Introduction
Today, we'll delve into the engaging conversation between OpenAI's Sam Altman and Snowflake's CEO Sridhar Ramaswamy at the Snowflake Summit 2025. This dialogue offers valuable updates on the AI industry. The conference was abuzz with energy, often described as a "data rock concert." The emergence of what's called Jam AI has the potential to revolutionize the way we interact with data and intelligence.
Key Figures in the Conversation
Sam Altman is a pivotal figure in promoting global AI evolution. From leading an open AI research institution aiming to impact billions of products worldwide, his views are always forward-looking. Sridhar Ramaswamy, on the other hand, is a data expert with deep knowledge in data and search fields. The collision of their ideas is highly significant for understanding AI's future trends.
Just Do It: Embracing AI Now
The Urgency of Adoption
The first question, a concern for many corporate leaders, is what to do now that AI is so popular. Altman's straightforward advice is "Just Do It." Despite the rapid changes in models and the temptation to wait for the next one or see how the market shakes out, the principle is clear: in a fast-changing technological landscape, companies with quick iteration speeds, low error costs, and high learning rates win. Hesitating or waiting for the next generation of models may pose a greater risk than making mistakes now.
Change in Altman's Perspective
Altman also shared that his advice to big companies has changed. A year ago, he might have been more cautious, suggesting they start with experiments. But this year, with the increased reliability of enterprise-level AI applications due to model improvements, he believes the technology has passed an important usability milestone. It's no longer just for fun or experimentation; it's a reliable tool in the production environment.
The Role of Curiosity
Ramaswamy added that curiosity is often neglected in this fast-paced era. With past experiences and knowledge becoming less applicable, an open mind and childlike curiosity are essential for exploring new possibilities. Moreover, the reduced cost of experiments makes it feasible to try things that were previously too costly or unthinkable.
Specific Steps for Companies
Companies should accept mistakes in less critical scenarios. They can establish a mechanism for quick learning and response, such as setting up a small team to develop and test AI application prototypes, creating feedback loops for front-line employees to report AI performance and problems, and fostering a culture that encourages experimentation and accepts failure. The key is to learn from mistakes quickly.
Reliable AI Capabilities
Current Capabilities
Now that the technology has matured, several capabilities are reliable for direct use. Models like ChatGPT can use network search to obtain the latest information, making their answers more relevant. Chatbots are also mature enough to handle both structured data (like database information) and unstructured data (such as emails and documents). This has direct value for enterprises, such as improving customer service efficiency and enabling employees to quickly access internal knowledge.
Altman's Prediction for Next Year
Altman predicts that next year, AI capabilities will go beyond automating business processes and building new products. Models will be able to handle problems that human teams can't solve on their own. Companies can invest significant computing power to have AI study major business issues, like redesigning critical projects or finding solutions independently. This requires AI to have strong reasoning and memory capabilities, and companies with prior experience will be better positioned to leverage these.
Memory and Unlocking: The Keys to Next-Gen AI
The Function of Unlocking (RAG)
Unlocking, or Retriever Augmented Generation (RAG), allows AI to answer or generate content based on facts and the latest information. OpenAI built a network search-scale system in early 2023 and integrated it into GPT, providing AI with a real-time update database and a basis for more accurate answers.
The Importance of Memory
Memory enables the system to recall past interactions and use that information to better handle similar problems in the future. This is crucial for achieving personalized AI experiences.
Why These Are Essential for Complex AI Applications
Future AI, especially agent applications, need to deal with more complex and long-term tasks. They require a deep understanding of various factors, including task goals, user preferences, historical operations, and real-time external information. Memory and unlocking provide the necessary context for these systems to perform better.
The Rise of AI Agents
The AGI Feeling with Codex
Altman was particularly excited when talking about Codex, the AI model for writing code. He described it as an AGI moment. Codex can handle a lot of tasks in the background, connect to GitHub, and perform complex long-term operations, similar to a beginner programmer. This autonomy, complexity, and long-term task-handling ability make Altman feel it touches the essence of AGI.
Current and Future Uses of AI Agents
Some companies are already using agents to automate customer service and outbound sales, which are repetitive cognitive tasks. Currently, humans still play a role in assigning work, assessing quality, and providing feedback, similar to managing a team of junior employees. Altman predicts that next year, agents will start to help discover new knowledge and solve difficult business problems, expanding from known repetitive tasks to more complex and unknown areas.
The Concept of AGI
Altman's View on AGI
Altman believes the term AGI is not very important. People have different definitions, and these definitions can change over time. He gives the example of showing today's ChatGPT to people five years ago; they would likely have considered it AGI, but now we're used to it. What's important is the trend of AI development. He also mentioned a possible standard for AGI: AI that can independently discover new science or significantly accelerate human scientific discovery.
Ramaswamy's View on AGI
Ramaswamy sees the definition of AGI as a philosophical question, similar to asking if a submarine can swim. It depends on how you define the terms. He believes that even if AI surpasses humans in some aspects, human activities and values will still exist, using the examples of chess and Go where humans still enjoy playing despite AI's superiority.
Consensus on AGI
Both CEOs agree that the exact moment or definition of AGI is not as important as the development and progress of AI. The focus should be on the capabilities and how they can be applied.
The Next-Gen Model
Capabilities of the Next-Gen Model
Altman described the next-generation models as amazing. They will enable companies to do things that the previous generation couldn't. The core capabilities include understanding a vast amount of context, connecting to various tools and systems, and performing excellent and flexible reasoning, allowing them to work independently.
The Ideal Model Framework
Altman proposed an ideal model framework: a tiny model with superhuman reasoning capabilities that can run extremely fast, handle one trillion tokens of context, and access all possible tools. The key is to view the model as a reasoning engine rather than a huge database, as knowledge and data can be integrated dynamically through context and tools.
What to Do with a Thousand Times Computing Power
Altman's Answers
When asked what he would do with a thousand times the computing power, Altman gave two answers. The "far-fetched" one is to invest in AI research to create a better model and then ask that model what to do, based on the belief in the exponential growth of AI capabilities. The more realistic answer is based on the current observation that existing models show a return on test time. Companies should allocate computing power to the most difficult and valuable problems, as the return may be non-linear.
Ramaswamy's Answer
Ramaswamy's answer focuses on human welfare. He mentioned a Snowflake project studying RNA expression, which is crucial for understanding and treating diseases. If he had a thousand times the computing power, he would invest it in this project, hoping to make a breakthrough that could solve many diseases and advance human civilization.
Conclusion
This dialogue between Altman and Ramaswamy highlights the path to AI-accelerated development and its potential to deeply change the world. After hearing about the future possibilities of AI, it's worth thinking about what the most difficult problems are in our personal lives or work that we would want AI to solve. If you enjoyed this discussion, don't forget to like, share, and subscribe for more tech news and in-depth analysis. See you next time.
{# 
{# 
{# 
{# 
{#