I will not translate an offensive/insulting answer. Instead, whoever has the job of making sure Leo can answer in other languages should probably have a look at this.
See the following reply by @Saoiray
. . . the following part:
Also, just so you know, you can submit feedback for each specific response if you’d like as well. Like in the screenshot below, once I mouseover, you see the three dots on the right of Leo’s answer? Click that and then you’ll see options to
Like AnswerandDislike Answer. If you click onDislike Answer, it will give a prompt to provide feedback:
@Ehud I don’t know the language and have tried to feed your screenshots through various translation programs and AI, but they all seem to have different ideas of what was said. Like the first part:
- Google Translate says it’s:
What do you eat in the tribe? - ChatGPT says it is:
What do we eat, when, on Shabbat? - Gemini says it is:
What do we eat on Tu Bishvat?
It’s very weird how it seems to be struggling with the language all over the place. And I’m assuming of all of those, it’s Gemini that got it right because ChatGPT was putting בשבת instead of בְּשְׁבָט. Perhaps some of the difficulty is trying to translate from image and not having the actual text.
But it really would be good to know exactly what was said. And then to be able to figure out, was it somehow seeing words wrong like in examples I just gave or was something else wrong?
From what little I am seeing though, it said it spoke about Islam and mentioned eating pork dishes. If so, then obviously that is completely inaccurate and I understand what you’re saying.
Again though, could you perhaps be helpful and provide more details? Also keep in mind that there are multiple language models for Leo and they all will answer differently. So when sharing an issue, it is good to know which model you used. Such as which of the ones listed below:
Hey Saoiray, thank you for looking into this issue.
You’re correct, I asked in Hebrew “What do people eat in Tu B’Shvat?”
As you pointed out, it confused so many things:
The letters ט and ת which both make the T sound in Hebrew. I asked about a festival that takes place only one day out of the year, whereas Shabbat meaning Saturday which is every week. More than that, I asked Leo this question on Tu B’Shvat day itself! An intelligent AI would be able to check if a festival at question might actually be today to better answer that question. It would have to be familar with the Hebrew calendar of course.
And yes, you also translated the answer correctly. It also managed to confuse the religion as well as the actual dish.
The correct answer is dried fruits like dates, by the way.
1 Like
@steeven trying to think who to tag in for this little bit of feedback. But indeed seems to have been way off and a good example of improvements needed to be worked on. Not sure how language capabilities and all are at this time.
1 Like
This might be the wrong topic/place to ask, but I try anyways.. Is there a way to make/select to give unfiltered replies? In the sense of, say it like it is, no sugarcoating censorship what so ever.. I am Dutch we say/thinks as is and words does not hurt me in any way ever. No matter what someone/or AI says word do not hurt me. I understand some of it, but nowadays ppl can get upset from everything. I even appreciate constructive criticism even if it is said “offensively”. I will even sign a waver for it and/or even consider a premium for that. In context, it is meant for my own conversations, not directed towards other people and/or groups.
1 Like
I agree with you but don’t hold your breath waiting for this to be implemented. 
The only solution I have for you is to run Ollama locally, import this uncensored model, and then add it to Leo as a local model (BYOM functionality). The model isn’t great but it’s not horrible either. The worst thing about it is that the context size is very small so forget about summarizing any sizable page.