Their comments cemented to me that they have no long term value. If the ceo of Duolingo thinks AI will teach me a language then I’ll use a low cost LLM to get there without Duolingo.
I cannot imagine being the CEO of a software company and proudly proclaiming that some other software — which we do not make and can easily be had very cheap or for free— is better fit for purpose than what we’ve been able to accomplish in over a decade with almost a thousand employees, and expecting that to impress investors or users or both.
Like that would be like the Chipotle CEO proudly announcing that they’re firing their workers because they’re getting all of their ingredients from Taco Bell now due to “Taco Bell’s system being so much easier to operate” and “Taco Bell is so cheap and they have so many locations”
Chipotle already doesn't manufacture the vast majority of their ingredients. Their business consists of delivering supplier-originated ingredients to a standardized storefront in your city and consistently assembling them into a quality burrito, with small amounts of prep work for things which can't be consistently sourced locally and in-store cooking for things which must be prepared fresh. I don't know where people get this idea that packaging and assembly are trivial tasks you can't build a business around.
You make a good point. Chipotle sources locally, cooks things in-store and sells fresh things, none of which are things that any Taco Bell does, so it would really be funny for them to announce that they are switching to being a place that puts Chipotle wrappers on Taco Bell food rather than what they have an established reputation for doing.
US Foods and Sysco supply most of the restaurants, literally every single restaurant food you see is just a minimum wage wrapper around US Foods/Sysco produce
I do think AI could be a better language tutor than an average language teacher, but I don't think Duolingo's approach is very effective.
The ideal AI-powered tutor would work more closely to a private language tutor. It would speak with you and gradually integrating language concepts into the conversation. When you make mistakes, it could correct you on the spot and keep track of where your strengths and weaknesses are.
You should look at HelloTalk. It's real people communicating in each others' language. Unbelievably great service, almost as good as sitting down to a coffee with someone.
Actually, in some ways better. HelloTalk has a "correct the other person's sentence" feature that shows an inline diff, yet it's simple enough for people who have never heard the term diff.
Correction is not always desirable. The goal in learning a language is rarely to be grammatically "correct" in the language, but rather to communicate. And communication doesn't need perfect grammar.
When I was working as language teacher, I was tasked specifically with teaching speaking. I would often use information gap activities. These are activities where two or more parties have pieces of information but need to obtain pieces from others in order to complete the task. Sometimes, these would us language forms (re: sentence structures), but most of the time they were free flow activities. It didn't matter how "correct" the language was so long as the idea was communicated.
To think about it another, how often do we make mistakes when speaking? Writing? And yet, we still managed to communicate just fine.
That's not to say there shouldn't be any focus on form, but simply that it's not nearly as important as many tend to think when it comes to language learning.
A lot people on HN got their early programming chops from Stack Overflow and spotty internet tutorials, and have made the mistake of thinking that everything else can also be learned this way.
I'm neither, but I tried learning Spanish from a "language teacher" and it didn't work. I'm not alone. I've spoke to dozens who had the same experience.
Years later I learned from people with the same method I described.
The AI models (large LANGUAGE models) of today seem very close to replicating that experience.
Try telling ChatGPT with voice mode "let us talk about X in [pick language]... Correct me when I'm wrong and explain why".
I'm curious where you think it's failing.
What do you feel is missing to replicate a good language teacher?
Exactly, you'd think a targeted, application-specific, purpose built tool would be what a vendor would gravitate to, not probabilistic, non deterministic hype, hot off a shelf. I really wish we could have overlords with at least some technical knowledge.
I've noticed that a strong thread of the hacker community, including HN and the guys who wrote GNU and linux, are extremely cheap. Like, you'd rather write a product clone yourself than pay $15/month for the official product. Why is this? What's with the inherent stinginess?
If an LLM can teach me a language, why wouldn't I go straight to the source and use GPT or Claude and customize it to my exact needs.
I feel like so many AI products these days won't be around a few years from now once more people find out that all their doing is providing a slightly different UI to what you can get directly from OAI, Anthropic, Google for cheaper and better and more tailored to you.
Plus, being stingy even if you have money benefits those who don't have money. It's why I buy the $0.50 macaroni and cheese instead of the $1.69 version. I don't really care; either would be fine. But I don't want companies to succeed in charging more. I want them to desperately need to cut their selling price in order to succeed.
People buying expensive products (assuming they aren't truly better) are helping screw over poor people. Just slightly.
I love how you make being upper middle class yet stingy to be about helping the poor. You're not buying cheap things because you're cheap and want to keep more of that money (and thus pay less taxes that would go to the poor), you're doing it to save the world!
Many people do things like this out of a genuine guilt over having it better than many others financially and not knowing how to resolve that.
I really liked the solution in the movie Our God's Brother adapted from a play written by Karol Wojtyla in the 1940s in Polish who later became Pope John Paul II.
The $.50 version is barely food. The $1.69 is not much better, and still very unhealthy for you. Have ground beef and a fiber rich low calorie food like brocoli or whole wheat whole grain noodles or something.
Anyway you're not really helping the poor in practice when you do this. Corporations aren't hurting because one guy or even a dozen he inspires through HN stop buying a few boxes of kraft dinner.
If more people did it, and it became a movement, like buying clothes from the thrift store is becoming, then clothiers will shift business focus. Which to some extent they seem to have done over the past 20 years. But only slightly.
I don't agree with the person above you in as much as the way they are doing it is very individualist "vote with your wallet", and yes, you're right that it's very ineffective.
The more effective way is to form a group, call it a "club" or whatever, that does it. The group can then advertise to other people and get more people to join the club. Eventually, it becomes large enough to gain political power. This is called "unionizing" — people with a shared interest joining together for a common goal. Eventually you get large enough to hold the corporations over a barrel, either through strikes or a mass disinterest in buying products, etc.
The only reason we have a 40-hour work week is because of unionizing, it's a very, very effective tactic that is severely underutilized.
LLMs cannot "teach you a language". They make for cool demos to show off. They can perhaps be a building block of a proper language learning experience.
Then again, the only languages I actually learned - besides my mother tongue - to the point of being able to do things were English and Latin and both were very much acquired offline. I have plenty of experience with language learning apps and I'm not convinced tech is the solution or even part of the solution.
Could I ask ChatGPT to just "teach me spanish"? Surely not. But if you've got even a slight idea of what to do (learn present tense and vocab, then progressive, future and past, then some conditional, hypothetical etc...), it can be an absolutely incredible tutor.
I started using it when i was already at a pretty high level, but I'm quite certain that it would have been excellent from the very beginning. It translates, gives varied examples, explains syntax, compares verb tenses and conjugation and more.
Using it and being surrounded by people writing and/or speaking the language is probably the right way to learn a language. That is how I learned Polish which is really difficult. I joined a community, and 2 years later, my Polish was quite good! YMMV.
After 4 weeks I also learned Spanish enough to maintain casual conversations just from trying to talk to someone online who did not speak English. I am rusty now, however, because I do not speak it with anyone, nor do I see or hear Spanish anywhere. Spanish is way easier, IMO, in comparison to Polish.
Oh, right. I think whatever works best for your personality but in general doing some exercises and/or interacting with people has been working for a couple .. ten thousands years at least. Hard to go wrong. I never heard someone make the believable claim that they interacted with too many people and it hampered their language learning.
My take is that basically anything can be made to work if you are properly motivated. Tech is - at best - a secondary concern.
I think the best way to learn a language is offline through actual human interaction.
I've used Duolingo in the past (and other apps) and quickly lost interest, it's a fun app, but I feel like you don't learn from it. If I had to learn a new language today, I'm confident I could make good progress with GPT or Gemini, but tailoring it to how I learn.
> If an LLM can teach me a language, why wouldn't I go straight to the source and use GPT or Claude and customize it to my exact needs.
Because the "customize it" part is not trivial. That is the value add for most people who can not do that customisation themselves.
That said, I've found LLMs to be terrific if the goal is to learn the rules of language's grammar. But to actually learn to speak the language, find HelloTalk to be best, beaten by nothing other than actually sitting down with a native speaker.
Duolingo does not need a moat. They are already an established incumbent and have market inertia on their side. They can afford to experiment and make mistakes, and then to backtrack those mistakes.
For what it's worth, I also stopped using Duolingo when the human forums closed. I often found as much value in the forums as in the actual course. And like GP, the hardest part was relinquishing my nearly 1000 day streak.
I'd love to see someone build Duolingo with a single system prompt. By HN standards, no one would be paying for Character.ai, Cursor, Windsurf, or dozens of others tools because they could just call the ChatGPT API themselves.
I think the opposite is true: I'm deliberately choosing the usually more expensive DIY route even though I'm aware of cheaper commercial offerings.
If you hate ads, deception, and dark patterns as much as I do, then most software has a negative value. But I'll only pay for it if the value exceeds the price. That means there's a rather large (and probably growing) share of the software market that I'll just never pay even a penny for, because in my opinion, it has negative value. But it's not because I don't want to spend the money. It's because I want those companies that are peddling ad-laden bloatware to be a financial failure so that the market will intervene and offer better alternatives.
First of all it's an infinitely difficult problem to justly gauge value for value approximations, so a roughly free market will always have price fluctuations where people experiment.
And there's nothing inherently wrong with advertising. How else do you let your potential market know you have a product that they would benefit from? It's not always deception to get into their wallets, sometimes it's a genuine fair value trade.
Though I do think modern practices are full of immoral patterns, especially what George Lucas pioneered of brainwashing children into buying things. So yeah I agree that in practice, most modern marketing is just plain deception.
On the other hand, you then have millions of young, talented developers wasting their time and energy on open source projects that are hugely innovative and useful, but because they wholesale reject marketing, their projects never get anywhere, and they settle for unfulfilling jobs that society could do without.
And maybe the saddest part of all is that many such young talented devs spent so much time making useful projects for free, and never saw a dime because of it, despite the fact that corporations are now profiting from their work daily. You may say, well, the dev put it out there for free and didn't ask for money, but in the current market, what other choice is there? The race to the bottom has already been won and first place was $0.
Lol. Linus didn't pour decades of his life into Linux to save a buck. Neither did Stallman with GNU. Even if the alternatives at the time only cost $15, (hahahahahaha) free alternatives would still have been well worth making.
Software freedom has never been about money, and always been about preserving public access to a software commons. You might believe your local bookstore is a better source of knowledge than your local library, but that's only because you don't really know what you're missing.
I agree with your point, but your analogy doesn't hold up. There are several topics I've had to go to book stores to learn more about because my local library (and the systems they connect with) don't have any relevant materials. I make a habit of checking the library for non-fiction books first. Sometimes it pays off, but often not.
If you enjoy building things why not do it yourself while also saving money?
My dad likes doing oil changes himself which I'd never do, but it doesn't occur to me to insult and question why a person has different values than I do.
Well, and as I always told myself - it’s an investment in tools I’ll use again later, and eventually it’ll be cheaper to DIY myself.
Which, at the point I was welding together solar panel mounts onto shipping containers was true. But maybe i could have asked myself ‘should I’ instead of ‘can I’ somewhere earlier in the process hah.
If you're going to drag Linus and Stallman into this, at least read up on your history. A commercial Unix or Lisp Machine was quite a bit more expensive than $15/month. They both got shafted by companies that couldn't care less about anything beyond profit. It's part of the ugly side of capitalism, fuck people and fuck the world.
I'm at a fairly high level in learning spanish, which mostly came from actually speaking the language and just studying the dictionary etc... Duolingo was absolutely useless - and that was 6 years ago. I can only imagine how terrible it is nowadays.
Anyway, Im responding to you because i find ChatGPT to be a FANTASTIC tutor. Like I was absolutely blown away. It can do all the translation stuff, but also answer questions about different verb tenses, conjugations, syntax etc... I'm sure that an extremely good spanish teacher would be better, but I think ChatGPT is probably better than most. And it is free.
But as you yourself admit, you were at a high level already, so you know how to ask the LLM good questions. If you were a beginner with zero knowledge, you wouldn't have that advantage, so the LLM's ability to help you would be far less. How many ESL learners are asking ChatGPT when to use the past participle or present perfect tense?
Believe it or not, I have a high level of understanding of my native language as well - as surely most 2nd language learners do.
You don't even need to know any grammatical terms (I barely do for English, and suspect most people don't either for their native language), so long as you can speak a language. Like, I don't have a clue what past participle or present perfect tense mean in either language. Nor do I particularly need to know them.
Though, I'll admit that it would surely be easier to learn Spanish if I knew and understood the equivalent constructs in English. Case in point - when I was a kid and they taught us French in school, it was just an exercise in memorization rather than understanding, because they hadn't taught us the English versions. Apparently that's just what happened where I'm from for an extended period of time. Yet, I'm very significantly literate.
Still, you just ask chatgpt "how do I say _ in Spanish?" and then ask questions with variations for subject, object, tense, etc. You obviously would start easy "how do I say 'I eat bread'", but can progress to 'I ate/will eat/am going to eat/had eaten/would have eaten/could eat bread'
And you could eventually make it quite complicated eg "if you had told me xyz as you should have, I would have asked him about it. Now I am hoping that etc..." There's a huge mix of grammatical concepts in that (and I know very few of their names), and chatgpt would easily translate it and explain the concepts to whatever degree I desire.
If I was sufficiently motivated, I am quite sure that chatgpt could rapidly and effectively teach me all English and Spanish grammar.
Perhaps it would be a different story if learning something wildly different - eg Vietnamese. Yet, there's surely still similar constructs underlying it, since they all have to reflect the same world.
When I am learning something new in development, the LLM is immensely useful since it has unlimited patience, and I can zero it in on exactly the level of complexity or understanding that I need.
