bombobuffoon wrote:At one point I constructed a sentence that was perhaps too difficult for me.
And so I constructed the sentence and to my utter surprise ChatGPT scored the sentence correct 10/10.
I was suspicious of this and give it to a native to check.
In return I received a bit of a verbal clip around the ear for producing a "sentence with significant errors".
Yes, ChatGPT has some very significant limitations. It's a poor proofreader, and it's bad at explaining grammar, and it's usually blind to its own mistakes. Once it makes a mistake, it's almost impossible to correct.
It's a pretty good writer, and a decent translator (if you ask it correctly). Competing models like Claude 3 add the ability to summarize long texts well.
So if you're going to use ChatGPT, you need to play to its strengths. For major languages, you can ask it, "How could I say the following in ______?", and it will often do fairly well. Interestingly, the more expensive models are also pretty good at questions like, "I am a language learner. Please explain what the words __________ are doing in the text below." But keep it concrete, and don't ask for detailed grammar explanations.
And Finnish will make things a bit harder. Even for ChatGPT.
EDIT: Here's
an example of what I can get Anthropic's Claude 3 Opus to do with some panels from a graphic novel, in a language with lots of training data.
bombobuffoon wrote:To add to that I understood several other pages that also had a context that was set outside the scene. Perhaps this is some form of wider progress, or is it just yet another false dawn? The minor gains certainly has been thematic during the past 3 years. I have had to fight over each word, sentence, concept in this language. There has been no magic moments, only false dawns. I could write a whole post about that. Any victories have been minuscule, like this one. I suppose I should just enjoy it.
Graphic novels are fantastic. Yes, the pictures and the context can help you understand things, but you still internalize that understanding
anyways. I have gone far out of my way to take advantage of helpful context to temporarily boost my understanding, and had very good results.
As for sudden breakthroughs, some of the fastest and most dramatic progress I've ever made still took 30-60 hours of hard work. My knowledge of French was built one page, one episode, one flashcard and one conversation at a time. Some people have been lucky enough to get sudden epiphanies, but I never did.
If an hour a day for a month produces a noticeable improvement in one area, you're doing great.