The wave of artificial intelligence technology led by ChatGPT is still hot, but OpenAI, the company that made this product, survived the cold bench for many years, and finally soared into the sky. How did they do it?
Therefore, I have a strong interest in the organization construction of this enterprise. I searched and searched, and finally found some clues on the official OpenAI blog, hoping to get a glimpse of the leopard.
This article introduces the experience of OpenAI's learning day (Learning Day) every Thursday.
They found that at OpenAI, the greatest contributions came from cross-field experts, so the company either hired them or cultivated them. Before the implementation of learning days, it was rare to see employees grow across domains, for example, software programmers had little exposure to machine learning, and since implementing learning days, this growth has become very common.
1. Why is there a study day?
In November 2018, the initiator of the study day, Wojciech Zaremba (Woj for short), realized that in many areas he was stagnant, because he was always overwhelmed by urgent tasks.
These areas are increasingly important to Woj. For example, he kept trying to evaluate whether the team should switch to a deep learning framework, but after an hour or two of coding, he was always interrupted by various chores, preventing the work from making any progress.
Woj kept hearing about research in other fields that might apply to robotics, but knew nothing about those fields, and reading for half an hour at a time didn't help him.
He knew the best way to tackle this problem was to set aside one day a week to study. But if that's what he needs to be productive, chances are his team needs to, too, so Woj tries to apply this as an experiment to the entire team.
He expects short-term productivity losses across the team, but will see gains in one to two years. However, the effect came faster than expected, and within a month, he began to see that the communication between the researchers and engineers was smoother, and each began to use the terminology of the other's field of expertise correctly. Within half a year, he found that researchers were discussing how to refactor the codebase, and engineers were taking on research tasks.
Although the company previously encouraged self-study, it never seemed to work. Things are different now, a team member was able to go from knowing nothing about machine learning to contributing to computer vision within three months. A very good engineer studied reinforcement learning (RL) for half a year and is now producing results comparable to a PhD in the field.
2. What do you do on study day?
1. Read cutting-edge papers, including reading papers, reproducing papers, or corresponding artificial intelligence tutorials.
2. Learn basic knowledge, including basic knowledge of mathematics and science, basic knowledge of programming, writing, hands-on ability, etc.
3. How to maintain the vitality of the study day?
First of all, it is clear and strict that the study day is not a day of leisure, but a day dedicated to hard work. Team members should make the most of this day and really use it to learn, and it is forbidden to use this time to watch dramas, and don't use it as an extra weekend.
The study day may be used for work, in which case the study day will become a normal working day, as employees may want to finish their main project faster (due to internal or external pressure), so, by having each This can be avoided by members studying on the same day. This can have a positive effect and encourage everyone to make the most of their study day.
Learning days require each team member to proactively share their learning progress, and employees can share what they have learned, which keeps learning exciting and provides a mechanism for accountability.
This is the general content of the study day. If you are a team leader, you can try it out, maybe it will produce some unexpected positive effects, and look forward to your feedback.
Blog address:
https://openai.com/blog/learning-day#OpenAI