time leap, sci–fi, photorealistic, –niji 5 –ar 3:2 –s 1000
Since the release of ChatGPT, various industries have not stopped exploring its capabilities.
Many large companies have followed suit, competing to launch their own large language models, such as Google Bard and Claude from abroad, and Wenxin Yiyan, Pangu Model, and Tongyi Qianwen from China, all launched within the last month or two.
Due to the rapid development of ChatGPT, many industry leaders fear that an uncontrollable situation may arise in the future.
Musk, Apple co-founder Wozniak, and Turing Award winner Bengio have all called for the AI industry to immediately halt research on models stronger than GPT-4 and have opened a joint signature campaign. As of now, over 13,000 people have signed.
Coincidentally, two heavyweight figures in the AI field, Andrew Ng and Fei-Fei Li, also publicly stated during a live broadcast yesterday:AI research cannot stop!
The two opposing forces each hold their own views, and neither can persuade the other.
At this point, I can’t help but ask, will ChatGPT really be shut down?
My answer is very clear: impossible.
From the day ChatGPT was released, it has rapidly integrated into various industries, attempting to break the old patterns of the traditional internet and establish a new order.
Many friends who have used it have expressed that their daily lives and work have become inseparable from it.
So, how do we break the old patterns?
Let’s analyze it briefly.
First, the strength of AI lies in its strong logical thinking and learning ability.
Secondly, from a capital perspective, leveraging AI capabilities to prioritize the elimination of positions with high recruitment costs and salaries makes economic sense.
By now, I believe the conclusion is quite clear.
Programmers may become one of the professions most impacted by ChatGPT.
In addition to the points I mentioned above, there is another particularly important consideration.
To make their large models stand out in this AI battle, big companies will go to great lengths to accumulate computing power, hire industry experts at high salaries, and mine various valuable data to continuously optimize and iterate their AI models.
However, there is a hidden value in this, which also requires AI to possess strong programming capabilities for research direction, that is:
Enabling AI to achieve autonomous evolution
Doesn’t that sound like a sci-fi movie plot?
Here, I want to share with you a theory proposed by British mathematician I.J. Good in 1965 called the “intelligence explosion”:
If we were to define a “superintelligent machine”, it would be a machine that can surpass all human intelligence activities, regardless of how smart the person is.
Since designing machines is also one of these intellectual activities, a superintelligent machine could design even better machines.
Therefore, without a doubt, human civilization will enter the era of “intelligence explosion”, at which point human intelligence will be far behind.
Thus, the first superintelligent machine will become humanity’s last invention, as long as it is obedient enough and tells humans how to control it.
Currently, developers on GitHub are already attempting to explore this direction, even though they may not realize the impact these attempts will have on human society in the future.
A few days ago, I shared two open-source projects in the community, from which we may gain some insights.
Teaching ChatGPT to Iterate
To have AI assist humans in completing a complex task, it is necessary not only for it to understand the task content well but also to grasp the priorities of various tasks.
BabyAGI is an intelligent task management Python script developed based on GPT, used to test the practical effects of AI task-driven autonomy.
In this system, BabyAGI not only needs to understand the tasks assigned to it but also needs to explore independently, creating tasks, determining task priorities, and executing tasks.
GitHub:http://github.com/yoheinakajima/babyagi
Below is the execution process of this script:
-
Extract the first task from the task list;
-
Send the task to the execution agent, which uses OpenAI’s API to complete the task based on the context;
-
Enhance the results and store them in Pinecone;
-
Create new tasks and rearrange the task list priorities based on objectives and previous task results.
The underlying logic of the system is that it can create tasks based on task results and set goals.
With this step, we then need to teach AI to search the internet for information, obtain solutions, and iterate on its own.
Enabling GPT to Self-Iterate
Auto-GPT, although just an experimental open-source application, has already demonstrated the powerful capabilities of the GPT-4 language model to the outside world.
In simple terms, it can optimize its own code and automatically fix bugs.
This means it is an intelligent coding tool that understands how to enhance its capabilities through programming, marking a crucial step since the emergence of GPT-4.
GitHub: https://github.com/Torantulino/Auto-GPT
In addition to being able to automatically search the internet and gather various data, it can also attempt to access mainstream websites and platforms, using GPT for file storage and summarization.
We may find ways to break through the limits of AI possibilities from this project.
Final Thoughts
The two open-source projects mentioned above, in my view, are preliminary explorations of AI attempting self-evolution.
Although they are just initial explorations, they provide some research directions and implementation ideas for large companies.
I believe that the day when AI truly achieves autonomous evolution is not far off.
When that time comes, where will we go?
—
If you are interested in the future business applications of ChatGPT and the direction of technological exploration, feel free to scan the QR code below to join our community for mutual exchange and discussion.
