Open Mycheze opened 1 year ago
This enhancement should really be on the greatest priority list in my opinion. Achieving this will put Lute on steroids, and can possibly become THE greatest app there is for language learning.
I was actually thinking about this again today. I agree that this feature would be so killer and make Lute 10x more useful. Some more ideas and thoughts:
I don't trust GPT-4 or any other AI to be accurate enough (it's known to just straight-out lie and make stuff up) for any language that doesn't already have good dictionaries.
It's good with English and other major languages. We have the dictionaries and grammars. It's inaccurate with smaller languages - we'd rather use something more accurate. What's the actual value here?
GPT-3.5 is much worse than 4. I've only seen 4 make a few mistakes with Czech. I've heard it's much worse for even smaller languages, but Czech isn't very big.
I disagree that most languages have good dictionaries and grammars that are actually useful for the average language learner. Frequently, looking up a word requires you to do 3-4 more lookups to sorta understand what you looked up. And I personally find "linguistic grammars" to be totally useless. I'd much prefer a short and simple explanation from GPT-4.
I don't care if it's occasionally wrong. Natives are as well. I've gotten conflicting (and flat out wrong) grammar explanations from native speakers. I use what it says as a guide to help my understanding. Plus, that single look up is less than .0001% of my time spent with the language.
If it's wrong 20% of the time or more, then yeah, you should probably use something else. But incorporating this feature earlier to straighten out bugs and problems will make it even more useful with GPT-5 and other models come around.
This enhancement should really be on the greatest priority list in my opinion.
So hard to say what the right priority is. There are some things which are "just features", which don't really push the envelope. Possibly this chatgpt is one of those features which does push the envelope, but there are others too. Some of the other pending requests feel more like foundational things, which should get priority over this, the biggest of which is #99 (changing how data is stored, which has many knock-on effects such as sharing books/vocab lists, reducing some parsing inconsistences, etc).
The challenges I see offhand with chatgpt -- or essentially any other service:
These are all just questions to explore the issue, not reasons to rule it out this idea. Cheers all, jz
So, I fiddled around a bit and without any substantial work, I got this sweet result. Just hosted another python server and used curl and OpenAI api, and added the localhost url as a dictionary link. chatgpt acts whimsically btw, sometimes gives a nice conjugation list and sometimes not :(
Whole sentence selection :) (no change in code or prompt)
Feel free to use as a makeshift feature: https://github.com/Shantibhushan95/lute-gpt-server
Hey there @Shantibhushan95 that looks pretty great! I’ll check out the code, thank you.
Also, @jzohrab I guess the Discord invite link is broken
Also, @jzohrab I guess the Discord invite link is broken
Whoops! What was the link you tried?
The one under Getting Started. https://discord.com/invite/CzFUQP5m8u Nevermind, logged in from Safari and it worked
Is your feature request related to a problem? Please describe.
Many small issues with texts or dictionaries are easily solved by ChatGPT and a few prompts. But copy/pasting things between Lute and ChatGPT can be cumbersome and slow. It would be great to have the integrated a little.
Describe the solution you'd like
I would like to see a few ChatGPT features.
Additional context
Here're some example prompts that I've been using:
For defining stubborn words:
For reformatting: