Conversation
buherator
buherator
I recently saw a lot of examples of using LLMs when you could get away e.g. with and API call, but now I think I found the perfect example of LLMs being more niche then even skeptics (like myself) think they are:
Even skeptics have to admit that LLMs are very good at natural language translation. @kagihq introduced a fast ("Standard") #LLM for its translation service that seems to fail miserably if you try to translate single words for less common languages:
https://kagifeedback.org/d/9373-standard-translation-is-unusable-for-hungarian
My point is that doing a dictionary lookup for all words (~1mil for English) could be done on a disposable vape in no time with better results, incl. clear indication of lookup failures, so you can fall back to your GPUs when needed.
Even skeptics have to admit that LLMs are very good at natural language translation. @kagihq introduced a fast ("Standard") #LLM for its translation service that seems to fail miserably if you try to translate single words for less common languages:
https://kagifeedback.org/d/9373-standard-translation-is-unusable-for-hungarian
My point is that doing a dictionary lookup for all words (~1mil for English) could be done on a disposable vape in no time with better results, incl. clear indication of lookup failures, so you can fall back to your GPUs when needed.
0
1
0