It learns from reams of general conversation as well as articles, news stories, books, and other digital texts describing people like Elon Musk, Queen Elizabeth, and William Shakespeare.
The system also has a way of combining different concepts learned during training. The result is an almost endless collection of bots that can imitate an almost endless collection of people, repeating an almost endless number of topics, as Mr. Thiel discovered while chatting with the Karl Kautsky bot:
Jonas Thiel: The Marxists of your day were very convinced that revolution was a historical inevitability. Given the recent defeats suffered by socialists around the world and the lack of class struggle, even in times of great suffering, do you think it is time to revise this view?
Karl Kautsky: I have not. We may have erred strategically, but I don’t believe socialism is a pipe dream. I still believe that the proletariat must break the shackles of capitalism and that the working class must unite around Marxism. We must remember that capitalism is still deeply flawed, and our job as socialists is to agitate and spread class consciousness.
Jonas Thiel: I totally agree, but isn’t it wrong to assume that workers united to overthrow the bourgeoisie is inevitable? After all, the bourgeoisie is so powerful that it is difficult for workers to unite in collective resistance instead of acting alone.
Karkautsky: You are right, it would be a mistake to be overly optimistic in this regard. But I’ve found that it’s best to believe in class consciousness and revolution until I’m proven wrong. What are the options? Do you believe that workers will be chained forever? I disagree with this. Rather than succumbing to it, fight capitalism recklessly. I do not believe that the workers will never wake up. Even if I am overly optimistic, we cannot give in to pessimism. I refuse to be proven wrong. I hope you can understand.
Sometimes, a chatbot will get things done. Sometimes, it doesn’t. When Mr Thiel chats with an avatar meant to emulate Mr Reed, an American political thinker of the 20th century, it turns him into “some kind of militant Maoist, which is absolutely wrong”.
Like Google and OpenAI, among other top labs, Mr. De Freitas, Mr. Shazeer and their colleagues plan to train their systems using ever-increasing amounts of numerical data. This training can take months and millions of dollars; it also improves the skills of human interlocutors.
The rapid improvements of the past few years can only last so long, the researchers said. Richard Socher, the former head of AI at Salesforce who now runs a startup called You.com, thinks these exponential improvements will start to level off in the next few years, when language models can analyze almost anything Degree. Words on the Internet.
But Mr Shazeer thinks the runway is much longer. “Billions of people in the world are generating text all the time,” he said. “People will continue to spend more and more money to train ever smarter systems. We’re nowhere near the end of this trend.”