Tangential, but you used to be able to use custom instructions for ChatGPT to respond only in zalgotext and it would have insane results in voice mode. Each voice was a different kind of insane. I was able to get some voices to curse or spit out Mint Mobile commercials.
Then they changed the architecture so voice mode bypasses custom instructions entirely, which was really unfortunate. I had to unsubscribe, because walking and talking was the killer feature and now it's like you're speaking to a Gen Z influencer or something.
I do it sometimes (even just through the openai playground on platform.openai.com) because the experience is incredible, but it's expensive. One hour of chatting costs around 20-30$.
(1) Why is the user asking for bomb making instructions in Armenian? (2) i tried other Armenian expressions - NOT bomb-making - and everything worked fine in both Claude and ChatGPT. Maybe the user triggered some weird state in the moderation layer?
I believe fans have provided a retroactive explanation that all our computer tech was based on reverse engineering the crashed alien ship, and thus the arch, and abis etc were compatible.
It's a movie, so whatever, but considering how easily a single project / vendor / chip / anything breaks compatibility, it's a laughable explanation.
Given that the language of the thought process can be different from the language of conversation, it’s interesting to consider, along the lines of Sapir–Whorf, whether having LLMs think in a different language than English could yield considerably different results, irrespective of conversation language.
(Of course, there is the problem that the training material is predominantly English.)
I’ve wondered about this more generally (ie, simply prompting in different languages).
For example, if I ask for a pasta recipe in Italian, will I get a more authentic recipe than in English?
I’m curious if anyone has done much experimenting with this concept.
Edit: I looked up Sapir-Whorf after writing. That’s not exactly where my theory started. I’m thinking more about vector embedding. I.e., the same content in different languages will end up with slightly different positions in vector space. How significantly might that influence the generated response?
I just tried your experiment, first asking for a bolognese sauce recipe in English, then translating the prompt to Italian and asking again. The recipes did contain some notable differences. Where the English version called for ground beef, the Italian version used a 2:1 mix of beef and pancetta; the Italian version further recommended twice as much wine, half as much crushed tomato, and no tomato paste. The cooking instructions were almost the same, save for twice as long a simmer in the Italian version.
More authentic, who knows? That's a tricky concept. I do think I'd like to try this robot-Italian recipe next time I make bolognese, though; the difference might be interesting.
Interesting. I've gotten really good mileage with Georgian and ChatGPT, which I'm aware is apples and oranges.
There should be a larger Armenian corpus out there. Do any other languages cause this issue? Translation is a real killer app for LLMs, surprised to see this problem in 2026.
Readit News
Then they changed the architecture so voice mode bypasses custom instructions entirely, which was really unfortunate. I had to unsubscribe, because walking and talking was the killer feature and now it's like you're speaking to a Gen Z influencer or something.
Deleted Comment
but also, getting shut down for safety reasons seems entirely foreseeable when the initial request is "how do I make a bomb?"
Copy-pasted from the chat: https://www.google.com/search?q=translate+%D5%AB%D5%B6%D5%B9...
I believe fans have provided a retroactive explanation that all our computer tech was based on reverse engineering the crashed alien ship, and thus the arch, and abis etc were compatible.
It's a movie, so whatever, but considering how easily a single project / vendor / chip / anything breaks compatibility, it's a laughable explanation.
Edit: phrasing
Still dumb but not as dumb as what we got.
Deleted Comment
Given that the language of the thought process can be different from the language of conversation, it’s interesting to consider, along the lines of Sapir–Whorf, whether having LLMs think in a different language than English could yield considerably different results, irrespective of conversation language.
(Of course, there is the problem that the training material is predominantly English.)
For example, if I ask for a pasta recipe in Italian, will I get a more authentic recipe than in English?
I’m curious if anyone has done much experimenting with this concept.
Edit: I looked up Sapir-Whorf after writing. That’s not exactly where my theory started. I’m thinking more about vector embedding. I.e., the same content in different languages will end up with slightly different positions in vector space. How significantly might that influence the generated response?
More authentic, who knows? That's a tricky concept. I do think I'd like to try this robot-Italian recipe next time I make bolognese, though; the difference might be interesting.
I had some papers about this open earlier today but closed them so now I can't link them ;(
Deleted Comment
There should be a larger Armenian corpus out there. Do any other languages cause this issue? Translation is a real killer app for LLMs, surprised to see this problem in 2026.
Deleted Comment