OpenAI plans to release a very powerful open-source model.
It will enable people to run extremely powerful models locally, redefining the possibilities of "local deployment".
Altman announced this news during a dialogue at the AI startup school in San Francisco, hosted by Y Combinator's CEO Garry Tan.
During the discussion, they also talked about OpenAI's development process, future direction, the origins of ChatGPT, and AI hardware construction.
Altman mentioned that the o3 model's running cost last week was 5 times that of this week, with a continuing price decline trend, and API costs will continue to drop significantly, making open-source models excellent. The memory function of ChatGPT is not just a small piece of hardware, but more like an AI companion.
In the future, GPT-5 and various reasoning models will develop towards integrated models, connecting with AI intelligent terminals and robots to become people's life operating systems.
He also said that if users register for the highest level of ChatGPT subscription, they will be given a free robot.
The following is a compilation without changing the original meaning.
The Future of GPT Reasoning Models
Tan: What surprised you most about the latest o3 model, and what emerging behaviors or use cases have impressed you?
Altman: I think we are in a very interesting era. Although we haven't yet seen reasoning model products reaching a new level of innovation, the model capabilities of this world have entered a new domain, and we still have a lot of entirely new things to build.
We will soon launch an open-source model that will surprise you. I think it will be much better than you hope, and you will be able to run very powerful models locally. At the same time, API costs will continue to drop significantly, and open-source models will be excellent.
[The translation continues in the same manner for the rest of the text, maintaining the original structure and meaning while translating to English.]Open AI Development: Talent Attraction and Competitiveness
Tan: Are you one of the best in the world at gathering the smartest people? What is the most difficult lesson you've learned in recruitment?
Altman: Recruiting truly intelligent people who are motivated, efficient, and good at team collaboration can make you 90% successful, and I'm always surprised by how much people focus on other things during the recruitment process.
Hiring someone who maintains a good performance record, is curious, actively integrates into work, and aligns with the company's vision works very well.
Tan: Does maintaining a good performance record mean someone who has held management positions at top institutions for twenty years and has a top-tier reputation? Altman: In the early stages of a startup, I don't recommend hiring such people. To be honest, YC hired people with extensive management experience in the early stages, but the results were not ideal.
We prefer to hire young, energetic people who can get things done, rather than those with extremely glorious resumes. I would ask, what is the most impressive thing you've done.
Tan: As the CEO of Open AI, what is the most difficult lesson overall?
Altman: We must do many things simultaneously, and many large companies are challenging us in various ways. We must spend more effort dealing with these issues, and we need to switch from one major decision-making approach to a completely unrelated but equally important one to respond and handle them.
Tan: For many software engineers wanting to create B2B, SAS, how to accomplish a complex and troublesome task currently?
Altman: Now is the best time for startups in tech history. The success lies in the fact that startups can iterate more easily and at a lower cost compared to large companies. Large companies have many advantages, but their iteration speed is very slow. However, cheap things are also easily replaceable, so there are many angles to look at this issue.
I suggest looking at it this way: Everyone faces the same challenges and opportunities, but when industry cycles change so dramatically, startups are almost always the winners. We may have never seen such a massive change.
Taking action from this direction, I believe you will be in a very advantageous position. Perhaps you can invite me to discuss, for example, what defensive areas you can build over time. I think this is an internal issue.
Tan: What would you say to your student-age self?
Altman: I wish someone had taught me long-term belief and resilience. Many people give up after one failure. Learning how to persist, I think, is really important.
Cultivate trust in your own intuition and continuously improve your decision-making and intuition over time, thereby enhancing this trust. The good parts are really much better than you imagine, and the difficult parts are hard to express in any way you can understand. You must persist.
One More Thing
Coincidentally, a few days later, he publicly launched a vote: In which year will the o3-mini level model run on mobile phones?
By the time of writing, over 40% of voters believe this model can run in 2025.
Okay, okay, just waiting (Doge).
Reference links:
https://www.youtube.com/watch?v=V979Wd1gmTU
https://x.com/WesRothMoney/status/1937148640575009176
This article is from the WeChat public account "Quantum Bit", authors: Shiling Yiran, authorized by 36kr for publication.
