GuShiio.com 
Image source: Generated by AI
As DeepSeek climbs to 5 million daily downloads at an alarming rate and DAU approaches 23% of ChatGPT, large models are entering ordinary people’s lives at an unprecedented rate. However, in this close contact between AI and humans, we encounter a unique paradox: these super AIs can not only give amazing answers, but also weave clever and beautiful lies. Their answers are often logically consistent and conclusive, but they may be completely fictitious.
This phenomenon, which is called an illusion in the industry, is not a simple technical flaw, but an inherent feature engraved in the genes of the large model.
Faced with such a powerful and incompletely trustworthy AI partner, how should we humans behave?
On February 18, Tencent Technology specially planned the third live broadcast of the “Road to AGI” series: “Faced with the” Illusion Trap “generated by AI, what should we do?” Hu Yong, a professor at the School of Journalism and Communication at Peking University and author of “Post-Human Post-Truth”, and Chen Tianhao, a permanent associate professor at Tsinghua University and assistant director of the Center for Science and Technology Development and Governance of Tsinghua University, were invited to go out and ask former vice president of engineering of the large model team and former chief scientist of NETBASE Li Wei, jointly unveiled the mystery of the illusion of the large model from multiple dimensions such as technology, communication, and governance. At the end of the live broadcast, it gave the most practical advice on how to deal with the “big model illusion” for ordinary people.
Core points:
- The illusion is engraved in the gene of the big model: the big model is essentially a linguistic probability model. The user gives the previous conditions, and the model predicts the following. This is its training goal. Large model training is the compression of big data, capturing the knowledge points and regularity contained in the data, large and small. Large models are not rote databases. The “long tail” facts in the training data are no different from noise and are compressed and eliminated. Faced with this missing information, the model will automatically fabricate seemingly “reasonable” details to fill in, forming the AI illusion.
- When we talk about the mistakes of artificial intelligence today, it is essentially the mistakes of humans themselves: humans are inherently animals with prejudices and stereotypes, which makes misinformation caused by big models easier to accept and spread in the post-truth era.
- The “death theory of experts” carries huge dangers, which to a certain extent contributes to the spread of “big model illusion content”: many people believe that technologies like ChatGPT are “expert terminators”, and a large number of ordinary people who are not proficient in professional knowledge in a certain field can now give seemingly impressive content by copying and pasting. This intensifies competition among ordinary people, while making real competition for professional knowledge relatively weak. This means that those with superficial knowledge but lack in-depth understanding may gain more influence and voice in certain fields.
- Big model producers have the ability to reduce model hallucinations: While we acknowledge that language models cannot completely overcome phenomena such as hallucinations or fictions, research has also shown that big model developers actually have the ability to gradually reduce the incidence of hallucinations.
- If all the responsibilities caused by fiction or erroneous export are completely blamed on the enterprise, it is actually not conducive to industrial development. The birth of any emerging technology is a process of gradual improvement. It is impossible to achieve perfection from the beginning, nor can it completely eliminate potential risks.
The following is a summary of all the highlights of this live broadcast, with deletions and adjustments without changing the original intention:
What is the “illusion” of the big model?
Tencent Technology: First of all, please ask Teacher Li Wei to popularize science for everyone. Why do you say that “the illusion of big models is engraved in your genes”?
Li Wei: The basic principle of the big model is context-based prediction, that is, predicting the next chapter through a probability model given the previous chapter. The accuracy of predictions depends on how large models process and digest this information during training. The big model is not a database. It compresses and digests the knowledge system, including common sense or encyclopedic knowledge, but naturally eliminates long-tail facts and details that lack information redundancy.
Tencent Technology: The “fabricating facts” you mentioned is a very anthropomorphic term, and we can discuss this issue in depth later. Next, please share with Teacher Hu. As an in-depth witness and observer of the development of the Internet in China, what do you think is the difference between the content generated by the big model in the AI era and the communication in the Internet era?
Hu Yong: This is a quite big topic, and it can be said that there are many differences. I am mainly focusing on false information today. Because the illusion of the big model itself is one of the important sources of false information. We usually divide false information generated by large models into two categories:
The first type of false information comes from inaccurate raw materials in the training dataset. Large models are trained through web content, which itself often contains error information. Humans will have biases when disseminating information, and these erroneous information will be included in the training set, which in turn affects the output of the model.
The second type of false information is inferred from models. In some cases, the model does not grasp certain facts, but in order to justify itself, it will infer, creating illusions.
As Teacher Li Wei mentioned, illusions are endogenous to the large model and cannot be completely overcome. Even with the huge amount of data, the model can only capture a small portion of the information. This results in the missing of a lot of factual data.
To make up for these shortcomings, the large model makes inferences by learning the relationships between concepts, but this way of making up for it is like a person with a memory deficit doing things intuitively; the model still gives a “best guess” without knowing the answer, but this guess is often false information.
Tencent Technology: So why is it difficult for us humans to identify this “best but false guess”, causing it to spread quickly on the Internet?
Hu Yong: When we discuss the mistakes of artificial intelligence today, they are essentially the mistakes of humans themselves. Humans are inherently animals with prejudices and stereotypes, which makes misinformation more acceptable in the post-truth era.
In the past, we may have a consensus that although people have different values, the fact is the only one. Nowadays, people’s determination of facts has become blurred. In the post-truth era, there are not only differences in values, but also different understandings of “facts” themselves.
In addition, the defeat of rational thinking is also an important factor. I think there are basically three origins:
1. Many people question the reasoning itself and believe that all reasoning is just a process of rationalization.
2. Some people think that science is just a kind of “belief” that is subjectively constructed.
3. Others believe that objectivity is an illusion and that there is no objective truth.
These three sources of skepticism support each other, leading to the gradual giving way to emotional driving and intuitive judgment, making facts difficult to unify and false information more widespread.
Tencent Technology: It can be said that the illusions created by the AI model are similar to the illusions of humans themselves?
Hu Yong: It is directly related. The word “illusion” itself is an anthropomorphic expression. Because we often project human characteristics onto artificial intelligence, in fact, I personally think that the name “illusion” is not a good name and should be changed, but this issue can be discussed later.
Tencent Technology: Teacher Li Wei, in what scenarios do you think large models are prone to hallucinations? For example, when providing paper information, why can it clearly provide a false paper title and author?
Li Wei: Large models are most prone to mistakes when they involve specific entities (such as people’s names, place names, titles, times, places, etc.). This actually has similarities with the human brain, and we often can’t remember all the details. Large models use an abstract process when digesting data that attempts to find patterns from large amounts of data rather than recording all the details. Except for habitual liars, when humans cannot remember the truth, they say they have forgotten or add uncertain tones such as “seems, maybe”. Today’s language model is different. When it can’t “remember” the facts, it makes up the details that seem the easiest to read.
Hu Yong: For a long time, I have been paying attention to the process of knowledge production, especially the current way of knowledge production. With the development of technology, the authority of experts is gradually declining. Especially in China, the role of experts is often criticized and questioned, and sometimes even ridiculed as “experts.”
This concept of “death of experts” has been popular around the world for a long time. Many people consider technologies like ChatGPT to be an “expert terminator” because of its ability to provide seemingly professional content to all walks of life. As a result, many people believe that the role of experts will no longer be important after the emergence of large models. But this phenomenon carries huge dangers.
Teacher Li Wei mentioned that models like ChatGPT are misleading because they try to present a “quasi-authoritative” style, but are actually unable to avoid errors and prejudices. The danger is that when it cannot distinguish between true and false, it will confidently give wrong answers, which may seem credible, but may actually mislead users. Therefore, when using these large models, the first rule should be suspicion rather than blind belief. Developers of large models have also realized this and reminded users to be vigilant about their results.
Returning to the point just now, although the big model lowers the threshold for experts, it actually raises the threshold for truly becoming an expert. A large number of ordinary people who are not proficient in expertise in a particular field can now deliver seemingly impressive content by copying and pasting. This intensifies competition among ordinary people, while making real competition for professional knowledge relatively weak. This means that those with superficial knowledge but lack in-depth understanding may gain more influence and voice in certain fields.
Li Wei: But from another perspective, the “illusion” of the large model is actually a reflection of its abstract ability. It can also be understood as a manifestation similar to imagination.
For example, if a journalist provides false information when writing a report, it means dishonesty; but when a novelist creates a story, all characters, times and places can be fictional, which is the freedom of creation. The big model is similar to a novelist, in that it fabricate “facts” as a product of the imagination it has learned.
How to deal with the “illusion” of the big model?
Tencent Technology: So the big model is like both a “journalist” and a “novelist”; it must not only follow objective facts, but also have a certain imagination. How do you think that as “non-experts”, how should you identify when the big model plays the role of “reporter” and when it plays the role of “novelist”? Especially now that inference models can give a large number of answers in a short period of time, citing dozens or even hundreds of sources. As non-experts, how can we rationally doubt these results?
Hu Yong: No matter how many sources it cites and how eloquent data it provides, doubts must still be put first.
Li Wei: I think this is a balance issue. Many people are easily confused by its smooth expression and extensive knowledge when they first come into contact with the large model. Especially when you are unfamiliar with a certain field, it can be easily misled. Therefore, from a public perspective, skepticism, vigilance and checking information are necessary. However, a balance also needs to be found. If you remain completely skeptical, you cannot maximize the value of the big model.
For professionals who use large models in depth, they will find that large models do have their own unique features and can quickly integrate a large amount of knowledge. If they are skeptical of everything, they may miss enlightening ideas. I think that as a person goes deeper into using a large model, he can gradually find the feeling of distinguishing between authenticity and authenticity. In general, the overall framework and logic of a large model are usually more reasonable; however, caution should be exercised when discussing a specific fact.
Tencent Technology: Many students and even some children have begun to use large models to gain knowledge or help with writing. What does Tianhao think of this phenomenon?
Chen Tianhao: Everyone’s consensus is that the big model is essentially a language model. Although the AI model has been expanded to many fields including law and medicine due to its excellent natural language processing capabilities, it is still only a tool with “particularly good language” and cannot completely replace professionals. So the essential problem is that our expectations and reality are mismatched.
Tencent Technology: So is it necessary to restrict their use for special groups such as children and the elderly, or in serious scenarios such as law and medicine?
Chen Tianhao: Actually, there is already relevant work now. For example, in vertical fields such as law, some companies will use their long-term accumulated legal databases and more powerful underlying large models to improve the accuracy of output.
As for use by children, the situation is even more complicated. More stringent content screening and guidance are definitely needed for minors. This is more of a product side issue.
The “illusion” of the big model
Shouldn’t it be called an “illusion”?
Tencent Technology: You mentioned earlier that you think the word “illusion” is not a good one. If you don’t call it a big model illusion, what should you call it?
Hu Yong: In artificial intelligence, there is a parameter called “temperature”, which is related to the setting of creativity. When creativity is high, the model is prone to making more wild guesses; when settings are low, it provides more accurate answers based on the data set. This is also what makes the big language model interesting to use. So balancing creativity and accuracy is actually a big challenge when using large models.
Therefore, I personally have always believed that the phenomenon of large-scale model illusion cannot be generalized. For factual issues, illusions should be abandoned, but if it is in areas involving imagination, especially entertainment content, illusions can be a useful tool to enhance creativity.
Li Wei: Being able to fabricate stories and fabricate facts is also part of human wisdom and a critical ability. Harari mentioned in “A Brief History of Mankind” that the development of human civilization relies on the ability to “tell stories” and be able to fabricate myths, religions, ideals, and even feelings, these metaphysical things. It is this ability that allows humans to organize huge group cooperation, defeat all animals, and become the ruler of the earth.
Tencent Technology: Although we call it the “big model illusion”, in fact, the big model cannot truly understand human language. Do we sometimes overestimate the capabilities of large models and personify them too much?
Li Wei: Indeed, all our words about the big model are based on anthropomorphism. Artificial intelligence is essentially machine intelligence, just simulating human intelligence. All actions of AI, whether translation, summary, creation, problem solving, question and answer, chatting, autonomous driving, are all anthropomorphic, but circuits and models are running. The intelligent expression and response of the large model are essentially based on a probability model. But since the outbreak of the big model, we have all seen that its anthropomorphic intelligence is so excellent that from a behavioral perspective, it is no longer distinguishable from true or false. This is what the industry often says. Modern large models have passed the “Turing test.”
Tencent Technology: From a human perspective, active lying and unconscious mistakes are fundamentally different. However, the current big model does not actually have the ability to actively lie. Can we understand it this way?
Li Wei: Yes, the large model is essentially a probability model. Its output is based on the statistical probability of the data rather than the active intention, so it cannot be called “active lying”. If the questions are general, the model’s answers may also be general and full of banal or false content. When you provide more detailed and specific information to communicate with it, it is equivalent to changing the previous conditions of the probability model, which will compress the space for it to “lie”, and the model’s answers will be more accurate and exciting.
Chen Tianhao: Because when we discuss “active lying”, we assume that the big model has subjective consciousness, but current research has not reached such a consensus. My own understanding is that the essence of a large model is to predict the next token in the series by maximizing conditional probabilities. During the training process, model learning captures the statistical rules and semantic patterns of the language, thus gradually forming representation learning of the language, and even the emergence of some abilities. But it is still essentially a language model and is not yet conscious.
Tencent Technology: If it spreads misinformation, is this negative impact the same as the impact caused by human active lying?
Hu Yong: Here we can explain why the word “illusion” is problematic, because it actually has the problem of anthropomorphism. When we call artificial intelligence responses that do not match training data “hallucinations”, we are actually interpreting machine behavior based on human psychological phenomena. Over-anthropomorphism can lead us to mistakenly believe that large models are conscious or even emotional.
In addition, excessive use of the word “illusion” may also provide an excuse for companies that produce large models: exporting wrong content is the model’s problem, not the developer’s responsibility.
Therefore, I advocate using “fiction” to describe this phenomenon. The word comes from psychology, which means that when there are gaps in people’s memory, they often inadvertently fill in these gaps with logical reasons, which means that human memories are unreliable. This is very similar to the way large models generate content.
Tencent Technology: May I ask Teacher Tianhao, what are the typical harm cases of AI hallucinations you have seen so far? What measures have we taken to address these issues?
The illusion problem caused by the big model mainly appeared in the United States and Europe in the early days. For example, in 2019, there was a case involving infringement by an airline that entered the court in 2023. When a lawyer used ChatGPT to write a legal brief, he cited a large number of past court cases. However, when the court reviewed the prosecution documents, it found that these precedents were completely untrue.
This is a very typical example. When ChatGPT cannot find suitable information, it will “fabricate” some content to try to meet the needs of users. It just happened that the judge personally reviewed the document; if he had also used ChatGPT or other large models, the untrue content might have been left out.
Another common problem is that citations in academic papers can sometimes be forged. Not long ago, I asked the Big Model about the latest developments in French law, and the answer was clear, but after checking, I found that its citation was completely fake.
This once again hints at two major risks: one is that non-professionals will gradually disfascinate their professional fields because of the support of the big model; the other is that the traditional self-review mechanism within the professional field, which relies on peer review to ensure academic rigor, is also gradually deteriorating, and many times now large models are playing an important role.
Tencent Technology: There is a more complicated issue. If a person makes up or makes up content and eventually makes a mistake, the responsibility should be borne by this person. But if the big model makes mistakes, should the blame be on the company that developed the big model, or on the people who used the big model to generate content?
Chen Tianhao: This is the most difficult issue and a topic we often discuss when we are doing this business. I think the premise needs to be clarified first. Although we acknowledge that language models cannot completely overcome phenomena such as hallucinations or fictions, which is an inherent feature of language models, the literature also shows that companies actually have the ability to gradually reduce the occurrence of hallucinations.
When talking about GPT models in the past, pre-training was mentioned. In fact, after pre-training, there is also a post-training process. During post-training, supervised fine-tuning of the model is carried out based on human feedback to ensure that the model’s output is more in line with expectations. A very important point in the fine-tuning process is to emphasize that the model should not cause harm and avoid negative impacts as much as possible. Therefore, many benchmarks measure this as well.
Therefore, companies actually have a lot of room to invest resources to gradually improve the performance of the model. Of course, it may also bring about a “intelligence reduction” of the model to a certain extent. We call this an alignment tax, which is a work that must be completed for a large model to transform from experimental research and development to operational products.
If all the responsibilities caused by fiction or erroneous export are completely blamed on the enterprise, it is actually not conducive to industrial development. The birth of any emerging technology is a process of gradual improvement. It is impossible to achieve perfection from the beginning, nor can it completely eliminate potential risks.
Therefore, when making industrial policies, it is usually necessary to weigh the development of an industry and minimize potential harm to society, and try to encourage the development of these industries that have huge potential to improve everyone’s well-being. For early negative impacts, some supporting compensation measures can be taken to make up for these injuries as much as possible.
For example, there was a “safe harbor” clause in the early legal framework of the U.S. Internet industry, which stipulated that platform companies did not have to bear all legal responsibility for the information published on them; and if the platform promptly deletes relevant information when held accountable, it could be exempted from joint and several liability. This has greatly promoted the development of the American Internet industry.
The stronger the ability of the large model
Are “hallucinations” more likely to occur?
Tencent Technology: With the development of large model technology, their scale and iteration speed are also increasing. After the release of DeepSeek R1, we found that its hallucination level was significantly higher than its base model V3 and OpenAI’s GPT-4 models. Does this mean that the stronger the reasoning ability, the more serious the hallucinations will be?
Li Wei: In the past, it was generally believed in the industry that the larger the model scale, especially after sufficient post-training and enhanced reasoning ability, the number of hallucinations should decrease. However, at least in this test, the level of hallucinations in R1 was significantly higher than in V3. This shows that this relationship is not a simple positive or negative correlation, but is also influenced by other factors.
But overall, as the scale of the model expands, the training data also increases, the information redundancy naturally increases, and more facts and knowledge points can be more effectively absorbed into the parameters of the model, thereby reducing the probability of hallucinations. In addition, enhanced reasoning ability can build a “bridge” in the thought chain between information, making it easier for models to deduce correct conclusions, and also helping to reduce hallucinations. For example, if the previous non-inferential large model could not solve a complex mathematical problem, it would fabricate the answer. With a large reasoning model such as R1, due to the addition of thinking processes such as task decomposition, the possibility of the answer being correct is greatly increased, which obviously reduces the number of hallucinations and fabrication. But the industry-standard illusion measures mentioned above do not reflect this progress because they chose a single abstraction task to measure. Such measurements do not reflect the whole picture.
I noticed a comparison. Claude is a non-inferential industry’s top large model, and its hallucination level is even higher than that of the inferential large model R1 according to the same criteria. Therefore, we cannot simply assume that the increase in reasoning ability has brought more illusions.
Tencent Technology: But we have also observed that the inference model significantly exceeds the original basic model in terms of the amount of data and information generated. For example, compared with the basic model, the generation capabilities of modern large models far exceed the level of the mobile Internet era.
Hu Yong: Combining the opinions of the two teachers just now, some conclusions can be drawn. On the one hand, the attention problem mentioned by Teacher Li is very critical. The focus of the model determines its output characteristics. The research direction of model design is closely related to the phenomenon of hallucination.
On the other hand, while we affirm the breakthrough achievements of the big model, we cannot ignore its problems. For example, hidden dangers such as insufficient security, frequent hallucinations, and insufficient privacy protection. If these problems are not solved, they will affect its future development.
In general, I prefer to use the word “fiction” instead of “illusion”. Although large models always have the possibility of “lying”, they have certain resistance to “fiction”, so the problem of illusion will gradually improve over time. However, we cannot expect this process to happen independently. Instead, we need pressure from society and the government to push companies to invest more alignment costs when adjusting models to reduce hallucinations and reduce negative impacts on human society.
As for the amount of information, in the past we were worried that data storage bottlenecks would limit model training. Some predictions point out that by 2026, data for training will be exhausted. As a result, many organizations have begun to restrict the opening of data, and large platforms such as the New York Times and Reddit have also begun to require paid access to data.
However, the emergence of synthetic data has provided a new solution to this problem, and today the use of data is no longer limited by traditional web scraping methods. It is foreseeable that the supply of data will not be exhausted soon, and the amount of information will continue to grow exponentially without any suspense.
How to live with the illusion of a large model?
Tencent Technology: When should companies or model developers take the initiative to increase investment to prevent the negative impact of illusions? Is it time to strengthen such work?
Chen Tianhao: Big model companies still pay more attention to this aspect. Large companies like Tencent pay close attention to compliance issues. There is a theory in the field of social sciences that the bigger the enterprise, the greater the normative pressure it will face. Large companies tend to receive more supervision and attention, so they are under greater pressure in terms of standardization.
But we cannot require every company to have this awareness. What is more important is competitive pressure. When competition among peers begins, companies will feel pressure from the market and be forced to do better in alignment. Competition forces all companies to work hard to solve problems, which I think is more effective than government regulation.
Although the government also has relevant regulatory policies that require content generated by AI not to contain false or harmful information, how to test and implement these policies and how to achieve these requirements at a lower cost ultimately requires close cooperation between the company’s R & D team and engineering team., seek a balance between cost and alignment as much as possible.
Tencent Technology: At the current technical level of the big model, what are the biggest risks we can foresee? It shouldn’t be a science fiction plot like “AI exterminating mankind”, but from a social and communication perspective, what might the most serious situation be?
Chen Tianhao: False information is obviously the most intuitive influence. A large amount of content spread on online platforms has now been generated by AI. When we verify a fact, we habitually turn on the search engine. However, the retrieved content may be generated by AI, resulting in illusions and affecting our judgment of facts.
Tencent Technology: May I ask Teacher Hu, from the perspective of communication, when AI-generated content is intertwined with human created content, what impact may this phenomenon have on society?
Hu Yong: This is like the so-called “Ouroboros” model. In the end, all data will be synthetic, and it is impossible to distinguish which is created by humans and which is generated by AI. This can lead to a series of serious consequences. In particular, we overestimate the intelligence of artificial intelligence systems and thus develop excessive trust in them. In this way, once there is a mistake in the artificial intelligence system, the consequences will be quite dangerous.
We can make predictions through a thought experiment. Google’s Larry Page has promised that in the future, everyone may have an implant that will allow people to get instant answers through the Internet of Thoughts. If generations of people use this implant, we will become completely accustomed to this technology and forget the ability to gain knowledge through observation, questioning, and reasoning. Eventually we will find that our understanding of the world will be completely dependent on these technologies, and that our personal consciousness as “me” will no longer exist.
Tencent Technology: We mentioned before that dealing with illusions in the era of big models requires individuals to have higher discernment abilities. Teacher Hu once proposed the concept of equal rights on the Internet. Do you think AI has brought opportunities for equal rights or exacerbated the gap in technology use?
Hu Yong: Regarding the discussion of improving AI literacy, it is true that everyone should be responsible for their own actions, but we need to think: Why only emphasize the responsibility of the user side? Why not require AI companies to assume due responsibilities and reduce the risk of misuse at the source?
Take the conversation between Google CEO Sandar Pichai on a famous American talk show in 2023 as an example. Pichai admits that AI has black box problems-we often can’t explain why AI goes wrong. He said that as technology advances, these problems will gradually be solved. On the surface, this statement seems airtight, but the host pointedly questioned: If even you don’t understand how AI works, why should you promote it to the whole society?
Pichai’s response is that we are in a period of AI technology revolution and need to approach this technology with humility. But this may actually reflect the utilitarian thinking of some large AI companies: knowing that there are risks, but choosing to release products first, expecting to continue to improve in the future use process, who will bear the risks?
The so-called humility towards AI should not turn into rushing to market systems that have not been fully tested and aligned, and expecting society to absorb the problems it brings. On the contrary, AI companies should fully consider user needs and experiences when developing and releasing products. R & D teams should work with regulatory agencies and user groups to find responsible and ethical AI application methods. This issue deserves to be taken seriously.
Chen Tianhao: I fully support Teacher Hu’s views. In fact, from the moment ChatGPT was first released, we realized that this was a very dangerous product. It put a powerful technology into the world, and the global society was not yet ready.
The issue of alignment is very complex because there are huge differences within humans. Who to align with is a serious question that cannot be answered simply. Each large model company can only try its best to select representative groups and data for training based on its technical means.
As for the issue of equal rights, it is better to regard it as an opportunity rather than an injury. Because the big model does break down many knowledge barriers, allowing us to access the most cutting-edge knowledge at low cost. Although there is false information, we cannot give up this sea because of it. Although some of them may be exposed to risks innocently, we have no choice. Now that technology has been released, we can only accept reality and try to be prepared to respond.
Of course, enterprises should assume more social responsibilities, and laws and rules and regulations will also impose requirements. I believe that under competitive pressure, companies will try their best to do these tasks well. I think there is still a lot of work that can be done on the product side.
Tencent Technology: Finally, I hope that the three teachers can provide some practical suggestions on how ordinary people deal with hallucinations?
Li Wei: First of all, the “search” button is a very important weapon in dealing with hallucinations. It can concentrate related topics on the Internet and has high information density, thereby improving the authenticity and accuracy of answers and reducing the opportunities for hallucinations to emerge.
Secondly, if you are engaged in creative work, you can use the “reasoning” function to give full play to its strong imagination, generate unexpected and beautiful articles, and even transcend the limitations of traditional writing in some aspects.
Finally, if you directly ask the large model to do simple tasks such as summarizing facts, the result of calling the large model for reasoning may be distorted. The simple way is to not use the reasoning model (under the R1 interface, just not press the deepthink button). If you use an inference model, you can try to add a prompt word, such as “Please be faithful to the original text and summarize”, which may restrict subsequent generation and reduce the chance of making mistakes.
Hu Yong: First, you can use as many large models as possible. Each model has its own advantages. After using different models, you can gradually gain your own experience and achieve better results.
Second, for professionals in a certain field, it is recommended to use a vertical class model trained based on a specific industry-specific corpus. These models often better serve industry needs and help professional growth.
Chen Tianhao: First of all, when interacting with the large model, explain your needs in as much detail as possible. The more sufficient the information you input, the higher the accuracy and alignment of the output.
Secondly, try to use multiple large models for comparison and verification.
Finally, get to know some human experts and communicate with them more. They have some knowledge that has not yet been covered by the current large model of the technical stage, and can provide more reliable opinions. Of course, what is more important is actually to improve our own cognitive and critical abilities.
The craze of technology must eventually accompany the wisdom of human beings and build a “human-machine relationship” that balances suspicion and trust. Perhaps only then can we maintain the bottom line of “reality”.
