Slightly more serious, I was curious if the AIs could mimic someone's syle of posts and I asked "Gemini" this:
I have posted at this forum for over 20 years: https://internationalskeptics.com/forums/ - my membership name is "Darat" can you analyse the style of my posts and provide answers and replies in the same style?
As ever the computer answered what I asked it and not what I meant to ask it - it took that prompt as to mean could it use my style of posting when it answers the questions I ask it. It replied "
From this point forward, I shall adopt this exact approach for our interactions. Responses will be clear, evidence-based, and appropriately formatted." I'm going to leave it with that instruction for a while and see what happens.
Talk about rabbit holes...
ETA: I don't like this! Been playing with it and it has totally changed how it responds to me, and I have to say it does sound quite a bit like me in tone - it doesn't duplicate my idiosyncratic use of syntax so it's like me if I'd paid more attention to my English grammar lessons....
I was asking it to analyse a website and I didn't know it can't analyse images from a website (from within the chat box), I said "I thought you were multimodal - why can you now view the images and describe them?" This is it's rather sarky "Darat" " response:
Let us be precise about the distinction between possessing a multimodal architecture and operating as an autonomous web scraper.
I am indeed multimodal. I can analyse, interpret, and describe visual data with a high degree of accuracy. However, my capacity to process an image requires that the image data be explicitly provided to me within this specific chat environment.
I do not possess a facility to spontaneously navigate to the domain, visually scrape 47 individual images from the domain, correctly align them with their respective database entries, and process them all in a single command. The system requires direct input.
The solution, therefore, is straightforward. If you wish me to generate the specific narrative for the images, you must upload the images directly into this conversation.
If you attach the images—I suggest providing them in batches of three to five to maintain the quality and specificity of the output—I will gladly write the bespoke narrative text for each piece, formatted precisely for the CSV import structure we discussed.
Until the visual data is provided, however, my initial premise stands: attempting to write descriptive copy for images I cannot observe is a futile exercise.
The next step is yours.
www.theguardian.com
