I just started to test this new feature and I think it’s awesome, Could it have a more natural-sounding voice, like the text-to-speech from Google?
Really loving this guys. Thank you for all the hard work you must have put into it.
The only question I have is, is there a way to make her less flirty? I find it a bit off-putting to always be called “mon cherie” or to be asked things like what my most romantic French phrase is
Brilliant idea, but his talking is like a robot not active like Google translation sound, need to make good effort on that to complete this fantastic idea, thank you.
I love the chat bot, but I’d love a separate tab/option that only corrects my language mistakes and do not chat with me. The chat bot language correction feature is absolutely insane, the fact that I can mix between my native language and my second language is really helpful, but I don’t want to be chatting with the bot to use this feature. Could you guys make it a separate tab? or maybe an option so that the chat bot does not talk and only corrects writing mistakes. Keep up the great work!
Good stuff. I would like option to turn off the audio completely though!
Still a bit weird! Is ChatGTP too expensive to integrate?
Very cool feature! I’ve been playing with this for Italian and overall it’s quite useful. I have noticed some odd interactions and was wondering if I should screenshot to give feedback?
For example, we were talking about shrimp and somehow the bot got confused and started talking about griffons.
Similarly- getting confused between subtitles and submarines.
Or the most recent was the subjunctive and conjunctions.
Hello, awesome feature! but could you please remove these extra words between the asterisks in the bot’s speech. for example giggles or something like this
Awesome, I like that I’m able to have the conversation translation already in every text just in case I need to look for a better understanding. Also, integrating it with my vocabulary of known words inside Language Reactor is so awesome. oh and I can have furigana on top of my kanjis too, because of this integration
Sadly, is still not working for me. Communication happens in English (as the post says it) but the 3 times I have tried, my conversations on the Japanese side don’t make sense at all, meanwhile, if I ask ChatGPT to have a conversation with me in Japanese, it will have a real conversation in Japanese without any translation in English happening. So, I will have to come back to ChatGPT for this conversation practice and miss all of the great features that this chat could provide.
I really really hope there is a way the conversation can ignore the English translation. oh, and also, the voice is really bad. I thought we were already leaving behind those robotic voices because I’ve seen AI voices with a more natural way of talking, but I haven’t tried other languages than Japanese, so this might be only in some languages, including Japanese? I don’t know. Also, it should be able to recognize that the conversation is happening as a “written” type of text where you add things like “wink, giggles, laugh” and those kinds of emotes, and if you are gonna speak that conversation, you should ignore those parts.
In general, it has a lot of potential, especially because of the integration with LR, but still not working for me, ChatGPT still gets me better results when practicing Japanese
Thank you for the hard work!
I want to suggest exporting the chat as text (html, excel - like with subs). There are a few useful sentences I would like to take a look later, but if I close the chat I believe I’ll loose everything there.
Thanks
Unfortunelly, she isn’t so good in italian
Love the feature! I can’t play the sound though when using any mobile browser. It would also be great to choose from different accents in the future. Maybe start just with male or female. The LLM is also a bit weak at times. I guess the GPT 4 API is still too expensive…
Hey, you may want to deploy this new model: Mistral 7B.
- Outperforms Llama 2 13B on all benchmarks
- Outperforms Llama 1 34B on many benchmarks
- Approaches CodeLlama 7B performance on code, while remaining good at English tasks
- Uses Grouped-query attention (GQA) for faster inference
- Uses Sliding Window Attention (SWA) to handle longer sequences at smaller cost
We’re releasing Mistral 7B under the Apache 2.0 license, it can be used without restrictions.
Mistral’s demonstration of a small model delivering high performance across a range of tasks could mean major benefits for businesses.
For example, in MMLU, Mistral 7B delivers the performance of a Llama 2 that would be more than 3x its size (23 billion parameters). This would directly save memory and provide cost benefits – without affecting final outputs.
The company says it achieves faster inference using [grouped-query attention (GQA)](arxiv org/pdf/2305.13245.pdf) and handles longer sequences at a smaller cost using Sliding Window Attention (SWA).
OMG - just tried it - it was so fun! She was a ‘psychologist’ and I got to tell her about my frustration with learning spanish. it was very therapeutic! and then we played a game with vocabulary - and that was fun too! - way to go!!!
Audio for the chat feature still not working… Tried Chrome, tried Edge, tried Brave. Even tried on my phone.
No audio gets produced, although when the dictionary for individual words is opened, that specific word is uttered.
Is this supposed to be a pro feature or something?
Hi @J_32,
Sound is not tied to the Pro Plan/premium. However, I’m not sure if the chat feature has sound.
I’d recommend emailing support:
I quickly did try the ChatBot in my mother language so I can see the quality of answer. In only six sentences there were two weird incorrect sentences. One sentence in a way that I never heard of in my language (wrong transalation, too literal). And the other sentence as switching to very formal talk although before it was very informal. Seems like the model is not so good.
But awesome interface. Really like that its intertwined with all the vocabulary and sentence functionality. Awesome.
Btw, why you didnt just use GPT4 as a model? (sorry if its a noob question :-D)
It’s nice. I like how it says a bunch of phrases back at you. This makes it more interesting than other chat bots where you are kind of lost on where to take the conversation.
Looks like the “play” button is now removed. It used to be there, with the obvious implication that it was for the text-to-speech feature, but it’s not there anymore, so I suppose it’s still a work in progress before implementation.
Yes @J_32, I was also noticing that the TTS play button had also been removed from the Text/Reader mode on my android phone version of the PWA (pseudo-app) and I messaged support about it.
I wonder why it’s missing on some things (like the chat feature) maybe they are working some things out with it like you mentioned in your comment.