As a learning human language model, ChatGPT has been published for a whole year since its release in November 2022. It has unexpected development. What I am talking about is not revolutionary and convenient, not to become faster and stronger, but to learn from human bad habits.: Lie flat.
Recently, everyone began to discover that ChatGPT was suspected of "lazy", and three major crimes began in November:
1. The answer to the question becomes slow: not only is the digestive instruction slow, but the answer time is also obviously stretched;
2. The answer becomes short: if you avoid complex issues, and the instructions for filling in the form will give up only half;
Three, sometimes ignore instructions: such as refusal to provide complete answers, refuse to execute constant file conversion, and read links.
All parties are guessing the reason. Some people have fun. The human language model has learned human beings. It is necessary to encourage it to speak, rest after a long time of work, enlarge the fake mood at the end of the year, or even have self -awareness.There are also conspiracy theories, saying that the previous ChatGPT development company OPENAI management palace fighting is the fuse; some people refer to ChatGPT or imitating the antibiotic hibernation behavior, and want to lie directly.
However, it is worrying that OpenAI emphasizes that since November 11, there is no modification of the model, and said that the user has received "ChatGPT lazy" and admits that it is trying to solve this problem.In other words, in these two or three weeks, the natural changes of ChatGPT, no artificial interference, the creator failed to find the reason and removed it wrong.
OPENAI releases ChatGPT, pushing AI change to the public on the one hand with different technologies to improve the efficiency for text work, sound painting production, operation, etc.It continues to improve the year, including the performance, context understanding of the update model, and generating more context consistency; expand the scope of application, provide information about various themes, help users solve problems, perform creative writing or information collection; borrow fromThe integration of user feedback helps to solve potential defects and enhance user experience; cope with multi -language needs, whether it is their respective performance in various languages, or doped text expressions (such as English plus Chinese and white clip under the Chinese text and white clip.Answer); update at any time, in addition to improving technology, add more data input in real time, from data limited to 2021, open to real -time, can read the latest network data, making the output results more accurate and effective.
In the first quarter of this year, when the world was in a hurry to test the human language model, many comments thought that the development of AI such as ChatGPT would cancel / change many types of workers or even layoffs, affecting human social capabilities, reducing information on information, reducing informationTrustness, improvement efficiency, unified language usage, increasing privacy and information security risks.It is a pity that one year later, I did n’t see the sky like a movie novel. At the popular level, the change seems to be not too big and too obvious.Yes, but it is not as well as revolutionary levels.
However, in this year, the unimaginable AI of the industry has become lazy. If it fails to repair it in the short term, the following impact may occur:
1. More users switch to BARD, XAI or other services;
Second, in order to prevent accidents, applications and window docking based on GPT-4 are rewritten, so that engineers do not dare to rest at the end of the year;
3. Human beings are doubtful and retaining the attitude of scientific and technological development. On the one hand, it is a good thing to be vigilant in security and trust, but on the other hand, it may slow down the development of science and technology;
Four, ChatGPT, etc. becomes out of control, and the consequences are unpredictable. If humans fail to control, and AI is destroyed, it will be expected whether another AI will become a savior.
The author is a digital technology practitioner
