Please note that these ideas are preliminary and merely interesting speculations. I have not had sufficient time to fully consider/research all of them, and I may be incorrect in many of them. However, I do hope this content is engaging for those attempting to comprehend current developments.
OpenAI 's o3 model should not be a surprise. OpenAI demonstrated the scaling laws in testing to us two months ago, and the history of computing dictates that we should trust the trendline, regardless of how improbable it may appear. The truly astonishing aspect is the rapidity with which it has occurred in just two months. We have transitioned from university-level AI to doctoral-level AI in such a short timeframe. For humanity, change is exciting, but rapid change is alarming.
The trajectory of future developments is becoming increasingly clear. o3-level models excel at optimizing for any task for which a reward function can be defined. Mathematics and programming lend themselves readily to the design of reward functions, whereas novel writing presents a greater challenge.
Consequently, in the short term (within one year), we will observe models exhibiting varying proficiencies. They will essentially attain AGI-level capabilities in mathematics, programming, and general reasoning, but their written fiction will be mediocre.
Will AI swiftly replace all software engineers? No. Software engineering is not solely about constructing pull requests based on highly specific prompts. Unlike mathematicians, software engineers constantly interact with the physical/ reality world, namely, with other people.
Engineers must collaborate with clients to understand their needs and with team members to comprehend their requirements. When engineers design architectures or write code, they operate within a substantial organizational context. o4 will not be capable of replicating this. However, o4 will empower engineers who possess this context to operate at 10x speed.
If software engineers operate at 10x speed, will fewer be required? Considering a specific company, they might require fewer software engineers, achieving the same output with a leaner team. However, the global demand for software engineers may increase, as the world can certainly utilize significantly more high-quality software. Therefore, we may witness a golden age of applications from leaner companies, offering personalized micro-applications for individuals and businesses alike.
In the long term (>2 years is considered long term now, ironically), software engineering will be drastically different, making predictions difficult. How could it not change when o6 systems exist and are fully integrated into our applications? Roles such as front-end engineers may cease to exist within three years. Is this unusual? Not particularly—the role of front-end engineer did not exist 30 years ago.
We should acknowledge that software has been self-disrupting with each generation. Software has always been, and will continue to be, about translating needs into pure logic. This translation process has ascended from binary to the abstraction level of Python. The current difference is that it is now ascending to English.
This shift to English opens coding to non-technical individuals. However, the most effective builders will remain those capable of navigating between abstraction levels and reality
In summary, as software engineering is fundamentally about understanding and solving organizational needs through code, the complete automation of software engineering will coincide with the complete automation of all organizations.
We have discussed knowledge workers, but what about manual laborers? AI will impact them as well, but at a slower pace, as it must contend with gravity and friction. o-level models will not be of significant assistance to robotics, as a model requiring an hour for processing is not conducive to a factory production line. Enhanced intelligence in foundation models is beneficial, and o-level models will aid in training these models, but this will not resolve the primary bottleneck in robotics advancement. The primary bottlenecks are hardware improvements and fast/reliable models for perception and action. These will require more time to improve (i.e., several years). The truly rapid advancement of robotics will likely occur when robots begin manufacturing robots and AI begins conducting AI research. This may originate from o-level models, but this is likely several years away.
