Struggle for power
When Anton Karinek, 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 do: He started playing with them to see how they could help him. work. He carefully documented their results in an article in February, noting how well they did in 25 “use cases,” from brainstorming and text editing (very useful) to coding (pretty good with some help) to math (not so much).
Karinek says ChatGPT has misinterpreted one of the most fundamental tenets of economics: “It’s very bad.” But the mistake, which was easily noticed, was quickly forgiven in light of the advantages. “I can tell you that it makes me more productive as a cognitive worker,” he says. “I’m definitely more productive when I use the language model.”
When the GPT-4 came out, he tested its performance on the same 25 questions he documented in February, and it performed much better. There were fewer instances of fabrication; he also did much better on math problems, Karinek says.
Because ChatGPT and other AI bots automate cognitive work as opposed to physical tasks that require investment in hardware and infrastructure, improvements in economic productivity can happen much faster than in past technological revolutions, Korinek says. “I think we could see more productivity growth by the end of the year — certainly by 2024,” he says.
Who will control the future of this amazing technology?
Moreover, he said, in the long run, how AI models can make researchers like him more productive could help drive technological progress.
This potential of large language models is already evident in research in the physical sciences. Berend Smith, who runs the Chemical Engineering Laboratory at EPFL in Lausanne, Switzerland, is an expert on using machine learning to discover new materials. Last year, after one of his graduate students, Kevin Mike Jablonka, showed some interesting results using GPT-3, Smith asked him to demonstrate that GPT-3 was actually useless for the kind of complex machine learning research his group was doing. predict the properties of compounds.
“He completely failed,” Smith jokes.
After a few minutes of tweaking with a few relevant examples, the model was found to perform as well as advanced machine learning tools specifically designed for chemistry, answering basic questions about things like the solubility of a compound or its reactivity. Just give it the name of the compound and it can predict various properties based on the structure.
