How ChatGPT will revolutionize the economy
When Anton Korinek, an economist at the University of Virginia and a fellow at the Brookings Institution, got access to a new generation of big language models like ChatGPT, he did what many of us have done. : he started playing with them to see how they could help his work. Be careful document their performance in an article in February, noting how well they handled 25 “use cases”, ranging from brainstorming and text editing (very helpful) to coding (pretty good with some help). help) to do math (not excellent).
ChatGPT misinterpreted one of the most basic principles in economics, Korinek said: “It was really bad.” But the mistake, easily discovered, was quickly forgiven for the sake of good. “I can tell you that makes me, as a conscious worker, more productive,” he says. “It’s great, no doubt I’m more productive using a language model.”
When GPT-4 came out, he tested its performance on the same 25 questions he recorded in February and it performed much better. There are fewer instances of fabrication; Korinek says it also does much better in math exercises.
Since ChatGPT and other AI bots automate cognitive work, as opposed to physical tasks that require investments in equipment and infrastructure, economic productivity gains can happen much faster. compared to previous technological revolutions, Korinek said. “I think we could see a stronger productivity increase later this year—in 2024 for sure,” he said.
What’s more, he says, in the long run, the way that AI models can help researchers like him be more productive has the potential to drive technological progress.
That potential of large linguistic models has emerged in the study of the physical sciences. Berend Smit, who runs the chemical engineering lab at EPFL in Lausanne, Switzerland, is an expert in using machine learning to discover new materials. Last year, after one of his graduate students, Kevin Maik Jablonka, showed some interesting results using GPT-3, Smit asked him to demonstrate that GPT-3 is, in fact, invaluable. for the kind of sophisticated machine learning studies his team does to predict the properties of compounds.
“He failed completely,” Smit joked.
Turns out that after tweaking for a few minutes with a few related examples, operational models as well as advanced machine learning tools was developed specifically for chemistry to answer basic questions about things like a compound’s solubility or its reactivity. Just give it the name of a compound and it can predict different properties based on the structure.