessible as possible, which is very beneficial to the academic community and the entire world.
"We often find that what researchers or users truly want is for OpenAI to comprehensively improve the performance of general models."
So this is where 90% of OpenAI's current efforts are focused - dedicated to enhancing model capabilities.
Programming Capabilities at the Core of OpenAI
Another aspect worth highlighting is programming.
An audience member asked whether programming is just another vertical application or if it is more important for OpenAI's future.
Altman said he believes programming is more crucial for OpenAI's future.
Programming will become a core method for these models - currently, if you ask ChatGPT to respond to a question, you'll get a text passage, perhaps an image.
But what you most want is a complete program.
You want each response to generate custom-tailored code, at least that's what I personally hope for.
You want these models to have the ability to react and act in reality.
And writing code, I believe, will be the core way to execute tasks in reality, call numerous APIs, or do other things.
Therefore, Altman believes programming will belong to a core category.
Model Customization
Regarding future model customization, will logging in with an OpenAI account allow bringing in each person's memories and historical context?
Altman described his ideal model state as:
A very compact inference model with trillion-level context Token capacity, sufficient to encompass your entire life experience.
This model will never be retrained.
Its parameters will never be customized.
But it can efficiently reason across your entire context.
All conversations in your life, every book you've read, every email, everything you've seen would be included, and connected with data obtained from other sources.
Your life will continuously add new content to this context.
Of course, there might be better models, but Altman says anything else is a compromise to this ideal form.
Prediction for 2025-2027
At the end of his talk, Altman made predictions about the AI development roadmap for 2025-2027.
AI Value Sources in the Next 12 Months
Altman pointed out that AI value creation will focus on three major areas: robust infrastructure, smarter models, and frameworks for technological integration into society.
He believes that by driving these areas, issues such as memory, safety, and real-world interaction protocols will be resolved.
Below, he elaborated on his views from a more granular perspective over the next three years.
2025: Rise of Intelligent Agents, Programming Leads
This year will be when AI agents become practical and take on work tasks.
Particularly in programming, AI will dominate related tasks, significantly improving efficiency.
2026: AI-Driven Scientific Breakthroughs
In 2026, AI will begin to independently or collaboratively achieve major scientific discoveries.
Advances in scientific knowledge will lay the foundation for sustainable economic growth.
2027: Robots Reshaping the Economy
By 2027, robots will gradually transform from experimental products to core entities creating economic value.
This interview outlined OpenAI's development from early exploration to becoming a leader in the AI field, along with Altman's profound insights into AI's future development.
The core points revolve around building more intelligent and user-friendly AI models, creating core AI subscription services, and empowering a broader ecosystem through APIs and future new protocols.
He emphasized programming's core position in AI's future while sharing valuable experiences about entrepreneurial resilience, challenges for large enterprises adapting to AI transformation, and research laboratory operating principles.
This interview not only showcased OpenAI's ambitions and strategic direction but also provided an important perspective on how AI technology will shape future society and economy.
References
https://x.com/sequoia/status/1922010755987067096
https://www.youtube.com/watch?v=ctcMA6chfDY
https://x.com/slow_developer/status/1922020202969006120
This article is from the WeChat public account "New Intelligence", authors: Ding Hui and Peach, published by 36Kr with authorization.
