Teuken-7B-Base and Teuken-7B-Instruct: Towards European LLMs

Re:Teuken-7B-Base and Teuken-7B-Instruct: Towards European LLMs by mrweasel on Tuesday April 15, 2025 @08:26AM (#43691751)

Someone apparently did observe ChatGPT (I think it was ChatGPT) switch to Chinese for some parts of it's reasoning/calculations and then back to English for the final answer. That's somehow even weirder than the LLM giving different answers depending on the input.

[ Parent | on HN ]

Re:Teuken-7B-Base and Teuken-7B-Instruct: Towards European LLMs by laurent_du on Tuesday April 15, 2025 @09:41AM (#43692574)

Reminds me of this funny video: https://www.youtube.com/watch?v=NY3yWXWjYjA ("You know something has gone wrong when he switches to Chinese")

[ Parent | on HN ]

Re:Teuken-7B-Base and Teuken-7B-Instruct: Towards European LLMs by ApolloFortyNine on Tuesday April 15, 2025 @09:10AM (#43692168)

I've seen this happen as well with o3-mini, but I'm honestly not sure what triggered it. I use it all the time but have only had it switch to Chinese during reasoning maybe twice.

[ Parent | on HN ]

Re:Teuken-7B-Base and Teuken-7B-Instruct: Towards European LLMs by Telemakhos on Tuesday April 15, 2025 @09:57AM (#43692757)

I've seen Grok sprinkle random Chinese characters into responses I asked for in ancient Greek and Latin.

[ Parent | on HN ]

Re:Teuken-7B-Base and Teuken-7B-Instruct: Towards European LLMs by andai on Tuesday April 15, 2025 @11:12AM (#43693959)

I get strange languages sprinkled through my Gemini responses, including some very obscure ones. It just randomly changes language for one or two words.

[ Parent | on HN ]

Re:Teuken-7B-Base and Teuken-7B-Instruct: Towards European LLMs by genewitch on Tuesday April 15, 2025 @02:40PM (#43696825)

Is it possible the "vector" is more accurate in another language? Like espirit d'esclair or schadenfreude, or any number of other things that are a single word in a language but paragraphs or more in others?

[ Parent | on HN ]

Re:Teuken-7B-Base and Teuken-7B-Instruct: Towards European LLMs by sanxiyn on Tuesday April 15, 2025 @06:57PM (#43699401)

Possibly. I have seen Claude switching to Russian for a word or two when it is about revolution!

[ Parent | on HN ]

Re:Teuken-7B-Base and Teuken-7B-Instruct: Towards European LLMs by numpad0 on Tuesday April 15, 2025 @04:04PM (#43697704)

Isn't it just it getting increasingly incoherent as non-English data fraction increases?

Last I checked, none of open weight LLMs has languages other than English as its sole dominant language represented in the dataset.

[ Parent | on HN ]

Re:Teuken-7B-Base and Teuken-7B-Instruct: Towards European LLMs by maxloh on Tuesday April 15, 2025 @09:18AM (#43692261)

> the LLM giving different answers depending on the input.

LLMs are actually designed to have some randomness in their responses.

To make the answer reproducible, set the temperature to O (eliminating randomness) and provide a static seed (ensuring consistent results) in the LLM's configuration.

[ Parent | on HN ]

Re:Teuken-7B-Base and Teuken-7B-Instruct: Towards European LLMs by lolinder on Tuesday April 15, 2025 @09:23AM (#43692350)

It's not necessary in most inference engines I've seen to set the temperature to 0—the randomness in the temperature is drawn from the seed, so a static seed will work for any temperature.

[ Parent | on HN ]

Re:Teuken-7B-Base and Teuken-7B-Instruct: Towards European LLMs by jll29 on Tuesday April 15, 2025 @09:38AM (#43692529)

The influence of the (pseudo-)random number generator is called "temperature" in most models.

Setting it to 0 in theory eliminates all randomness, and instead of choosing one from a list of next words that may be predicted, always only the MOST PROBABLY word would be chosen.

However, in practice, setting the temperature to 0 in most GUIs does not actually set the temperature to 0, but to a "very small" value ("epsilon"), the reason being to avoid a division by zero exception/crawsh in a mathematical formula. So don't be surprised if you cannot get rid of random behavior entirely.

[ Parent | on HN ]

Re:Teuken-7B-Base and Teuken-7B-Instruct: Towards European LLMs by Asraelite on Tuesday April 15, 2025 @02:12PM (#43696479)

> the reason being to avoid a division by zero exception/crawsh in a mathematical formula

Why don't they just special-case it?

[ Parent | on HN ]

Re:Teuken-7B-Base and Teuken-7B-Instruct: Towards European LLMs by ricochet11 on Tuesday April 15, 2025 @10:15AM (#43693043)

i’ve seen that with deepseek

[ Parent | on HN ]

Re:Teuken-7B-Base and Teuken-7B-Instruct: Towards European LLMs by rzz3 on Tuesday April 15, 2025 @12:53PM (#43695468)

I saw Claude 3.7 write a comment in my code in Russian followed by, likely from a previous modification, the English text “Russian coding” for no reason.

[ Parent | on HN ]

Re:Teuken-7B-Base and Teuken-7B-Instruct: Towards European LLMs by jananas on Tuesday April 15, 2025 @01:23PM (#43695859)

In had it doing the reasoning in Turkish and English despite the question being in German.

[ Parent | on HN ]