An inaccurate or incorrect response of a Large Language Model, sometimes referred to as "Hallucinations", is attributed to an error in the code not the training data which represents input material manipulated by the code to give a desired output.
The prompt as well is not responsible for the inaccuracy of the output which it triggers given that the LLM is a general chatting application and hence an improvised inquiry should yield at least a correct response for an error free code.
The process of releasing updated versions of such LLMs with the aim of achieving higher accuracy or more intelligence could be represented as the maintenance part of a software development lifecycle to fix errors and increase reliability of an AI system.
An AI is a computer system which executes instructions by the algorithm through utilizing available resources of input data and computing infrastructure
Dear Mohamed el Nawawy ,
Large language model (LLM) systems, such as ChatGPT1 or Gemini2, can show impressive reasoning and question-answering capabilities but often ‘hallucinate’ false outputs and unsubstantiated answers,. Answering unreliably or without the necessary information prevents adoption in diverse fields, with problems including fabrication of legal precedents5 or untrue facts in news articles6 and even posing a risk to human life in medical domains such as radiology..
https://www.nature.com/articles/s41586-024-07421-0
Regards,
Shafagat
@Shafagat Mahmudova
Thank you so much for your contribution
I agree that these false outputs pose high risks and prevent adoption especially in critical fields.
In the context of Large Language Models (LLMs) like GPT-4, "hallucinations" refer to instances where the model generates outputs that are factually incorrect, misleading, or nonsensical. These can occur despite the model's ability to produce fluent and coherent text. Here’s how to view and understand these hallucinations:
Understanding and addressing hallucinations in LLMs is crucial for developing more reliable and effective AI systems, ensuring that their outputs are both accurate and useful.
AI hallucinations can’t be stopped — but these techniques can limit their damage
Developers have tricks to stop artificial intelligence from making things up, but large language models are still struggling to tell the truth, the whole truth and nothing but the truth...
"Ask an artificial intelligence (AI) chatbot for a scientific reference and it might give you the citation you’re looking for. Or, it could ‘hallucinate’, spitting out an only partially correct, or totally false, answer. Researchers say that preventing AI from erring altogether is impossible, but they’re working on measures that could make hallucinations less frequent and less problematic. These include making a chatbot refer to a trusted text before responding and conducting “brain scans” of a large language model’s artificial neurons to reveal patterns of deception. For now, the best defence might simply be to take the output of a chatbot with a large pinch of salt..."
https://www.nature.com/articles/d41586-025-00068-5
- Badges
- Science method
- Similar topics
- Biological Science
- Ecology