I am not sure what is more scary for me; the capabilities of Large Language Models (pros and cons) or their propensity to get things wrong. And wrong in this case is contextual. Is it wrong in fact because of error or a lack of capability? Or, is it wrong because of self-interest (and the term is used advisedly)? In both ways it can be uncanningly human.
Some time ago Tom Lander, Kingston, sent me the image above, of a Woodhouse's Toad, Anaxyrus woodhousii. A nice image, other than adding it to one of our species galleries I wondered how I might use it. About that time I was asked to review Google’s Gemini image generator. I was playing so I asked Gemini to take this image and (giving it a set of parameters) generate something. The image below is what I got back.
A Woodhouse’s Toad model for your home, complete with box and a computer showing the LiDAR imaging software which was used to generate the model. Complete with coffee cups and assorted cables. None of this ever existed. None of this is real.
We do not use AI generated material in any of our publications or on our websites - except in the April 1 press release we include in the Black Range Naturalist.