It seems that voice assistants have made no progress over the past decade. Users have become accustomed to lowering their expectations and giving clear, direct commands such as “lower the volume,” “turn off the lights,” and “play the next song.”
However, the emergence of ChatGPT has changed everything. People are experiencing the sharp, fluent, and rich conversational abilities of ChatGPT, the new Bing robot, Google Bard, and other large language models driven by robots, making Siri’s stiffness more apparent than ever before.
The big companies behind voice assistants have also felt the pressure.
On April 28, Beijing time, Amazon released its first-quarter 2023 financial report, and CEO Andy Jassy stated during a conference call, “We have built a large language model under Alexa, but we are building a larger, more versatile, and more capable large language model, which I think will accelerate our vision of becoming the world’s best personal assistant.”
Almost at the same time, The Information reported that Apple engineers are working on combining the LLM large language model with Siri to launch a smarter Siri, which will meet users with subsequent iOS updates.
Google seems to have reallocated resources between Google Assistant and Bard and integrated the two in its architecture.
In China, earlier this month, Alibaba announced that all of its products will be integrated with the “Thousand Questions with Common Sense” large model, which includes Tmall Genie, of course. In the official demonstration demo, the new Tmall Genie supports free dialogue and generates content based on user needs and scenarios at any time.
The direct reason for Siri’s “stupidity,” or stiffness compared to ChatGPT, is, of course, the technology behind it.
The new generation of chatbots represented by ChatGPT is driven by large language models (LLMs). These model systems are trained on large network datasets and can recognize and generate text based on the dataset to generate complete sentences.
In contrast, Siri, Alexa, and Google Assistant are basically command and control systems. These systems can intelligently understand limited problems and requests. If the user’s question is not in its code, the assistant is helpless.
According to a report in The New York Times, John Burkey, an engineer who once worked on Siri at Apple, said that adding new features to Siri is difficult due to the complexity of the design.
Siri has already realized the threat posed by ChatGPT, and the situation of not advancing is retreating has already formed. As for how to advance, that is a problem.
Hopefully, one day we can pick up our phones and ask Siri, “Is Beijing hotter than Shanghai today?” and get a satisfactory and effective answer.