I’m surprised at how well the AI made the piece though. Had it not been for the tag bottom right it might have taken me a while to notice. It’s actually surprisingly consistent, even the reflection in the table bottom center is mostly correct. The thing that tipped me off (after the tag) was the vase on the mantle top right and one of the chains disappearing on the ceiling lamp top center (Edit: and just now the missing eye in the reflection bottom center).
4 Likes
Fair, but I prefer drawing planes so I’ll leave this one to someone else
1 Like
Skynet taking note of anyone hating on AI
5 Likes
I call it Roko like the Basilisk.
2 Likes
Oh god not this rationalist nonsense, please. Begone Scott Alexander!
That’s because LLMs do not know what they’re saying, like at all, they just (essentially) follow a flow chart of statistically linked words and phrases, with adjustable weights dependent on the input, thus can “make stuff up” or make obvious errors.
LLMs also pull heavily from reddit and similar, e.g. sources of opinions, not facts. Nor do they discriminate between factual information or fiction, e.g. wikipedia vs a fan-wiki for a book series.
2 Likes
Even AI knows that Britain can’t make a mobile tank to save its life.
bingo, calling it “AI” or an “Agent” is a gross exaggeration of what these tools are.
1 Like
Torsion Bar is not better than Hydropneumatic lmao
Gaijin infilitrating AI confirmed
1 Like
According to Gaijin, the stock Swift F.7 has a climbrate half as good as an Me-262. This is obviously a complete lie, because it climbs at 45m/s stock.
I really want to know how Gaijin gets those numbers.
1 Like
I didn’t even spot that one xd
And yeah, @AblativeKitten, while its not technically thinking, it is still funny to me that it can make mistakes like this with absolute confidence :)
