It will make the AI powered features (context aware translations, context aware explanations, reader review summaries, ask AI) noticeably slower.
If this graduates from experimental mode, it will almost certainly be as part of a more expensive premium plan option, since it costs more to run. For the time being, any premium user can enable it.
That’s awesome Steve! I am happy to pay more for more accurate and powerful AI models. I think gpt4o is significantly better than the mini so it’s worth it in my opinion.
I have a quick question about the context translations ( not the explanations feature). I noticed that in Albanian it’s not using the gpt4o for the translations but it looks like it’s still using the mini, but for the explanations it’s now using the new model. I have ticked the box for the new model. I noticed because the mini is less accurate for Albanian than gpt4o and the explanations feature is doing a much better job than the translation.
Is it not available yet for all languages in context translations ? Thanks!
You’re absolutely right! Good catch! I’ve fixed it now so that when betterAIModel is enabled it will use the full gpt-4o model for the context translations as intended.
Thanks Steve. I just tried it out and it seemed to work and it got the translations right, but I have the remove flashcard option enabled when you deselect a word ,and when I translated the same words/phrases again it seemed it reverted back to the mini and the translations were different and more inaccurate.
Any suggestions?, I’m not sure if the problem is on my end. I have all the boxes ticked in the settings.
If you actually deleted the word then it shouldn’t be using the old translation. My two best guesses are that a) the gpt-4o model is returning the same result or b) the word still happened to be in your word list. (There is also a decent chance of an option (c) which is that there’s a bug making it behave differently from what I intended.)
If (b) was the reason, then I’ve just fixed that issue. What would happen before, is that if you translated a word which was already in your word list again in a different context, it would just return the translation you had already saved to your word list. This made sense for the simple context unaware translations, but for the new context aware translations I’ve just changed this behaviour so that it will do a new lookup to ensure you are getting a translation that makes sense given the context.
Thanks for the quick reply Steve. So I have no words in my saved words list and I get very different translations from the explanations and the context translations. The explanations gives me the correct translation, I’ll try and attach the screenshots. When I select the correct translation from the explanations now it seems like it’s overwritten by the context translations and shifts in second place( hope that makes sense). Just to show you one example I’ve attached photos. This happens with nearly all words. When I try and delete the context translations meaning it just pops back up as the default translation.
Thanks so much, it’s working now! Big difference in the accuracy. Is deleting the cache something I need to do or can do by myself for future reference? I’m not sure how it works on readlang, I thought deleting the words from the list did just that.
No, that was on my end. After fixing the bug where it was using the mini model instead of the full one I forgot that I’d been caching (saving) the results in my database. I do this to avoid repeat requests to ChatGPT in the case where the exact same word + context + model is being used.