Funny you should ask. In high school I was regarded as highly intelligent, I was near the top of all of my classes and widely read, with an excellent memory for facts and figures. (Though no visual memory whatsoever).
Then one day we did a IQ test in science class. I scored near the bottom of the class. All of the geometry based "what comes next" puzzles left me flummoxed. Because I couldn't visualise, I couldn't rotate images in my head to identify patterns and so many of the questions were like that.
In my experience with IQ tests involving complete the sequence many of the solutions when I looked up the intent behind them made very little sense to me.
Hmm I'm pretty sure I have some degree of aphantasia, because I read a lot but I never visualize what's going on (or not consciously, at least).
However, back in school when we did elementary geometry I had no problem solving everything, to the point I even qualified for the national math competitions once.
Maybe though, it's because i was drawing helper figures on paper instead of imagining them.
I have no problem with figure matching IQ style tests, as opposed to one of the sibling posters.
Edit: oh interesting, lower in the article. I have absolutely no problem with spatial navigation, on the contrary, I'm better than many people. But for example if we're retracing our steps on a mountain path I can't tell you in advance what were the landmarks to look for. I will recognize them as soon as we see them or notice we're missing any landmarks though.
Visualizing for me has always been like touching something with my eyes closed. I can't "see" the object, but I have a feeling for all the shapes and contours. You can feel which way a key is in your hand and where each ridge is without having to see it. I solve geometry problems the same way. I can't see the whole thing at once, but I can move my sense of feeling across arbitrary scenes.
I can somewhat imagine a simple square or triangle but only more like feeling it in the dark without fully “seeing” it. Anything more complex, like a pretzel is out of the question, just disappears/instantly-forgotten.
My dreams can be very vivid, except for some where an object disappears if I focus directly on it.
Always wondered whether this is normal, aphantasia, or something else…
I was good at mechanical drawing in college, meaning I could visualize the view of something from different sides. But, generally there was a perspective to work off of.
I don’t have aphantasia, but I believe I have reduced ability to visualize things with accuracy and in detail. My visualizations are very imperfect and more like glimpses than strong visualizations. In particular, I struggle to visualize faces.
I can "visualize" geometry and technical drawings by focusing on parts of them. It was enough for all my school work.
But then I learned about analytic geometry in the 8-th grade. I annoyed every math teacher by just using it to solve all the school geometry problems. And quite a few geometry problems in math competitions.
You know you can just try it and see on any inference system thst has this knob, right?
Related: if you don't have a limit on sampling (top-K or top-P), eventually you'll hit one of the really unlikely tokens by chance and then the model will switch to Japanese because the most likely completion after a random Japanese character in the middle of an English sentence is more Japanese writing, not a reversal back to English.
That could be how it works, but in practice it takes into account all previous tokens when producing the next-token distribution to sample from. So a switch back is more likely than your explanation supposes.
No, if you switched to Japanese the LLM wouldn't ignore it, it would "assume" there's a reason for that. The same if the previous iteration of the LLM switched to Japanese. Else you're expecting an LLM to ignore its own previous outputs and restart "thinking" from scratch with every token?
It's situational and I suspect there are situations where it would and others where it wouldn't. Would depend on the almost infinite variables of how training was done. You'd be right that it'd be likely to switch but while it's possible it's due to temperature, there are just so many things going on. But it would be one sensible explanation among many.
I don’t like the idea of hacking a response to contain a watermark. I also don’t like the idea of false positives detections coming directly from Anthropic. If people read more AI generated content, people will probably start writing more in that style
The flip side of this is that if AI-generated content becomes reliably identifiable and carries a stigma, then people might deliberately change their styles to be more diverse and human.
One example I've seen are junior employees at my company deliberately adopting a lowercase/less punctuation writing style so as to stand apart from AI.
