Yesterday, OpenAI released an exciting news, including functional enhancements, fee reductions, etc. The biggest highlight is that I think it opens the door to building an ecosystem: GPTs.
For specific content, please see the link below:
Heavy! OpenAI releases GPT-4 Turbo, the most powerful ChatGPT in history is here!
Press conference link
https://www.youtube.com/watch?v=U9mJuUkhUzk&t=185s
We have also been discussing how to combine the optimization solution system and the (large) language model to provide information to each other, iterate each other, and promote each other. To solve intelligent decision-making problems in various professional fields and industries.
For example:
Dynamically analyze price signal influencing factors (weather, events) in the supply chain (logistics warehouse network) to form a dynamic pricing model and dynamic intelligent scheduling.
-
How to analyze safety operation manuals and operating specifications, and combine them with process design to dynamically embed checkpoints into corresponding process nodes to minimize risks and simplify operating steps and inspection steps.
-
How to obtain the information captured in the execution steps and the information fed back by the executives, and analyze the feedback on the process.
-
How to update the risk point check logic and steps based on feedback during the execution of operation steps.
-
How to form a dialogue with the operator in real time and dynamically. (Similar to the application of ChatGPT in specific scenarios.)
-
How to analyze customer feedback information, order dynamic execution information, system log information, etc. Form root cause analysis of KPIs and find weak links and risk points in the process. Continuously optimize and iterate the customer service system and provide feedback to the supply chain control tower and intelligent decision-making brain.
This all involves combining language models and optimization models. It also lowers the threshold for modeling, allowing business analysts and industry personnel to use decision-making models just like large models.
In the future, data-inspired models can be used to build bridges to assist algorithm analysts and business personnel, and then use quantum computing to solve some NP-hard optimization problems, which will greatly improve production efficiency.
How to distribute the value generated by the new species of large model + optimization + quantum computing? How to regulate the use of new species?
Will this new species become conscious? What if consciousness arises?
A brief discussion on the mathematical basis of surface classification
Teacher Gu Xianfeng’s early introduction article:
Connectionism and symbolism in artificial intelligence
Cai Shaowei’s Course: Introduction to Constraint Solving
https://www.bilibili.com/video/BV1gP411h7dp/
SOTA! ChatGLM3 refreshes the best results of the Chinese agent benchmark SuperCLUE-Agent
There will be a discussion meeting on the GPT logistics model in Fudan this weekend. I hope that through learning and exchange, some new ideas and implementable ideas can be generated.
Machine-translated from the Chinese original 确定性求解系统和基于统计的学习系统相结合：符号主义+ 连接主义，互相促进，共同进步，帮助解决行业中的问题 (published 2023-11-08). WeChat original ↗