For those of you making use of Chat GPT, please bear in mind that you cannot rely on anything it says to be valid and true. Firstly, it is grossly out of date (contains data only up to 2021), and has limited access to research output. I asked it to “Write a 500 word overview of recent research on the biology of the jackalberry tree, and provide references “. I did this because I was curious what it would come up with, and I had just been reading a lot of literature on jackalberry trees. So I had enough knowledge to know what was wrong.

Some of the information it provided was correct, some terribly incorrect, some of it marginally correct but poorly composed, and all of the references were fake.

Let me repeat that, all of the references were fake.

The references look real, and one could easily be tricked into imagining them to be true. Here is one example:

Bortolomeazzi, R., Nespitali, R., & Favretto, A. (2019). The jackalberry tree: a keystone species in Zambezi Valley ecosystems. African Journal of Ecology, 57(1), 1-10.
It is 100% fake.

So believe Chat GPT at your own peril, and never use it for something that you don’t already have considerable expertise in. And for goodness sake, never cite its references unless you have

1. verified that they actually exist
2. verified that the citation is correct.

In another test, I asked it something fairly obscure, “Write a 400 word essay on Lithophylluym cuneatum, and cite references”. This is a species that I described, and published in an international research journal. The ChatGPT output was as close to 100% wrong as you can get. For example it claimed that “the distinctive pink, purple, or red coloration of L. cuneatum is due to the presence of pigments such as phycoerythrin and phycocyanin”. In fact, L. cuneatum is entirely unpigmented! There was absolutely not a single thing in the 400 words that was true. Nothing. Nix. And of course the reference were wrong or irrelevant, and the actual paper describing L. cuneatum was not cited.

The correct paper to cite would be:
Keats, D. 2006. Lithophyllum cuneatum sp. nov. (Corallinaceae, Rhodophyta), a new species of non‐geniculate coralline alga semi‐endophytic in Hydrolithon onkodes and Neogoniolithon sp. from Fiji, South Pacific. Phycological Research 43(3):151-160.

In all fairness, Chat GPT does not have full access to the scientific literature. However, it seems likely that not everyone realises this, and I can imagine the Internet is already full of misinformation masquerading as truth. One can only wonder about how much pure nonsense it produces, and that goes undetected because for the uninformed it seems true.

In relation to the title of this post, how wrong can you be? Wrong. Very wrong.