Skip to main content

Artificial intelligence hallucinations

The Original Article was published on 18 April 2023

The Original Article was published on 25 February 2023

Dear Editor,

The anecdote about a GPT hallucinating under the influence of LSD is intriguing and amusing, but it also raises significant issues to consider regarding the utilization of this tool. As pointed out by Beutel et al., ChatGPT is a promising large language model (LLM), i.e., a text-based virtual assistant, with knowledge derived from vast training data updated to 2021. It cannot directly access internet data, and, at the moment, it does not have any knowledge after this date. Nonetheless, ChatGPT can retain the information the user provides during a conversation to improve its responses to subsequent questions and inquiries. Indeed, even if it does not have the ability to learn like humans, it can analyze and reprocess what it has learned during the conversation, depending on the complexity of the prompts and the language used. This may account for the divergent response patterns elicited by our prompts compared to those of Beutel et al.

Nevertheless, as stated in the correspondence and previously by Azamfirei et al. [1], the content of the figure in our article, which aimed to compare different studies on a specific intervention, is substantially incorrect. Even though ChatGPT, with which the text was written, does not meet the authorship criteria [2], we are ultimately always responsible for the accuracy of the information provided. Additionally, while using this tool, we have often encountered another type of tricky AI “hallucinations,” which involves apparently correct bibliographic references with known authors and coherent titles, but which are entirely non-existent. Alternative AI tools could be used for this objective, allowing for exploring current scientific databases.

As it should always be done with any step in scientific production and already highlighted in our article, an expert in the field should validate the content generated by ChatGPT, even if it were a simple translation into English from another language. The worldwide enthusiasm and awareness of this tool's benefits and risks bring to mind the Dunning-Kruger effect [3], potentially applicable in the setting of scientific writing, as illustrated in Fig. 1.

Fig. 1
figure 1

A revised Dunning-Kruger effect may be applied to using ChatGPT and other Artificial Intelligence (AI) in scientific writing. Initially, excessive confidence and enthusiasm for the potential of this tool may lead to the belief that it is possible to produce papers and publish quickly and effortlessly. Over time, as the limits and risks of ChatGPT and other AI are learned, as well as the complexity of their functioning with the need for specific prompts, enthusiasm and confidence decrease. As this awareness increases, ChatGPT and other AIs can become effective and supportive tools in scientific writing, such as computers and internet search engines, finally achieving a conscious and correct usage

Finally, as a colorful detail, we report the case of Kary Mullis, who used LSD and engaged in conversations with an extraterrestrial entity disguised as a raccoon. Still, Mullis' groundbreaking invention of the polymerase chain reaction (PCR) revolutionized medicine, which earned him the Nobel Prize in Chemistry [4]. What would ChatGPT achieve under the influence of LSD? Only time will tell. (However, we do not recommend its consumption).

Availability of data and materials

Not applicable.


  1. Azamfirei R, Kudchadkar SR, Fackler J. Large language models and the perils of their hallucinations. Crit Care. 2023;27(1):1–2.

    Article  Google Scholar 

  2. Salvagno M, Taccone FS, Gerli AG. Erratum: Correction to: Can artificial intelligence help for scientific writing? (Critical care (London, England) (2023) 27 1 (75)). Crit Care. 2023;27(1):99.

    Article  PubMed  PubMed Central  Google Scholar 

  3. Kruger J, Dunning D. Unskilled and unaware of it: How difficulties in recognizing one’s own incompetence lead to inflated self-assessments. J Pers Soc Psychol. 1999;77(6):1121–34.

    Article  CAS  PubMed  Google Scholar 

  4. Carlson P. Nobel Chemist Kary Mullis, Making waves as a mind surfer. The Washington Post. Published November 3, 1998.

Download references


Not applicable.

Author information

Authors and Affiliations



MS, AGG and FST wrote the main text and agreed in its final form. All authors read and approved the final manuscript.

Corresponding author

Correspondence to Michele Salvagno.

Ethics declarations

Ethics approval and consent to participate

Not applicable.

Competing interests

The authors declare that they have no competing interests.

Additional information

Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit The Creative Commons Public Domain Dedication waiver ( applies to the data made available in this article, unless otherwise stated in a credit line to the data.

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Salvagno, M., Taccone, F.S. & Gerli, A.G. Artificial intelligence hallucinations. Crit Care 27, 180 (2023).

Download citation

  • Received:

  • Accepted:

  • Published:

  • DOI: