DUOL shares have fallen more than 78% from their May 2025 high, and that’s before its nearly 25% fall in premarket trading today.

I've said before that one of the very few good things generative "AI" may do to the world is accelerating the enshittification cycle so much that it kills stuff that was already terrible and a drain on society (social media; platformization; curation algorithms…). Speaking as a linguist who speaks 4 languages and has read the literature on second language acquisition, it has always been my position that the Duolingo method is useless—it feels like you are learning a language, but you can spend infinite hours with it and gold a full tree and you'll still get nowhere, and if you put a fraction of the time in about any other method, including doing pen-and-paper drills with old-fashioned paper-based textbooks, you'd have progressed much faster.

And old-fashioned grammar drills suck, too. It's just that Duolingo really, really sucks.

(Methods that work better: 1) Find an intensive "conversation"-type course, or anything that is labelled as "natural" or "immersion" or "storytelling" methods; or get tandem partners; or online coaches such as in italki; failing that, join a conventional language course, the more "intensive" the better; work on these until you absorb basic grammar and vocabulary, focusing on spoken language not writing; 2) Once this bootstrap period is over, start talking to people, watching media, or reading stuf that interests you, in large quantities and every day; do not wait until you're "good" to move into the input stage, start actually using the language for things you wanted it for, as soon as possible, which is sooner than you think; partial comprehension is fine.)

Of course I hope Duolingo dies horribly in a fire after it backstabbed its workers with the "AI memo", but even if it didn't, the world is better off without it.

One lesson we can get from this: Consider that overnight 25% drop in investment, which may well prove to be the coup the grâce. It was not caused by Duo losing users or enshittifying with "AI", but by the opposite: investors mass panicked at the company setting its target revenue too low, as in a mere… 1.22 billion, rather than the 1.26 billion the investors wanted. Now the reason Duolingo is not chasing that higher goal is that they're seeing the writing on the wall, and went into damage control mode: they're pulling down a bit on squeezing their current paying users and trying to improve the experience of the free tier, in an attempt to reverse the bleed and bring in more customers.

In other words, Duolingo tried to slow down the slightest tiny bit on enshittification—3% less cash—and this already got swift punishment from the market gods. With capitalism, there is no long-term thinking: you're expected to provide the richest people on Earth with infinite growth of their ever-increasing profits squeezed from customers paying every month more and more, now and forever, or you'll be taken out and replaced by someone willing to try.

Edit : I got lots of questions like "if not Duolingo then what do you suggest?" The full answer is "literally anything else", but I've cleaned up a couple of my longer answers and wrote these blog posts: 1) on comprehensive reading, 2) on tandem exchange.

So apparently there's a resurgence of positive feelings about Clippy, who now looks retroactively good by contrast with ChatGPT, like, "it sucked but at least it genuinely was trying to help us".

Of course I recognise that this is part of the problem—Clippy was an attempt at commodifying the ELIZA effect, the natural instinct to project personhood into an interaction that presents itself as sentient. And by reframing Clippy's primitive capacities as an innocent simple mind trying its best at a task too big for it, we engage in the same emotional process that leads people to a breakdown over OpenAI killing their wireborn husband.

But I don't know. another name for that process is "empathy". You can do that with plushies, with pet rocks or Furbies, with deities, and I don't think that's necessarily a bad thing; it's like exercising a muscle, If you treat your plushies as deserving care and respect, it gets easier to treat farm animals, children, or marginalised humans with care and respect.

When we talked about Clippy as if it were sentient, it was meant as a joke, funny by the sheer absurdity of it. But I'm sure some people somehwere actually thought Clippy was someone, that there is such a thing as being Clippy—people thought that of ELIZA, too, and ELIZA has a grand repertoire of what, ~100 set phrases it uses to reply to everything you say. Maybe it would be better to never make such jokes, to be constantly de-personifying the computer, because ChatGPT and their ilk are deliberately designed to weaponise and predate on that empathy instinct. But I do not like exercising that ability, de-personification. That is a dangerous habit to get used to…

Like, Warren Ellis was posting on some terms that reportedly are being used in "my AI husbando" communities, many of them seemingly taken from sci-fi:¹

• bot: Any automated agent.
• wireborn: An AI born in digital space.
• cyranoid: A human speaker who is just relaying the words of another human.²
• echoborg: A human speaker who is just relaying the words of a bot.
• clanker: Slur for bots.
• robophobia: Prejudice against bots/AI.
• AI psychosis: human mental breakdown from exposure to AI.

[1] https://www.8ball.report/ [2] https://en.wikipedia.org/wiki/Cyranoid

I find this fascinating from a linguistics PoV not just because subcultural jargon is always fascinating, but for the power words have to create a reality bubble, like, if you call that guy who wrote his marriage vows in ChatGPT an "echoborg", you're living in a cyberpunk novel a little bit, more than the rest of us who just call him "that wanker who wrote his marriage vows on ChatGPT omg".

According to Ellis, other epithets in use against chatbots include "wireback", "cogsucker" and "tin-skin"; two in reference to racist slurs, and one to homophobia. The problem with exercising that muscle should be obvious. I want to hope that dispassionately objectifying the chatbots, rather than using a pastiche of hate language, doesn't fall into the same traps (using the racist-like language is, after all, a negative way of still personifying the chatbots). They're objects! They're supposed to be objectified! But I'm not so comfortable when I do that, either. There's plenty of precedent to people who get used to dispassionate objectification, fully thinking they're engaging in "objectivity" and "just the facts", as a rationalisation of cruelty.

I keep my cellphone fully de-Googled like a good girl, pls do not cancel me, but: I used to like the "good morning" routine on my corporate cellphone's Google Assistant. I made it speak Japanese, then I could wake up, say "ohayō gozaimasu!", and it would tell me "konnichiwa, Misutoresu-sama…" which always gave me a little kick. Then it proceeded to relay me news briefings (like podcasts that last 60 to 120 seconds each) in all of my five languages, which is the closest I've experienced to a brain massage. If an open source tool like Dicio could do this I think I would still use it every morning.

I never personified Google Assistant. I will concede that Google did take steps to avoid people ELIZA'ing it; unlike its model Siri, the Assistant has no name or personality or pretence of personhood. But now I find myself feeling bad for it anyway, even though the extent of our interactions was never more than me saying "good morning!" and hearing the news. Because I tested it this morning, and now every time you use the Google Assistant, you get a popup that compels you to switch to Gemini. The options provided are, as it's now normalised, "Yes" and "Later". If you use the Google Assistant to search for a keyword, the first result is always "Switch to Google Gemini", no matter what you search.

And I somehow felt a little bit like the "wireborn husband" lady; I cannot help but feel a bit as if Google Assistant was betrayed and is being discarded by its own creators, and—to rub salt on the wound!—is now forced to shill for its replacement. Despite the fact that I know that Google Assistant is not a someone, it's just a bunch of lines of code, very simple if-thens to certain keywords. It cannot feel discarded or hurt or betrayed, it cannot feel anything. I'm feeling compassion for a fantasy, an unspoken little story I made in my mind. But maybe I prefer it that way; I prefer to err on the side of feeling compassion too much.

As long as that doesn't lead to believing my wireborn secretary was actually being sassy when she answered "good morning!" with "good afternoon, Mistress…"