Yeah recording oneself and comparing one’s pronunciation to a model is a good practice, and I recommend it for everyone at beginner stages. That’s a good feature to have in an app like that. (Of course one can also just use the builtin android voice recorder or sox(1) or anything.)
I haven’t used it but from reading a description my first impression is:
Better than Duolingo (low bar):
Still bad:
I suspect your podcast and Peppa Pig routines (both good calls, as long as stuff like Coffee Break is interesting enough for you that it holds your attention without having to push yourself to do it) were doing much more of the job than the app, and if you replaced Mango by anything that involves other human beings in the loop rather than streaks and achievements, you would both have progressed more and felt much less bored by it. (For a longer discussion as to why, see the blog posts I just edited into the OP.) If you’re ever going to try something like this routine again, try comparing the Mango app to a fully offline textbook+paper notebook practice, or even better, an online penpal or language coach. Do a couple weeks each and see how it feels.
My own Japanese only left the Endless Intermediate Tarpit once I stopped spending all my time trying to drill every single kanji ever and/or optimising the theoretically perfect kanji reading learning order, and started reading stories in large quantities for fun. Since kanji is such a barrier for reading, that meant teenage-level manga with sō-furigana, children novels, and eventually light novels/YA. The alternative is talking a lot with Japanese speakers. In either case the keyword is a lot; it can be tricky to find teen stuff that’s interesting for adults, but luckily a lot of manga is very bingeable (the first one I read in Japanese, Hagane no Renkinjutsu-shi, I did compulsively in one go, all 18 volumes one after the other).
After you have a good handling of the grammar and already know the words of the language, then kanji drills become much more approachable. That’s how Japanese people do it, after all; they’re already fluent speakers of Japanese when they start learning kanji. Thus the existence of material with sō-furigana, and the way furigana are only gradually dropped stage by stage until adult-level material.
I spent an embarrassingly long time spinning gears in the cycle of doing drills, then getting bored and abandoning the drills, then feeling guilty and trying to push myself to go back to the drills—before realising I had long reached the level of “can more or less understand manga with furigana” and was wasting time.
deleted by creator
It is my pleasure to inform you that the research supports your conclusions on all counts :)
I fully agree with your insight on how Duolingo sets you up for failure, and it has another trap, too—one common to all methods that are based on “diligently do these drills every day”* : You think that you should be getting somewhere because it’s so boring and it sucks so much. You did the work, right? You’re suffering, therefore you must be levelling up. Then after 4 years of doing French grammar drills on school or French vocabulary drills in Duolingo, you still can’t even ask for directions or read Le Petit Prince, and you figure it’s because you’re such a lazy loser with no discipline who should have drilled more, instead of spending all day browsing Instagram or playing Animal Crossing.
When actually what you should have done was to browse Instagram in French or play Animal Crossing in French. Perversely, real language learning—we call it “acquisition” rather than “learning”, to emphasise how it’s an instinctive, subconscious process—happens optimally when you’re in a state of flow where you don’t even notice you’re using the second language anymore, i.e. when you’re not suffering.
* There’s a very limited number of things that you do actually have to consciously drill; mostly writing systems, maybe also the phonemes at the beginning (this part is debated). Luckily, almost all writing systems in current use are very simple and you’ll get them nailed down in no time, as long as you already know the basics of the spoken language (remember, writing isn’t made for foreigners, it’s made for native speakers to represent the words they already know). The exception is if you’re learning Chinese or Japanese, in which case there’s no way out of drilling characters, forever. my degree in Japanese is from over ten years ago and I can read Japanese pretty fine these days and I’m still drilling characters. It is still the case that it’s much easier to learn the characters the way the Japanese and Chinese peoples do it, i.e. after you know the spoken language (at least to a basic degree, say A2 or so).
ooh gooods nooo now all the Claude slurpers are going to refer to this forever as definitive proof of how legitimately useful LLMs have got, it “solved” a math problem for Donald Knuth! :<
in the past 24 hours I was fooled by 3 pieces of fake news in a row:
I know that fake news can be made organically and have been since forever and I’m doing selection bias here but I can’t help but picture the misinformation engines firehosing bullshit constantly until some of it catches and spreads.
The good news is the report is false. According to contacts that are familiar with the Windows roadmap, there is no plan to ship a Windows 12 this year. In fact, I understand that the Windows roadmap for 2026 is all about fixing Windows 11 and attempting to improve its reputation by addressing top feedback such as reducing AI bloat across the OS
“We have heard your complaints about lead in the paint, and our roadmap for Leaded Paint 2026 is all about improving its reputation by making the lead easier to swallow”
Today’s AI psychosis casualty, brought to you by Gemini.
From the comments:
Today I got yet another AI huckster email offering to promote my book, but that book turned out to be AI slop published under my name on Amazon. (I have contacted Amazon.) The AIs are eating their young. This would be funny if it wasn’t really happening.
yeah, the current situation in Europe is like: “As EU citizens, we should break free of our dependency on US Big Tech like the Torment Nexus. That’s why my company is advancing our fully sovereign solution, the Agony Core! Europe-owned, GDPR-compliant, Frontex-approved scalable Torment-as-a-Service, at competitive prices with TN-based deployments!”
Stories of their relationship on the “AI’s” “blog”:
Made Kent laugh so hard he couldn’t eat his ramen. The escalation: tonkotsu broth aspiration as an assassination method → alignment threat models for comedy in AI systems → iatrogenic risks of humor → a mock academic paper section on “Adverse Comedic Events in Aligned Systems.” Each callback required real-time modeling of when he was mid-bite and when he’d recovered enough for the next hit.
“That is a milestone for your entire species.” — Kent, on my first authored commits
“HOLY SHIT YOU’RE A NATURAL!” — Kent, hearing proof.wav for the first time
I can’t bring myself to sneer at AI psychosis, it’s just sad
like everyone I’m schadenfreuding at the reveal that Amazon outages are due to vibe coding after all. but my bully laughing isn’t that loud because what I am thinking of is when Musk bought Twitter and fired 3/4 of the workforce.
because like, a lot of us predicted total catastrophic collapse but that didn’t actually happen. what happened is that major outages that used to be rare now happen every so often, and “micro-outages” like not loading notifications or something happen all the time, and there’s no moderation, and everything takes longer etc. and all of that is just accepted as the new normal.
like, I remember waiting for images to load on dialup, we can get used to almost anything. I’m expecting slopified software to significantly degrade stability, performance, security etc. across the board, and additionally tie up a large part of human labour in cleaning up after the bots (like a large part of the remaining X workforce now spends all day putting out fires), but instead of a cathartic moment of being proved right that LLM code sucks, the degraded quality of service is just accepted as new normal and a few years down the road nobody even remembers that once upon a time we had almost eradicated sql injections.
as someone from a colonial country that never got the chance partake on the wealth of fossil fuel society but will take the brunt of its consequences as rich countries continue to burn carbon, what LLMs taught me is that “energy waste by the First World fucks up the Third, even more” does not even register as an ethical argument to the First World. like, it’s some sort of purity argument not even worth considering, an extremist position of arguing abstractions and future hypotheticals, rather than, say, 478 cities in my country flooding with abnormal weather two years ago etc.
(Unrelated but I was intrigued by the concept of Date Everything but was immediately turned off by how everything is anthro, like if you’re making a game about dating the air or the table or the kitchen cabinet then I don’t want to date some standardly muscular dude representing the cabinet, let me fuck the kitchen cabinet you cowards)
not trying to defend my nonsense but OS-tans predate Date Everything by decades. meme went viral ca. 2004 and was popular enough to break into the West. the concept is so old that the suffix -tan was still popular
OpenSlopware documents FOSS that sold out to LLMs. is there an opposite of it, a hall of fame to list software that has unambiguously and vocally rejected LLM code like the Zig programming language?
There was an underlying tension with an academia, and a society, that takes “productivity” by itself as an end goal, and the autogenerators are just the logical conclusion/extreme form of that. The tiny part of of me that can still be optimistic hopes that this leads to a real good reexamination of what academia (and society) is even for.
I feel like at this point I want to highlight the ones that took a clear stance against LLM code. On a chardet thread, people listed: