Playing with Meta AI in a Sandbox

Today I decided to download and play with Meta AI but I didn’t use my Facebook or Instagram accounts. I used a brand new account with no connection to me. I don’t trust Facebook and the Meta “pattte blanche”, especially with AI, given how AI can manipulate us without us realising.

I asked it about Mindf*ck and genocide and it answered to both questions without skirting them. It did try to confuse me with a disambiguation question with the book. Initially it felt as if it wouldn’t answer, until I clarified that I meant the book, rather than swearing gratuitously.

I then asked it about the Facebook rebrand to Meta to avoid taking responsibility and whitewashing and for this it gave a more corporate answer.

Being English I then asked it about the weather and then about whether I should cycle or go to an Ingress First Saturday event, and for this it said “the weather will be grey and drizzly so avoid cycling. That is the forecast for tomorrow but rain is meant to come in the evening, not the morning. It failed to ask “At what time are these events taking place?”.

When I asked it to describe a photo it did. I wouldn’t rely on it for alt text descriptions, mainly because it’s verbose, and doesn’t sound confident.

I then shared a photo of meerkats with “fluffy feather skirts” dressed for Valentine’s day and it inferred the context correctly.

For added fun I asked it about the current pollen situation and it was right. It used French and Swiss sources and when I used Lac Léman it kept using that name rather than switching to Lake Geneva. Obviously it has Vaud roots.

I asked it about the key news stories for day and it seems to have given a neutral answer, probably because it doesn’t know enough about me to give a manipulative, biased news summary.

I shared photos of a spider to see if it would be able to recognise it, and it hedged its bets. Google said “That’s a Nosferatu spider” and Meta AI said “It could be A, B, or C” but it didn’t use the nosferatu nickname.

And Finally

When I asked it about context splicing it gave me the same answer as Euria, the Infomaniak search engine. Specifically it said “Nope, not an issue for me because I can switch between context A, B, C and D without issues. That’s paraphrased. I haven’t pushed far enough to get it to hallucinate yet.I like to see how “self-aware” AI is. I don’t trust Facebook and its apps, therefore I don’t trust Meta AI.