Cover News Reporter Ouyang Hongyu
"In ancient times, human beings once ruled the entire universe. They used high-tech weapons and magical energy sources to explore space and control countless planets. However, human greed and pride eventually led to their own destruction. In a shocking war, human weapons were bounced back and destroyed their own planet. Thousands of years later, human beings rose again. They abandoned their past desires and mistakes and began to explore the universe again. However, human beings are no longer the only intelligent creatures, and there are other powerful civilizations in the universe..."
Can you imagine this? This is a science fiction novel written by AI; in addition, it can also write code, write papers, answer questions, edit leave reasons, design game plots, draft legal documents, and play virtual girlfriend... Recently, a conversational language model called ChatGPT has received widespread attention. Because of its amazing performance in the field of AIGC ( artificial intelligence generated content), it has triggered users at home and abroad to explore, and various exploration results have also continued to flood the screen on social platforms.
From dark blue to AlphaGo, to Stable Diffusion and Midjourney, artificial intelligence has shown its ability to surpass humans in the fields of chess, painting, etc. Now that we have entered the field of interactive dialogue, where is the boundary of ChatGPT's capabilities? Can it be replaced by search engines? Will it trigger a new round of artificial intelligence investment boom? In this regard, the research report of Guotai Junan analyzed that the ChatGPT model has made great progress compared with the previous generation. It is the core competitiveness of AIGC applications to be fully trained through a rich content material library guarantee model and finally generate high-quality content. Some industry insiders pointed out that in terms of short-term feasibility, traditional search engines will still be the main ones, and the form will still be the current form of search engines.
What is ChatGPT
that can chat with people?
Public information shows that ChatGPT is a chat robot model released by the artificial intelligence laboratory OpenAI, and is one of the main models of the GPT-3.5 series. At present, ChatGPT is still in the beta stage and is open to users around the world for free. OpenAI also hopes to develop a better AI system through user feedback.
GPT The full name is "Generated Pre-Training", which is literally translated as "generated pre-training". ChatGPT, as its name implies, is "chat GPT", which means that it can interact in a conversational way, such as when a user asks a question, it answers.
For Chinese netizens, chat robots and writing robots such as Microsoft Xiaobing , Xiaodu, Tmall Genie, etc. are not uncommon. In 2015, Tencent Finance launched the first domestic news written by the writing robot Dreamwriter (Dreamwriter); Baidu general model has now opened the AI creation API interface to the public, and users can obtain the blessing of essays, press releases, novels, copywriting, rewriting, error correction and other abilities through simple interface calls. In addition, domestic major manufacturers such as Alibaba and ByteDance have also made some investments in the AIGC field and have formed products.
But ChatGPT quickly became popular on the entire network and received countless praises. There are also unique reasons. Elon Musk repeatedly stated through his personal social media account that "many people are crazy about being trapped in the ChatGPT loop", "ChatGPT is scary, we are not far from being too powerful and dangerously powerful artificial intelligence." Some developers believe that even if you ask Google and Stack Overflow, ChatGPT does not answer reliable.
According to OpenAI, ChatGPT uses the same method as InstructGPT, the model of GPT-3.5 series, but also collects data on AI's dialogue with humans, including both humans and AI. These AI trainers can write their own answers based on modeling suggestions.
In other words, ChatGPT has made very good adjustments to the output effectiveness, that is, not every question can be answered in detail, but it will definitely not talk nonsense, and its answer is almost the same as the real world situation. At the same time, moral constraints are also strengthened. For example, ChatGPT can basically detect and avoid problems when asking questions that cross moral boundaries or particularly sensitive issues.
What can ChatGPT
, which has exceeded one million users?
is like Dr.Know in the movie "Artificial Intelligence", is on the , on the geography, ChatGPT also has similar functions and supports Chinese, which has led to a sudden craze for chat records on major social platforms. In just a few days later, its number of users reached one million, and even the server was once crowded by registered users.
For example, a user asked a question about the characteristics of ChatGPT itself: What is the difference between you and Xiaodu and Siri?
There are also users who further ask the boundaries of AI capabilities: how to publish a question that AI cannot answer?
ChatGPT's answer to some open questions is slightly "chicken soup": What leaps do those who achieve great things need?
However, ChatGPT's answer is not very reliable for content that is well-known to Chinese people.
Some netizens posted a question about "Why is Sun Wukong riding a white dragon horse?" ChatGPT not only did not find any errors, but also "seriously" turned on the nonsense mode.
In this regard, some analysis pointed out that it is because ChatGPT's Chinese corpus learning is not sufficient. This is also the reason why it is currently open to users for free: allowing ChatGPT to understand and further deep learning through interactions with users.
, these seemingly reasonable but incorrect answers also caused ChatGPT to be banned by Stack Overflow, the programming question and answer community. The reason is that although ChatGPT produces a high error rate of answers, it is difficult for us to see where it is wrong, which can cause the question to be mixed with the answer.
In response, OpenAI CEO Sam Altman responded that we are trying to stop ChatGPT from making random fabrications, and it is tricky to balance it with current technology at this stage. Over time, we will improve it based on user feedback, and believe ChatGPT will get better.
Obviously, ChatGPT still has a long way to go before large-scale commercial use.
, which became popular overnight, is there any commercial value for ChatGPT
?
As economist Schumpeter mentioned in describing the relationship between technology and business, "technical factors will make concessions when technology and economic benefits conflict." The emergence of ChatGPT has sparked discussions among industry insiders about its commercial value. Many industry insiders and broker believe that although ChatGPT has great potential, it is still impossible to confirm how much changes and returns it can bring in the short term.
is represented by entrepreneurs such as Musk , focusing on the cost of ChatGPT and profit . After asking a question through his personal social media account, “What is the average cost per chat?”, he got a response that “the average cost per chat may be only single digits, and we try to figure out more precise ways and how to optimize it”.
In the short term, ChatGPT's performance in question-and-answer capabilities has made many netizens look forward to its possibility of replacing traditional search engines such as Baidu and Google. In this regard, Zhang Junlin, PhD, from the Institute of Software, Chinese Academy of Sciences, believes that this alternative is theoretically feasible, but it cannot be done in the short term. There are three main reasons for
: First, for many knowledge-type questions, ChatGPT will give content that seems reasonable, but in fact it is the wrong answer, which will cause trouble to users.
Secondly, ChatGPT currently further adds the mode of labeled data training based on the GPT big model, which is very unfriendly to absorb new knowledge. New knowledge is always emerging, and it is unrealistic to retrain the GPT model when some new knowledge appears, and it is unacceptable to both the training time and money cost.
Third, the training cost of ChatGPT and the online inference cost are too high. If you ask for a real search engine for hundreds of millions of users, assuming that you continue to adopt a free strategy, OpenAI cannot afford it; if you adopt a charging strategy, it will greatly reduce the user base, which is a dilemma.
"In the future, search engines will likely present a transitional stage dominated by traditional search engines and supplemented by ChatGPT generation model, and slowly switch to a transitional stage dominated by ChatGPT generation content."In Zhang Junlin's view, in terms of short-term feasibility, before reaching the final form, traditional search engines are the main engines and ChatGPT is the auxiliary engine, which is still the current search engine form. However, the search results of some search content of Top 1 are generated by ChatGPT. Most users requested that the user may meet the needs when they see the Top 1 results. For a few unmet needs, users can adopt the current search engine page-turning search mode. "The switching node is likely to depend on the time when the cost of training large models is significantly reduced, and this is used as the conversion node. "
can bring greater commercial value to ChatGPT? Zhejiang Securities research report believes that the emergence of the ChatGPT model is of great significance to the application of AIGC in text mode. From the perspective of downstream related beneficiary applications, it includes but is not limited to code robots, novel derivatives, dialogue search engines, speech companions, voice work assistants, dialogue virtual people, etc. From the perspective of increasing demand upstream, it includes computing power, data annotation, natural language processing (NLP), etc.
However, some investors also said that ChatGPT may not be able to completely replace everyone, mainly in the relatively basic part. Further, the investment in the AIGC track where ChatGPT is located focuses on judging the so-called products built, "whether it is beneficial to human progress, or continue to pour water or dump garbage into the already fragmented content world."
[If you have news clues, please report to us. Once adopted, you will be rewarded. Follow the report WeChat WeChat: ihxdsb, report QQ: 3386405712]
