Bully AI profile pictures if you ever see them. Don't engage them, treat them as beneath respect

And I'm only hearing about this now
i gotta find articles on this later

If AI is so willing to regurgitate information sourced from an April Fools blog post, then what other, much less obvious misinformation is it feeding to everyone?

Sweatshops As A Service

There are not enough words to describe how enraging this is

The updated free version of Microsoft Copilot recommends using MAS (Microsoft Activation Scripts) upon request, from a repository hosted on GitHub. These scripts are created for activating Windows 7-11 and Office applications quickly and without cost. Obviously, the scripts are not official and are an activation crack that violates the license.

Notably, Microsoft has not blocked these requests or restricted access to the MAS repository on GitHub, which is owned by Microsoft, despite its existence for a significant period.

When AI does something useful.

When presented with a picture of a giraffe with no spots, an AI model manages to recognise that it is a giraffe but also hallucinates a bunch of details like chairs, a person, and an umbrella. When asked what is unusual about the giraffe, the AI completely fails to recognise that the giraffe doesn't have spots. Other models also struggled with the Giraffe With No Spots image, even failing to recognise that there is an animal in the image at all.

Bonus fact: it is impossible to get an AI image generation model to create an image of a giraffe with no spots, because of course there's nothing like that in their training data – the only other recent spotless giraffe was born in 1972, long before digital photography. This one was born in 2023.

Source and further reading: https://www.aiweirdness.com/ai-vs-a-giraffe-with-no-spots/

imagine you pull up to get surgery
and the surgeon is a hippopotamus

ChatGPT , Grok , Gemini and other LLMs are severely affected by russian propaganda.

They have a disinformation network called “Pravda,” which aims to influence AI chatbots to promote “correct” narratives. Over the past year, they have produced over 3,600,000 articles , some of which are outright fakes, and most of which are half-truths with the right angle.

Unfortunately, search models are the most affected. They consider sites from the Pravda network to be relevant sources of information.