According to a new study posted on arXiv, many large language models can have biases just like humans and can analyze complex data to form their own opinion. The study said that many LLMs like ChatGPT-4 and Llama can form their own biases about a controversial topic if they are prompted in a way that aligns with their coding results. If the user asks them to be analytical or skeptical, LLMs can analyze a topic thoroughly and can provide insights that a human may overlook.
Dr. Awais Hameed Khan from the University of Queensland led this research and said that LLMs can help assist human researchers. He and his team of researchers have also introduced a novel design toolkit named the AI Sub Zero Bias Cards which can help researchers use LLMs as analytical tools. The toolkit comprises 58 cards which are related to reformatting and reframing prompts to use LLMs for analysis.
Large Language Models can easily make sense of a controversial topic or any topic for research by using their cognitive, reasoning and philosophical skills. It doesn't mean that AI models should be used to replace human interpretations on different topics. But due to the analytics skills of AI models, many researchers can benefit from their assistance.
Image: DIW-Aigen
Read next: Study Shows AI Models are Not Capable of Performing Tasks Which Are Beyond their Training
Dr. Awais Hameed Khan from the University of Queensland led this research and said that LLMs can help assist human researchers. He and his team of researchers have also introduced a novel design toolkit named the AI Sub Zero Bias Cards which can help researchers use LLMs as analytical tools. The toolkit comprises 58 cards which are related to reformatting and reframing prompts to use LLMs for analysis.
Large Language Models can easily make sense of a controversial topic or any topic for research by using their cognitive, reasoning and philosophical skills. It doesn't mean that AI models should be used to replace human interpretations on different topics. But due to the analytics skills of AI models, many researchers can benefit from their assistance.
Image: DIW-Aigen
Read next: Study Shows AI Models are Not Capable of Performing Tasks Which Are Beyond their Training
