Nowadays I’m experimenting with ChatGPT, Bard and Bing (GPT-4). To generate reading materials.
So far my favorite prompt (for Chinese) is: “Make a long, casual conversation between friends in Chinese, hanzi only, about XYZ.”
ChatGPT 3.5 makes around 200 word long conversations, Bing with GPT-4 400 words but once it did a 900 word long.
I usually do narrow reading, so I ask the same topic for 4-5 times one after another. This way it uses similar words but different conversations.
Today I was reading about Euro Truck Simulator 2. I can tell that Bing made really nice texts and the fact that sometime it wondered away from the fact that it’s just a simulator, so started to talk about real personal topics, like food, petting animals along the way, made these conversations really enjoyable.
Give it a try! Do you have favorite prompts?
My most used chat gpt prompt currently is:
- Can you explain the grammar in this target language sentence: your sentence
Others I use or have used:
can you summarize the following article in target language. Also provide the key vocabulary and the english translation of that vocabulary. Here is the article with title and author included: your article
for the target language summarization you did above, could you do it in the style of two Germans have a conversation about it
Create a dialogue in target language of someone looking to buy a train ticket (or specific topic/subject of interest). Please use all of the typical phrases and collocations.
Can you do the target language dialogue, but only use A2 level vocabulary and grammar?
for output practice…type this in: “Give me a random text of 100 words in English at the B2 level that I will translate back to you in target language and you will then provide me with a correction and necessary explanations.”
I often forget to use it, I prefer natural interactions so far, and I don’t trust hallucinations. However, a part from things already said, I like to use it for stories to import to LingQ when I’m bored.
What I like is that the story is dynamic. I can ask chatGPT to continue the story in the way I want to, I can add characters, eliminate characters, and so on. I leave one chat dedicated for one story, and in any moment I can ask to continue for a bit, and increase the level of language difficulty.
However, I don’t like that they made it too soft.
This story continuation thing sounds good. Newer tried before. Thanks for the idea!
@davideroccato, what do you mean they made it “too soft”?
@ericb100 I mean that you cannot ask too much from the dynamic of the story, to make it darker for example, because chatGPT starts to answer that it cannot say this or doing this because it’s against the policy, and so on, and so on.
Even simple things like: “kill this character in the story”. ChatGPT says: “I cannot kill them because bla bla bla”. So you have to say: “ok, eliminate this character from the story”. And chatGPT it’s going to find a way to put the character out-of-the-picture in a softer way.
Not sure if the use of “soft” is correct, but you can suggest me a better way to define this. You could also try yourself to check if the limitation is because I’ve been using the free version, and maybe a paid user with GPT-4 has more room about it.
Softly is good. I figured you meant something along the lines of what you describe, but wanted to make sure.
By the way, you may or may not know, but there’s a song from the '70’s named “Killing me softly (with his song)”. Maybe you just tell chat gpt to kill them softly =D
ha ha ha, yes, I know that song. I might try that one next time, I still have that story going on.
Hallucinations are still rather frequent, especially when explaining points in grammar in certain languages. The way I deal with it is usually to ask for the explanation, then ask for examples, then point out that only about 1/3 of the examples actually fit the explanation, at which point ChatGPT will correct itself and give you a more accurate explanation
For story telling the hallucinations are really helpful, makes interesting stories!
Can You write a dialogue at (A2, B1, B2, C1 etc) level in German by using this list of words “fertig, erledigt, …etc”
For example, if I am listening to a podcast in German in my laptop I keep writing unknown words under this prompt while listening to the podcast at the same time" Once I am done listening to the podcast I simply create the dialogue and import it to LingQ , read it right away by looking up the meanings of these unknow words. It has become almost second nature to me.
I haven’t encountered hallucinations in my language queries – maybe I don’t know enough yet to tell!
I have encountered hallucinations in other cases. I’ll push back and ask for specifics. It’s amusing how ChatGPT backs down and thanks me for clearing up the confusion.
My most frequent language query is – Explain “this” vs “that” in French.
This is a really great exercise. For me it was very sobering… I estimate my reading level in German to be about B2, but my output is nowhere near that. I love the way it tries to be diplomatic with the corrections- “your German level is quite good, but here are 50 mistakes you made” lol.
I do think that the sentences GPT is outputting are a little above B2 though. Much of the vocab it asked for is rated C1 in Seedlang, for example. I also find asking it to do a sentence at a time, rather than 100 words is a bit more approachable (for my level, at least).