Digirise Co., Ltd. has released a detailed explanatory video about the latest model "o3" announced by OpenAI. The o3 is positioned as an evolution of the recently announced o1, and is an innovative model announced in a short period of just one week. The previous o1 series had sufficient performance, but the o3 has evolved further, showing a great leap forward especially in inference capabilities.

The distinctive feature of the o3 series is that it offers two different approaches: the full-spec "o3" and the cost-conscious lightweight model "o3 mini." The o3 basically demonstrates the strongest performance in all fields, and is particularly outstanding in the fields of mathematics and programming. On the other hand, the o3 mini emphasizes processing speed, allowing for more practical operation while maintaining the minimum required performance.

Breakthrough Achievement in AGI Benchmark

The latest o3 model achieved a breakthrough in the general-purpose inference benchmark known as Arc. The Arc test measures tasks that are easy for humans to solve, but that traditional AI models have had difficulty with. What is particularly noteworthy is that o3 achieved an astounding accuracy rate of 87.51 TP3T, beating the human-level standard of 851 TP3T.

Achieving this result requires a large amount of computing resources, and the current problem is that it costs between $17 and $20 per problem. Considering that a human can solve a similar task for $5, there is still room for improvement in terms of cost. However, it is predicted that the cost of processing will gradually decrease with the emergence of new versions such as O4 and O5.

Performance in the fields of mathematics and programming

o3 showed a high accuracy of 96.71 TP3T in the field of mathematics, and was confirmed to have the ability equivalent to the top 175 programmers in the world in the field of programming. This is an extremely high level of development ability when viewed relative to the population.

The high programming capabilities are particularly noteworthy, greatly expanding the applicability in actual software development. The o1 Pro already demonstrated high capabilities in strategy formulation and document creation, but the o3 achieves even greater performance.

Key Safety Enhancements

o3 brings significant improvements to the safety of AI models. OpenAI has strived to make AI models safer while improving their ability to respond appropriately to user requests.

The safety of AI models has always been an important issue. OpenAI takes a careful approach to developing safe AI, and we continue to adhere to this policy in the development of o3, because our number one priority is providing safe answers to user requests.

Currently, o3 is conducting public safety testing and is working to find vulnerabilities and areas for improvement in the model at an early stage. It plans to release the model in stages, as if to "check everything before crossing the bridge," and will accept applications for safety testing until January 10th. Those who wish to participate in the pre-test can apply through the dedicated form.

Skills and measures that will be required in the future

The o1 Pro already has an IQ of over 130, and has achieved a 98.51 TP3T correct answer rate on the Japanese National Medical Examination. The o3 will surpass that performance even further, which could dramatically change the value of intellectual labor.

What is particularly noteworthy is that "humanness" will become more important over the next 50 years or so. Specifically, the following skills will be important:

• Networking ability
• Face-to-face communication skills
• Ability to use people

Since average users still have a resistance to AI and machines, human values will become more important. Business people of the future will be required to have a sense of gratitude on a daily basis and to improve their human skills.

AI tool introduction and utilization support guide

We provide AI training and AI tools for businesses. We have already implemented our services in over 150 companies, and have supported the use of AI in companies. We are particularly proud of our top track record in Japan for corporate AI training.

To help you experience the concrete effects of using AI, we are providing a free collection of 100 prompts that could not be introduced in the video. This collection of prompts summarizes practical use cases that directly lead to improved business efficiency. We also have a white paper that summarizes the winning strategy for introducing AI to companies. You can check these materials immediately by answering the questionnaire in the video description.

We accept individual consultations regarding the use of AI in actual business operations and the introduction of tools. We will propose the most suitable method of using AI according to the size of the company and the industry. We also provide support for promoting and educating employees on the use of AI within the company based on our extensive track record.