I know you've gone and stated your purported age (presumably to increase engagement which I'm not going to fault you for), but that being said, I'm still going to treat you like an adult.
Vibe coding is not a substitute for basic testing.
Just a few seconds of trying this - and it seems to be riddled with issues.
Many times the part of speech doesn't show - giving an inscrutable number instead:
πεπόνθατε
Part of Speech: POS 17
Gloss: -
ἐμαυτοῦ
Part of Speech: POS 12
Gloss: -
Also the "Translation" is checked, but the Inspector doesn't seem to give gloss/translation for 90% of any of the words that I tried clicking on to highlight.
When used on smaller devices, the layout shifts result in many of the buttons being completely misaligned as well. Again this would have been easily caught with a cursory visual inspection.
For a free tool, it's pretty clearly pushing the AI subscription aspect. Given how LLMs tend to do better with overrepresented languages (Spanish, English, Chinese, etc.), I'm not sure I'd trust my studies of ancient languages like Greek and Latin to OpenAI.
Fair criticism. I should have tested more thoroughly before shipping! I've fixed the POS displa issues you flagged and am working on expanding the lexicon. I really appreciate you taking the time to report them!!
You're right that basic testing shouldn't be skipped regardless of how the code was written.
Major unlock I see coming with the advent of AI is the ability to access knowledge that has dropped out of the sociosphere because of obscurity and language barriers. This is a cool move in the right direction.
I'm not saying this to be harsh. I think that the idea is good, and it's highly relevant to my interests. But I threw it a real softball, Zeus' famous "O popoi" paragraph from the Odyssey:
The translation it sent back was terrible. Completely unfit for purpose:
> "How you, men of Athens, have been affected by my accusers, I do not know; but I, for my part, almost forgot my own identity, so persuasively did they talk; and yet there is hardly a word of truth in what they have said. But I was most amazed by one of the many lies that they told—when they said that you must be on your guard not to be deceived by me,"
And that's all it returned. If that's what you get from a very famous and thoroughly analyzed passage, which has already been translated dozens of times, how can you trust it to analyze anything more obscure or complex? Frankly, I think it needs a lot more work.
Also, separately, I think that you should rewrite all of the text on the website, as passages like "A focused workspace for classical texts: clean typography, trustworthy glosses, and AI tutoring that nudges—never floods" are too obviously AI-generated.
> The AI feature is completely secondary as obviously, they are likely to give you wrong answers
I didn't check, but isn't the entire translation mechanism handled by an LLM?
Not many high schools do ancient Greek, other than a handful of private ones, so I would consider gearing it a lot more towards the Koine Greek market. Then you would be able to attract people from colleges and universities where people are doing it as part of a religious studies/divinity course.
Latin appears to be undergoing an unexpected revival of interest just now... So there nay be a bigger market than you think. To advertise it you could sponsor various relevant YouTube videos.
I know you've gone and stated your purported age (presumably to increase engagement which I'm not going to fault you for), but that being said, I'm still going to treat you like an adult.
Vibe coding is not a substitute for basic testing.
Just a few seconds of trying this - and it seems to be riddled with issues.
https://the-lexicon-project.netlify.app/app#plato-apology/17...
Many times the part of speech doesn't show - giving an inscrutable number instead:
Also the "Translation" is checked, but the Inspector doesn't seem to give gloss/translation for 90% of any of the words that I tried clicking on to highlight.When used on smaller devices, the layout shifts result in many of the buttons being completely misaligned as well. Again this would have been easily caught with a cursory visual inspection.
For a free tool, it's pretty clearly pushing the AI subscription aspect. Given how LLMs tend to do better with overrepresented languages (Spanish, English, Chinese, etc.), I'm not sure I'd trust my studies of ancient languages like Greek and Latin to OpenAI.
Fair criticism. I should have tested more thoroughly before shipping! I've fixed the POS displa issues you flagged and am working on expanding the lexicon. I really appreciate you taking the time to report them!!
You're right that basic testing shouldn't be skipped regardless of how the code was written.
Major unlock I see coming with the advent of AI is the ability to access knowledge that has dropped out of the sociosphere because of obscurity and language barriers. This is a cool move in the right direction.
I'm not saying this to be harsh. I think that the idea is good, and it's highly relevant to my interests. But I threw it a real softball, Zeus' famous "O popoi" paragraph from the Odyssey:
ὢ πόποι, οἷον δή νυ θεοὺς βροτοὶ αἰτιόωνται: / ἐξ ἡμέων γάρ φασι κάκ᾽ ἔμμεναι, οἱ δὲ καὶ αὐτοὶ / σφῇσιν ἀτασθαλίῃσιν ὑπὲρ μόρον ἄλγε᾽ ἔχουσιν, / ὡς καὶ νῦν Αἴγισθος ὑπὲρ μόρον Ἀτρεΐδαο / γῆμ᾽ ἄλοχον μνηστήν, τὸν δ᾽ ἔκτανε νοστήσαντα, / εἰδὼς αἰπὺν ὄλεθρον, ἐπεὶ πρό οἱ εἴπομεν ἡμεῖς, / Ἑρμείαν πέμψαντες, ἐύσκοπον ἀργεϊφόντην, / μήτ᾽ αὐτὸν κτείνειν μήτε μνάασθαι ἄκοιτιν: / ἐκ γὰρ Ὀρέσταο τίσις ἔσσεται Ἀτρεΐδαο, / ὁππότ᾽ ἂν ἡβήσῃ τε καὶ ἧς ἱμείρεται αἴης. / ὣς ἔφαθ᾽ Ἑρμείας, ἀλλ᾽ οὐ φρένας Αἰγίσθοιο / πεῖθ᾽ ἀγαθὰ φρονέων: νῦν δ᾽ ἁθρόα πάντ᾽ ἀπέτισεν.
The translation it sent back was terrible. Completely unfit for purpose:
> "How you, men of Athens, have been affected by my accusers, I do not know; but I, for my part, almost forgot my own identity, so persuasively did they talk; and yet there is hardly a word of truth in what they have said. But I was most amazed by one of the many lies that they told—when they said that you must be on your guard not to be deceived by me,"
And that's all it returned. If that's what you get from a very famous and thoroughly analyzed passage, which has already been translated dozens of times, how can you trust it to analyze anything more obscure or complex? Frankly, I think it needs a lot more work.
Also, separately, I think that you should rewrite all of the text on the website, as passages like "A focused workspace for classical texts: clean typography, trustworthy glosses, and AI tutoring that nudges—never floods" are too obviously AI-generated.
> The AI feature is completely secondary as obviously, they are likely to give you wrong answers
I didn't check, but isn't the entire translation mechanism handled by an LLM?
Hiya! Thanks so much for trying it out The translations for texts already on the site are pulled from Persues.
Imported texts shouldn't be translated. You were seeing the first bit of the Apology:
> ὅτι μὲν ὑμεῖς ,ὦ ἄνδρες Ἀθηναῖοι ,πεπόνθατε ὑπὸ τῶν ἐμῶν κατηγόρων ,οὐκ οἶδα · ἐγὼ δʼ οὖν καὶ αὐτὸς ὑπʼ αὐτῶν ὀλίγου ἐμαυτοῦ ἐπελαθόμην ,οὕτω πιθανῶς ἔλεγον . καίτοι ἀληθές γε ὡς ἔπος εἰπεῖν οὐδὲν εἰρήκασιν . μάλιστα δὲ αὐτῶν ἓν ἐθαύμασα τῶν πολλῶν ὧν ἐψεύσαντο ,τοῦτο ἐν ᾧ ἔλεγον ὡς χρῆν ὑμᾶς εὐλαβεῖσθαι μὴ ὑπʼ ἐμοῦ ἐξαπατηθῆτε
It seems like the import feature is broken entirely as I'm testing it. Thank you so much for pointing me to it!!
Not many high schools do ancient Greek, other than a handful of private ones, so I would consider gearing it a lot more towards the Koine Greek market. Then you would be able to attract people from colleges and universities where people are doing it as part of a religious studies/divinity course.
Latin appears to be undergoing an unexpected revival of interest just now... So there nay be a bigger market than you think. To advertise it you could sponsor various relevant YouTube videos.