The popularity of ChatGPT is still going on. Although it is just a more intelligent chat robot, after all, it is just a form of AI expression. What fields can the technology be expanded to, and can it bring to various industries, what changes are more concerned by the industry.
With the deepening of the intelligentization of automobiles, the status of AI technology in automobiles has continued to increase in recent years, including autonomous driving and smart cockpits. AI technology is widely used. So the emergence of ChatGPT, along with this new round of upsurge, what changes will the field of smart cars bring?
Subvert the cockpit voice interaction experience?
For the application of ChatGPT, the answer given by Microsoft is the search engine, which uses dialogue to bring a subversive interactive experience to the search engine. In the car, a significant trend in the smart cockpit is to gradually reduce unnecessary physical buttons in terms of interaction, and to use touch and voice to operate different functions of the cockpit.
As an AI model based on the "Chat" function, we can easily imagine that ChatGPT technology is used in voice assistants, including the current car smart cockpit that focuses on voice interaction. The speech interaction process includes several technologies such as speech recognition, natural language understanding/processing, natural language generation, and text-to-speech. Judging from the products launched by mainstream manufacturers, the accuracy of speech recognition can already be very high, while text Speech-to-speech has been popularized to all major short video platforms, and video AI dubbing can almost achieve the effect of being fake.
However, although the application of voice assistants has been widely used on smartphones for more than ten years, the development seems to have stagnated to a certain extent. In many cases, voice assistants on smartphones can accurately recognize the words of users, but the problem is that they still cannot accurately understand users. Most of the commands, unable to answer or make corresponding commands. Nowadays, in-vehicle voice interaction has made some progress in recognizing commands, and can recognize continuous voice commands, but in essence, voice assistants can only recognize some fixed command combinations and perform corresponding fixed operations. Therefore, in-vehicle voice assistants are still not actually "smart".
The emergence of ChatGPT may bring some changes to the intelligent voice industry. From previous experience, we have realized that ChatGPT’s dialogue ability can not only "understand" the user's dialogue, but also combine the context of the context to carry out "logic " exchange. Although the accuracy of some content may not be high due to problems with data sources, in automotive cockpit applications, it may be possible to combine this type of voice model to greatly improve the user experience of in-vehicle voice interaction.
(Related: How Carmakers And Tier 1 Suppliers Can Secure Connected Cars)
What does autonomous driving have to do with ChatGPT?
From the perspective of underlying technology, ChatGPT belongs to the field of NLP natural language processing, but NLP is not limited to the application of text. In the field of visual recognition, the Transformer large model has long been widely used, and Transformer itself belongs to a training model in the NLP field.
Therefore, ChatGPT may not directly promote automatic driving, but will make a greater contribution to human-computer interaction. But the technical route behind ChatGPT may promote the development of autonomous driving technology.
In ChatGPT, the difference from the previous GPT model is the addition of human feedback reinforcement learning (RLHF) technology, which uses the content provided by humans to judge the quality of its own answers, so as to continuously improve iteratively.
In the future, the RLHF algorithm will be used in the large model to allow the model to learn the data taken over by human driving. For example, in the same driving environment, if the human driving action is consistent with the algorithm decision and does not choose to take over, it will be recorded as a good case; if the human driver takes over, it will be recorded as a bad case. That is, through good and bad judgments, the self-driving algorithm is guided to make decisions that are closer to humans.
write at the end
At present, from the perspective of in-vehicle applications, the best application scenario for ChatGPT is only in-vehicle voice interaction. Some people in the industry believe that there is not much room for the in-vehicle voice market. At present, iFlytek and Cerence, which almost monopolize the market, are not actually able to achieve ideal income in this field. In this case, using ChatGPT, which has a high training cost, seems a bit too small use. In the future, we can look forward to what kind of experience changes will be brought about by more intelligent voice dialogues on smart cars.
