Hello everyone,
how is it going :)?
So a couple of weeks ago, I showed you the example generator tool that I made with ChatGPT.
It then had a bug for quite a while, but now it’s working again and I’ve also improved it a little bit.
But not only that – I also created two new tools for you.
Drumroll please….
- AI Story Generator
- AI Proofreader
And I’ll show you both of them today, so let’s take a look :).
Let’s start with the proofreader.
ChatGPT checks your sentence
This tool lets you enter a sentence or a short paragraph and give you a corrected version of it.
The correction is usually pretty good, but as everything with ChatGPT, it is NOT 100 reliable. So sometimes it overlooks a mistake or even adds a new one, especially if your sentence is perfect. So take it with a small grain of salt.
Or actually… if the answer seems weird, you could just copy paste it back into the input, and see if it corrects itself again.
In fact, I’ll probably add this as a button. Like… “double check”. That should catch pretty much all mistakes then.
Anyway, it’s already a nice and quick way to check something.
***
***
Now, I know that many of you would love to have some feedback about WHY something was changed. And I have played around with this a bit, but the analysis that ChatGPT gives just wasn’t reliable enough, in my opinion. Like… its explanations don’t always relate to the actual mistakes and the reasoning is sometimes just nonsense.
I’ll work with this some more, and try out different versions of GPT, because it would be really cool to get some more info about the corrections, but for now it’s really just that.
Cool.
And now the second one.
Reading Comprehension with ChatGPT Stories
This is kind of the example generator on steroids. So instead of a short example, the AI creates a full short story for you, and you can pick the difficulty and the genre.
Just enter a topic in German or English (German is better). That can be a single word, or a whole setup like
“Maria finds women’s underwear in Thomas’ drawer and confronts him about it.”
Then, pick a difficulty and a genre and you’re good to go. It’ll come up with a short story and also give you a translation.
The creation sometimes takes a couple of minutes, especially if you pick “complex” and a more specific genre. But if it takes more than 5 minutes, it probably froze up.
Now, the stories are usually pretty okay, but some genres work better than others and the “beginner” option is often a bit too difficult, in my opinion. I’m still working on how best to communicate to the model what exactly it is that you want.
It actually works better with a less specific topic, so if you just say “eine Reise” for example. But play around and find out.
And as per usual, there is a non-zero chance that the German contains a mistake. But overall, it’s a fun feature if you want to get some quick reading practice about something that actually interests you :)
***
***
So yeah, these are my two new tools for German and I’m absolutely going to make more :).
And yes, we’re working to make these into an actual app.
I’ll also add them to the menu, so you can always find them easily on the website.
But yeah… let me know in the comments how you like them and if you think they’re helpful. Oh and let me know if you have any ideas for features. I feel like the option to print or save a story would be nice, for example, and I’m sure you have other ideas.
Anyway, I really hope you enjoy playing around with this, even if you’re skeptical about these new AI tools.
Have a great week and I’ll see you in the next one.
Hey Emanuel, I’m posting this from Iran. unfortunately I couldnt access your website earlier today. (it just didnt load- I tried to refresh page for 20mins and still didnt work) and I had to use VPN to open this. I had the same problem yesterday and I’m afraid if this happens more in coming days cause I find your website and your texts excellent.
p.s: website now loads ok after turning off my VPN. I hope it doesnt happen tomorrow.
I’ve got some other reports from reader from Iran about this.
It was “mostly” blocked a few months ago already.
There’s nothing I can do though. It’s really up to you to use a VPN or download pdfs if you have access.
Thanks for these tools! On the sentence correction, I compared this tool with using DeepL to translate from German to English and back. The score on two sentences was 1-1. So I tried a tiebreaker.
The first example, a grammatically correct German sentence, came back uncorrected in DeepL. In this tool, it was “corrected” from past to present tense:
Der magische Horn eines Einhorns leuchtete weit und breit.
corrected to:
Der magische Horn eines Einhorns leuchte weit und breit.
DeepL 1, Chat GPT 0.
However, this tool corrected an error in a different sentence that DeepL didn’t know enough to correct. The sentence was:
Die Augen des Einhorns untersuchten jeden Baum, als es schließlich ein Rundel Löwen entdeckte.
This tool corrected Rundel to Rudel.
DeepL “corrected” Rundel to Rondell. Now, I did notice the problem in DeepL, because I could see the English translation of the (non-existent?) word Rundel was roundel, and I knew I wasn’t trying to say “a roundel of lions.” However, to find the correct word via DeepL, I needed to go to the Löwe entry in Wikipedia and read it until I found the word, because my dictionary doesn’t have a translation of pride as Rudel. So this ChatGPT tool, by giving me the correct word right away, saved some time and effort.
DeepL 1, ChatGPT 1.
Tiebreaker time!
Input sentence:
Als es seiner Chance sah, stürmte er auf der Löwen zu.
ChatGPT: Als er seine Chance sah, stürmte er auf den Löwen zu.
DeepL: Als er seine Chance sah, stürzte er sich auf den Löwen.
Both saw the inconsistency of es/er and decided to go with er.
Both corrected seiner to seine.
Both saw the inconsistency of auf der Löwen and decided to go with auf den Löwen (instead of the equally plausible auf die Löwen).
DeepL decided to change the verb from stürmen to stürtzen, for some reason. Is that an improvement? From my dictionary, it seems either could be fine in this sentence. ChatGPT kept the verb the same.
So–I’m not sure. If stürtzen is clearly better than stürmen, I’ll give this one to DeepL. If either is fine, I’ll give this one to ChatGPT for not making an unnecessary change.
Anyway, it’s fun to play with!
Wow, this was amazing to read!! Thanks a lot for doing and sharing this!
I LOVE these kinds of tests for capabilities and I’ve done quite a few myself before deciding it was “good enough” to offer it to learners.
The main problem with ChatGPT is that it “invents” corrections sometimes and if there are too many mistakes, it’ll miss one, probably.
But I think that can be fixed by a dialogue option or a re-prompt like “are you sure?” and then it’ll check its own output. I think I’ll add this at some point to the app we’re currently building for this.
As for changes in phrasings… I think it’s bad for a tool like this, because to a learner it’ll imply that “zustürmen auf” is wrong for some reason, but it’s not. Also “zustürmen auf” and “sich stürzen auf” are not the same thing, so DeepL changed the meaning and that’s not good for a correction tool.
I have never used DeepL, but I think they are in panic mode right now, because out of nowhere a big competitor to their business emerged.
Lastly… your test examples were really well done, especially the last one!!
Thanks for the feedback on my feedback! Good point on it not being good behavior for a correction tool to change the meaning, even if either word could be appropriate. (The bit about either word being appropriate is still just my surmise from checking a dictionary; I realize that your comments neither confirm nor deny it.)
“stürzen auf” pretty much implies a direct attack.
“zustürmen auf” just implies that you approach with vigor and high velocity, but maybe still being 50 meters away and maybe it actually stops.
Big overlap, but depending on the context, interchanging them might lead to continuity mistakes.
Playing around with it and reading others examples, it seems like the characters always do good, have nice experiences, obey the rules. The computer has not yet learned to be bad, maybe?
I think that’s a mix of OpenAIs attempt at steering the model to be good and the general fact that most stories end in a good way and have a “hero” protagonist.
I am prompting the model to create a short story with several parameters, so that alone is enough to have a high likelyhood of “good” main characters. Maybe I’ll add anti-hero or no happyend as options :)
Interesting observation! It holds true for the story I got, with the prompt, “The hungry unicorn stalks a pride of lions on the veldt.” In the beginning, “Es durchstreifte das veldt, auf der Suche nach einem Festmahl.” But in the end, the unicorn just chases the lions away and is very proud of itself for doing so–no fictional lions were harmed in this story.
Haha, I think that’s just because “horses” (even the augmented ones) usually don’t eat lions, so the probability of it actually attacking and eating the lion (in terms of word probability) is rather low, no matter what the prompt is.
But I will add a “sad ending” option and see how that goes :).
I like the way you try to get inside ChatGPT’s “head” by thinking about probabilities. I suppose a counter-argument is that it’s probable that anything stalking lions in search of a feast actually has the capability of killing and eating a lion :-), but I have no idea how ChatGPT frames its determinations of probability.
Also, it occurs to me that though the unicorn was proud of its lion-scattering prowess, it was still hungry at the end of story, even though this went unsaid. So not a completely happy ending for everyone!
It’s an entirely probabilistic machine that goes from word to word making each decision based on millions of parameters word before, sentence, paragraph, your input, etc) that at times pull in different directions.
If you have a consistent narrative, then you’ll get a really coherent output, but if there are some subtle contradictions in the context, different forces will pull in different directions and one will come out on top.
I’ve done quite a few experiments with it, and my favorite one is this:
You can clearly see that it has no concept of random and no real concept of “if”.
Your input activated certain parts of the model, that’ll make it very likely that the third position is “article”, and it “forgets” the task of words having to start with “z”.
I would say, if you run the same unicorn prompt 10 times (in fresh instances), you have a 1/10 chance that the unicorn eats the lion.
Viel Spaß. Es hat für mich einen Krimi über einen Hund geschrieben, der Orangen mochte. Es war wirklich so dumm, dass es lustig war.
Cool, freut mich zu hören :)
Both tolls are helpful and fun to use. Danke sehr
Freut mich zu hören :)
Both tools are fun and useful. I wrote a couple of sentences in German knowing I was probably making errors, and it figured out what I was saying and corrected it.
For the story I said I went to a beach and everybody else was naked – beginner – erotica. It came back with a nice little “PG” story where I take off my clothes, run to the water, come back, put on my clothes and drive home. I like that you have both the German and English translation for the generated story. Great Job Emanuel or maybe I should give all the credit to ChatGPT auf seiner guten Seite zu bleiben.
Thanks for the nice feedback :)
Can I have it written in Perfekt?
I am trying to work basic speaking skills
I tried adding instructions in the Your topic box and it worked!
I typed:
I couldn’t find my keys when I was leaving for work. Use Perfekt tense
Und die kleine Geschichte über unseren lieben Emmanuel und ein Einhörnchen: Emmanuel war ein junger Mann mit einer tiefen Verbindung zur Natur. Eines Tages war er im Wald, als er auf einen bezaubernden Anblick stieß – ein Einhorn. Es hatte eine majestätische Schönheit, die ihn gefangen nahm. Er konnte seine Augen nicht von dem magischen Geschöpf abwenden und bald spürte er eine starke Verbindung zu ihm. Er dachte, es sei wirklich das wunderbarste Geschöpf, das er je gesehen hatte. Bevor er es wusste, hatten sich seine Gefühle für das Einhorn von Bewunderung und Ehrfurcht zu Liebe geändert. Er wollte für immer bei ihm sein und jeden Tag die Schönheit der Natur gemeinsam erkunden. Er konnte nicht glauben, dass ein Geschöpf wie das Einhorn seine Liebe und Zuneigung erwidern konnte, aber es schien so zu sein. Jedes Mal, wenn er es sah, kam das Einhorn näher und ließ ihn es streicheln und ihm Liebe zeigen. Emmanuel dachte, es sei ein Zeichen, dass seine Liebe erwidert wird. Er schätzte jeden Moment, den er mit dem Einhorn verbrachte, und sie verbrachten oft stundenlang zwischen den Bäumen. Emmanuel konnte nicht glücklicher sein und er fühlte, dass sein Leben durch die Liebe eines Einhorns gesegnet wurde.
Absolut unrealistisch :).
Das Modell weiß offensichtlich nicht, was für wilde, passiv-aggressive, drogensüchtige Monster Einhörner eigentlich sind.
Mein Satz: “Ich frage mich, ob Emmanuel alle diese doofe Sätze lesen wird, nur weil er noch krank ist, und hat nichts besseres zu tun, oder weil er wirklich doch ein bisschen spinnt.” Correction: Ich frage mich, ob Emmanuel alle diese doofen Sätze lesen wird, nur weil er noch krank ist und nichts besseres zu tun hat, oder weil er wirklich doch ein bisschen verrückt ist. –Interessant wie die Redewendung “er spinnt” zu “er ist verrückt” geändert wurde.
That happened because I am currently having the model translate the input to English and then back-translate it to German.
I found that this gave better results than asking directly for a correction.
But I’m going to try a different model of GPT soon, and hopefully that’ll improve things.
Ideally, it should really just correct mistakes and not add modifications that aren’t necessary.
Wow. I really like the story generator. I had it generate a story about my love of Kaffee. One of the sentences came back as “Ich kann nicht ohne meinen Kaffee.” <kind of leaves you hanging huh?> I put that sentence into the OTHER KI to check it, and it knew what it was supposed to say and added the actual verb it needed.
I’m a bit late to this comment, but I hope you see it.
I don’t know if this was “on purpose” by gpt, but it’s actually a fairly common colloquial way to phrase this.
The verb is missing, and there’s no definite verb implied. So I wouldn’t know what to put there. Could be anything from living, to working, to driving.
never seem to get the 2 free posts a week
Try refreshing the page or go to the article directly from the homepage. That usually fixes it.
These are brilliant. I’m super impressed
Thank you!!
I tried Beginner and Sci-Fi, and the topic “Kaffee”. I got a brilliant story about the future of coffee. And wasted a half hour of my coffee break. Needless to say, I love this feature.
Sweet, I’m happy to hear that!
Text Check – ich bin geil
What response did you get :)?
Hallo Emanuel — Ich habe in “Text Check” geschrieben — “Meinem Hund geht es gut.” Es antwortete — “Correction: Meinem Hund geht es gut. – My dog is doing well. – Mein Hund geht es gut.” Was sagt mir das Programm dass “Mein Hund geht es gut” richtig ist?
This is an unfortunate case where it actually creates a mistake. The original sentence was 100% correct, and the tool then tends to make stuff up.
I hope this will improve when I switch to a better gpt model, but it’ll never be perfect.
I liked this exchange with the first tool. My prompt was “Wie sagt man, “Ich spreche Deutsch” auf ukrainische?” and it came back with, “Correction: Ich spreche Deutsch heißt auf Ukrainisch „Я говорю по-німецьки“.”
Wow, I didn’t know it would do that :D
But check this out: It came up with this wacko story about a magnolia tree, and then forgot to translate the second half of it.
Die schönste MagnoliaEs war einmal ein kleines griechisches Dorf am Fuße eines hohen Berges, wo jeden Frühling eine wunderschöne Magnolie blühte. Obwohl der kleine Baum überall im Dorf bewundert und geliebt wurde, war er besonders für die Bewohner des größten Hauses im Dorf, einem prachtvollen alten Herrenhaus, etwas ganz Besonderes. Jeden Tag kamen die alten Herren aus dem Haus und saßen lange unter dem Baum, träumend und voller Erbarmen. Es dauerte nicht lange und die Nachricht vom Wunder des blühenden Baums verbreitete sich allmählich in der Region und über seine Grenzen hinaus. Eines Tages, als der Baum schon den ganzen Frühling hindurch blühte, kam ein großer Wagen in das Dorf gerollt, der von einem Mann gelenkt wurde, der als großer König des Landes bekannt war. Der König sah den großen und schönen Baum und war so beeindruckt, dass er ihn unbedingt haben wollte. Nachdem der König seine Meinung bekannt gemacht hatte, konnte nur eine Lösung gefunden werden: Der Besitzer des Herrenhauses, der alte Herr, versprach dem König, dass er ihm den Baum überlassen würde, wenn dieser verspräche, dass jeder, der in seinem Königreich eine Magnolie blühen lässt, an dem Tag, an dem sie blüht, Zugang zu seinem Palast erhalten würde. Der König willigte ein und der alte Herr übergab den Baum zufrieden an den König. Der König brachte den Baum zu seinem Palast und pflanzte ihn dort, wo er den ganzen Tag über die schönsten Blüten tragen konnte, und jedes Jahr, wenn die Magnolie blühte, machte sich der König mit seinen Gefolgsleuten auf den Weg, um ihrer Schönheit zu huldigen und sein Versprechen einzuhalten. Dieser Tag wurde in den Annalen des Königreichs als der Tag der Magnolie verewigt und war immer noch eine sehr beliebte Tradition sogar Jahrhunderte später.
The Most Beautiful MagnoliaOnce upon a time, there was a small Greek village at the foot of a tall mountain, where a magnificent magnolia bloomed every spring. Though the little tree was admired and beloved by everyone in the village, it was especially special to the inhabitants of the largest house in the village, a beautiful old manor. Every day the old people from the house came out and spent long hours under the tree, dreaming and filled with compassion. It didn’t take long for the word of the tree’s miraculous bloom to spread gradually throughout the region and beyond its borders. One day, when the tree had been blooming throughout the whole spring, a great carriage rolled into the village, driven by a man known as a great king of the land. The king saw the tall, beautiful tree and was so impressed that he wanted it at any cost. Once the king had made his opinion known, only one solution could be found: the owner of the manor, the old man, promised the king that he would give him the tree if he promised that anyone who let a Magnolia bloom in his kingdom would have access to his palace on the day it blooms. The king agreed and the old man gave the tree up contentedly to the king.
Actually, I have given it a token limit for the answer, and this story ended up so long that the token limit was reached while creating the English translation.
I’ll probably remove that cap. It’s not really necessary.
This was “complex” story, right? Some of the sentence structure sounds like it.
Es funktioniert ziemlich gut aber man muss dem Bot interessante Themen geben, weil er ganz einfallslos ist.
Ja, das stimmt :)
Wo erscheint die Geschichte? Ich habe es drei Mal versucht.
Die erscheint in dem kleinen Fenster unter dem Button.
Probier bitte nochmal, ich glaube jetzt funktioniert es.
I can’t read the AI article as it says I can only read two articles a week. That’s all fine but I’ve currently read zero! Maybe you should get some AI on your website….
Please try and reload the page. That usually fixes it.
Or copy this into the url directly.
https://yourdailygerman.com/chatgpt-learn-german-tools/