GPT-3 transforms chemical research
Artificial intelligence is growing into a pivotal tool in chemical research, offering novel methods to tackle complex challenges that traditional approaches struggle with. One subtype of artificial intelligence that has seen increasing use in chemistry is machine learning, which uses algorithms and statistical models to make decisions based on data and perform tasks that it has not been explicitly programmed for.
However, to make reliable predictions, machine learning also demands large amounts of data, which isn’t always available in chemical research. Small chemical datasets simply do not provide enough information for these algorithms to train on, which limits their effectiveness.
In a new study, scientists in the team of Berend Smit at EPFL, have found a solution in large language models such as GPT-3. Those models are pre-trained on massive amounts of texts, and are known for their broad capabilities in understanding and generating human-like text. GPT-3 forms the basis of the more popular artificial intelligence ChatGPT.
The study, now published in Nature Machine Intelligence, unveils a novel approach that significantly simplifies chemical analysis using artificial intelligence. Contrary to initial skepticism, the method doesn't directly ask GPT-3 chemical questions. “GPT-3 has not seen most of the chemical literature, so if we ask ChatGPT a chemical question, the answers are typically limited to what one can find on Wikipedia,” says Kevin Jablonka, the study’s lead researcher. “Instead, we fine-tune GPT-3 with a small data set converted into questions and answers, creating a new model capable of providing accurate chemical insights.”
