News

According to a new study by a team led by LOEWE top professor Iryna Gurevych from TU Darmstadt and Dr. Harish Tayyahr Madabushi from the University of Bath, UK, AI models such as ChatGPT are apparently less capable of learning independently than previously assumed. According to the study, there is no evidence that the so-called large language models (LLMs) are beginning to develop a general “intelligent” behavior that enables them to proceed in a planned or intuitive manner or to think in a complex way.

The research focuses on unforeseen and sudden leaps in the performance of the language models, which are referred to as “emergent abilities”. After the models were introduced, scientists found that they became more powerful with increasing size and the amount of data with which they were trained (scaling). For example, with increasing scale, the tools were able to recognize fake news or draw logical conclusions. This raised hopes that further scaling would make the models even better. However, there were also concerns that these capabilities could become dangerous, as the LLMs could take on a life of their own and potentially escape human control. In response, AI laws have been introduced around the world, including in the European Union and the USA.

However, the authors of the current study have now come to the conclusion that there is no evidence for the presumed development of differentiated thinking in the models. Instead, the LLMs acquired the superficial ability to follow relatively simple instructions, as the researchers showed. The systems are still a long way from what humans can do.

 “However, our results do not mean that AI poses no threat at all,” emphasized Gurevych. “Rather, we show that the alleged emergence of complex thinking abilities associated with certain threats is not supported by evidence and that we can control the learning process of LLMs well after all. Future research should therefore focus on other risks posed by the models, such as their potential to be used to generate fake news.

Importantly, the tendency of these models to produce plausible-sounding but false results - known as confabulation - is likely to persist, even though the quality of the models has improved dramatically in recent times.”

The study will be presented in August at the annual conference of the renowned Association for Computational Linguistics (ACL) in Bangkok, the largest international conference on automatic language processing.

Contact:

Prof. Dr. Iryna Gurevych

Hochschulstraße 10, S2|02 B110, 64289 Darmstadt

Phone: 06151/16-25290; e-mail: iryna.gurevych@tu-darmstadt.de

 Link to the study:

https://arxiv.org/abs/2309.01809