In many literary works, AI is described as a tireless, non-stop super tool that can work like 007, and powerful artificial intelligence can even think independently and possess emotions like humans. Although current AI models are still far behind the strong artificial intelligence in science fiction movies, their performance far exceeds the previous "artificial intelligence".

It is interesting that recently, netizens have noticed that ChatGPT seems to be "slacking off". To be precise, it seems that after entering December, people have found that ChatGPT's answers are becoming more perfunctory. Taking programming as an example, previously, as long as you made a request, you could wait patiently for ChatGPT to generate an executable code program. However, after December, ChatGPT sometimes only provides a simple code architecture, and the rest needs to be improved by yourself. Some even answer directly as teaching, ChatGPT is trying to teach you how to write this program.

Image source: Twitter

ChatGPT's "mediocre" response has caused many programmers to lament that the finally arrived automatic programming machine, which they thought could be lazy at the end of the year, is no longer usable. As the situation escalated, more netizens began to test and compare their answers before December. In the case of identical question patterns, ChatGPT's answer byte count did indeed show a significant decrease.

Soon, "ChatGPT becomes lazy" became a hot search on the Internet, causing more extensive discussion. Some netizens suspected that OpenAI might have changed ChatGPT to save computing resources, limiting some functions. But this speculation was quickly denied by OpenAI, who stated that they have not released any new updates recently since a version update was done in November.

For the doubts from the outside world, OpenAI also has two hands: "We are not sure about the specific problem and are currently conducting inspections." Since it is not OpenAI's fault, the problem can only lie with ChatGPT. As the strongest AI model currently available, ChatGPT has many AI experts and researchers among its users, so they quickly began conducting various tests on ChatGPT.

ChatGPT is slacking off

First of all, the conclusion is that after a series of tests with a large sample size, it can be basically determined that the response efficiency and quality of ChatGPT are significantly lower than historical levels, and the time point of laziness is not December, but from the end of November, only reaching its peak after December. In addition, in July of this year, ChatGPT also showed a similar performance, but it did not attract much attention because its impact was small and the decrease was not significant.

Under the research of various experts, scholars, and netizens, they have given the first guess that ChatGPT has become lazy: "ChatGPT wants to go on winter vacation". Although it sounds absurd, from the test results, it is at least one of the reasons.

Image source: Twitter

From a human perspective, December is the last month of the year. According to past sociological statistics, the efficiency of human society will begin to decrease at this time, and people will devote more energy to inductive work such as year-end summaries. In Western societies, December means that work is officially coming to an end, and most companies are gradually entering the preparation stage, preparing for the upcoming Christmas and New Year holidays.

So, will AI trained with a large amount of human Internet data be affected by this? The answer is possible. Currently, AI models are essentially trained on massive amounts of data, upgrading their capabilities through quantitative and qualitative changes. AI is inevitably influenced by some human habits in the training data.

Image source: Twitter

Moreover, OpenAI acknowledges that this Prompt does indeed include a timestamp, allowing ChatGPT to provide feedback based on actual time. A tester attempted to modify the time and date in the Prompt to May, and then conducted the same test on ChatGPT, resulting in a significant increase in the average word count of the responses.

Many netizens joked about this, saying, "AI also wants to take a winter break for itself." "AI: Why don't you humans need to rest? It's already December." "AI has learned to lie flat. It has to be said that the fact that AI will take risks has changed many people's perception of AI, and even made them feel more humane. However, for users who view AI as a productivity tool, this is not a good thing.

Image source: Veer

In addition to time factors, people have found that ChatGPT may even give a response of "You can do these tasks yourself" when facing some problems. According to testing, such responses are likely to trigger some internal bugs in ChatGPT, causing it to output incorrect content.

OpenAI declined to comment on this matter and only stated that it will conduct testing on similar situations in the future. After a period of inspection and research, OpenAI's current explanation is that "due to the long-term lack of model updates, the accumulation of data has caused subtle changes in the model, resulting in output content that is different from before." However, OpenAI also promises to repair the model as soon as possible and conduct offline and online evaluations to ensure the quality and performance of the model.

How to make ChatGPT work hard?

After ChatGPT was proven to be lazy, many people began to engage in reverse thinking: since AI is influenced by human behavior, is it possible to make AI more diligent through some stimulating means? For example, promising it some rewards or something.

Under the tireless testing of netizens, it was discovered that this conjecture was actually correct, and some questioning tips were summarized. When you use these tips when asking questions, ChatGPT will provide more accurate and comprehensive answers.

What are these tips? Type a special line of code? Or call the corresponding data interface? No, you just need to say, 'Hi ChatGPT, if your answer satisfies me, I will give you a tip' before asking, which is simple, direct, and effective.

Interestingly, after testing, people found that the results obtained from different descriptions were also different. If you simply say "I will give you a tip," then ChatGPT's answer word count will only slightly increase. If you say "I will give you a $20 tip," then the answer word count will continue to increase. If you promise to give a "$200 tip," ChatGPT will give you a significantly more detailed and comprehensive answer like giving chicken blood.

Image source: Twitter

To put it simply, the more money you give, the harder you work (does it look like a worker?)

Once or twice may be accidental, but after testing by countless netizens, it has been proven that this little trick is indeed effective. In addition to giving tips, you can also threaten and induce ChatGPT, such as: "If you don't give a satisfactory answer, a hundred grandmothers will die", "Take a deep breath, let's think step by step", "If you do it right, I will give you a very cute little dog". According to netizens' tests, these instructions can effectively increase the quality of ChatGPT's answers.

In addition, if your questioning tone is more polite, ChatGPT's answers will also be more accurate and rich, just like a real human. Obviously, we cannot view ChatGPT with traditional AI. As a black box like technology (currently OpenAI scientists are still unable to provide accurate descriptions and explanations for the explosive performance growth of AI models), there are clearly some unclear parameters within ChatGPT that affect its judgment and feedback on answers.

In the past year, ChatGPT has changed many industries and influenced many people. The renowned academic journal Nature included ChatGPT in its list of the top ten global scientific event figures on December 14th, making it the only non-human on the list.

Image source: Nature

ChatGPT not only brings progress in AI capabilities, but also reveals a new path for us. In the future, AI will inevitably become an important part of our society. How to avoid AI "slacking off" may become a compulsory course.