HN

Comments (57)

• Menu

  LeoPanthera
  This is very similar to the Google Trends results for "frqnce":https://trends.google.com/trends/explore?date=all&q=frqnce&h...You'll notice it peaks every northern hemisphere summer. On French keyboards, Q and A are reversed compared to US keyboards, and every summer, millions of French people go on vacation, and start Google searching for things back home on unfamiliar keyboards.It declines with the rise of the smartphone, as they're bringing their keyboards with them.Why it suddenly spikes in the last few years, I don't know.
• Menu

  anyfoo
  This is actually pretty common. It's less obvious with Chinese or Japanese, as the input method there usually matches the transliteration based on how the word is spoken (romaji in Japanese, pinyin in Chinese), which of course does not look unusual.For example, you wouldn't think twice about it if for the Japanese word for washing machine, you not only saw "洗濯機" (which is how it's written in Kanji), but also "sentakuki" or "sentakki" in the search results, because even to non-Japanese speakers it's pretty clear that that's probably the Japanese word for washing machine written with latin character transliteration, and pretty much exactly what you'd say.With Korean, it looks more jarring, as the input method is apparently very different, and seems to map the keys for unrelated latin letters to Hangul letters? (I have no idea, I don't know anything about Hangul other than it's based on syllables, kind of like Hiragana/Katakana, and apparently very logical.)
• janalsncm
  > It's actually insane the levels of understanding the algorithms that are responsible for serving us information have and how little we, the creators of said algorithms, understand what's going on in said algorithms.As others have said, keyboard mismatches are common enough that Google might have built out logic for it specifically. But thats not necessary and even “old school” search engines could learn these things.The first time “alemwjsl” is searched you might not have any data, but the user will probably fix their keyboard and retype in Korean. That gives you a query correction mapping. And you can assume if query1 yields no clicks and they update to query2, q1 is a synonym for q2 and serve results for q2 instead.Then, if a session contains a query “alemwjsl” and a click on midjourney.com and another session “midj” also contains a click on midjourney.com, those are co-clicked queries.You can also even start to represent queries by the words in their associated clicked documents or vice versa. This helps to get around the fact that people might search “how much superbowl tickets” and “superbowl tickets price” but the official page might not contain either of those strings.Of course there’s more advanced methods now (neural nets) but it’s cool to see how it worked in the past.https://www.kdd.org/kdd2016/papers/files/adf0361-yinA.pdf
• Menu

  yongjik
  > Turns out that somehow Midjourney is so commonly searched for, that Google has started serving them in search results for a meaningless English phrase that just means a Korean forgot to switch off of their English keyboard when searching for.BTW, this happens all the time in Korea, because it's extremely common for someone to type something while forgetting to switch to the correct input method. Try these, for example: 추ㅜ gozjsbtm elwmsl vkdlTjs
• Menu

  yorwba
  > It's actually insane the levels of understanding the algorithms that are responsible for serving us information have and how little we, the creators of said algorithms, understand what's going on in said algorithms.Keyboard layout mismatches are common enough that I assume Google has a layout detection stage hardcoded just like they have typo correction hardcoded. And the creators of said algorithms probably understand very well how they work. (The naïve way would be to convert from every possible layout to every other layout, but I think you could build something more lightweight using Hidden Markov Models.)
• Menu

  FeteCommuniste
  The Greek string υοθτθβε (meaningless and nearly unpronounceable, would sound roughly like "eeohtht-thveh") will get you YouTube as the top search result because those letters are what you get from typing "YouTube" with your keyboard set to Greek mode, at least on Windows.
• Menu

  bee_rider
  Adding to the confusion, alemwjsl just almost looks like a plausible name for something. Looking at it, you start parsing, right? “Clearly this is Alex… uh, something… JavaScript Library…”
• sergiotapia
  Google also knows what you're searching for if you touch type the wrong thing like one key shifted to the right.
• Menu

  rudolftheone
  I though I will test this case with Gemini 3 pro and bam: it nailed it perfectly on the first attempt:The string "alemwjsl" is a classic example of a keyboard input error specific to Korean users. Here is the explanation: The Hypothesis: The "Han/Yeong" (Korean/English) Toggle Error In South Korea, keyboards are bilingual. Users frequently switch between the Korean script (Hangul) and English (QWERTY) using a toggle key. If a user intends to type the Korean word for Midjourney (미드저니) but forgets to toggle the keyboard input from English to Korean, the output corresponds to the physical location of the keys on a standard QWERTY layout. The Proof (Mapping the Keys) Let’s break down the Korean word 미드저니 (Midjourney) key by key on a standard "2-Set" Korean keyboard: 미 (Mi) ㅁ corresponds to the A key. ㅣ corresponds to the L key. Result: al 드 (Deu) ㄷ corresponds to the E key. ㅡ corresponds to the M key. Result: em 저 (Jeo) ㅈ corresponds to the W key. ㅓ corresponds to the J key. Result: wj 니 (Ni) ㄴ corresponds to the S key. ㅣ corresponds to the L key. Result: sl Put it all together: al + em + wj + sl = alemwjsl Why this happens and why they bid on it Muscle Memory: Midjourney is a very popular search term in Korea (113K volume for the main keyword). Thousands of users type it quickly without looking at the screen. By the time they realize they are typing in English mode, they have already hit enter or the search bar has auto-suggested the "gibberish" term. Smart SEO/SEM Strategy: High Intent: Anyone typing "alemwjsl" is 100% looking for "Midjourney." There is no ambiguity. Lower Cost: While "미드저니" might have high competition (CPC 0.31), "alemwjsl" often has lower competition because many advertisers overlook "gibberish" keywords, though in this specific case, the CPC is quite similar (0.28 vs 0.31), indicating the secret is out. Capture All Traffic: By bidding on this, Midjourney ensures that even clumsy typists find their website immediately rather than being redirected by Google to a "Did you mean...?" page or a competitor. Conclusion: "alemwjsl" is simply 미드저니 typed with the keyboard set to English. It represents high-intent users making a very common technical mistake.Not the first time ChatGPT being inferior in such tasks.
• xp84
  > branded keywords aren't that great to run ads on anyway, you pretty much get all the traffic from them anyway since that's what the user wanted anyway, you don't really have to pay for it.Haven’t finished the article yet but this jumped out at me. This doesn’t ring true to me. Google runs an extortion scheme - since you can buy ads on your competitors’ trademarks, and since no users can tell ads from results (and since the organic results are now buried so far, they rarely get clicks anyway) if you don’t buy your brand keywords your competitors will get all your traffic.
• KitsumiTheFox
  I saw a video about this exact same topic! https://www.youtube.com/watch?v=N5Fj3YVok6U
• Menu

  GenerocUsername
  Going into article I guessed it s the Dvorak -> Qwerty mismatch.Korean -> English makes more sense.
• Menu

  magic_hamster
  Utter disappointment. The post was building up to this great reveal only to end up with the most mundane explanation. Anyone who uses a bilingual (or more) keyboard has had something similar happen to them a few times.
• mjk3026
  Reading this, it’s pretty clear the author doesn’t have any Korean friends. Koreans spend a non-trivial amount of time pressing the 한/영 (Korean/English) key.
• Menu

  echelon
  aadillpickle, fantastic blog!I've got nothing to add there that people haven't already been saying - this was a fascinating quirk of humanity and technology. Really good full-circle adventure uncovering the source.I'm commenting because I have to know what you're doing with your website and blog. It looks like a markdown/obsidian/static site generator. It's gorgeous and amazing. Did you write it yourself? Is it open source software?
• Menu

  msephton
  Very cool! Please add an RSS feed to your blog.
• N_Lens
  Somehow something common has wrapped around to being 'insane'.
• lifthrasiir
  Fun fact: intentional input method mismatch is commonly used for censoring profanities among internet streamers. For example, 시발 sibal (approximately corresponds to fuck in its ubiquity) transliterates to "tlqkf", so many Koreans can understand that without a written Korean text. Not that Koreans can generally read transliterated Hangul texts though.
• jihadjihad
  But no one ever figured out what the deal was with “covfefe”?
• phyzome
  I think I could have done without the pages of LLM output.
• nadermx
  This is a great example in tenacity. Pleasent to read too.
• yieldcrv
  ah, could have been worse. like something made up in a synthetic data set being training data for the world we experience
• Menu

  floren
  > I scroll up a bit to reread ChatGPT's analysis, and I realize it mentions "transliteration". I have no idea what that word means, so I look it up.How?
• shinhyeok
  As a Korean, this is hilarious
• anon
  undefined