News for hackers. Stuff that matters.
Re:Don't post generated/AI-edited comments. HN is for conver... by bruckie on Wednesday March 11, 2026 @04:35PM (#47341211)
My elementary school kid came home yesterday and showed me a piece of writing that he was really proud of. It seemed more sophisticated than his typical writing (like, for example, it used the word "sophisticated"). He can be precocious and reads a ton, though, so it was still plausible that he wrote it. I asked him some questions about the writing process to try to tease out what happened, and he said (seemingly credibly) that he hadn't copied it from anywhere or referenced anything. He also said he didn't use any AI tools. After further discussion, I found out that Google Docs Smart Compose (suggested-next-few-words feature) is enabled by default on his school-issued Chromebook, and he had been using it. The structure of the writing was all his, but he said he sometimes used the Smart Compose suggestions (and sometimes didn't). He liked a lot of the suggestions and pressed tab to accept them, which probably bumped up the word choice by several grade levels in some places.

So yeah, it can change the character of your writing, even if it's just relatively subtle nudges here or there.

edit: we suggested that he disable that feature to help him learn to write independently, and he happily agreed.

Re:Don't post generated/AI-edited comments. HN is for conver... by Terr_ on Wednesday March 11, 2026 @04:49PM (#47341494)
To rationalize my gut-feelings on this, I think it comes down to the spectrum between:

1. A system that suggests words, the child learns the word, determines whether it matches their intent, and proceeds if they like the result.

2. A system that suggests words, and the child almost-blindly accepts them to get the task over with ASAP.

The end-results may look the same for any single short document, but in the long run... Well, I fear #2 is going to be way more common.

Re:Don't post generated/AI-edited comments. HN is for conver... by zahlman on Wednesday March 11, 2026 @05:25PM (#47342176)
The analogy with tab-completion of code seems apt. At first you blindly accept something because it has at least as good a chance of working as what you would have typed. Then you start to pay attention, and critically evaluate suggestions. Then you quickly if not blindly accept most suggestions, because they're clearly what you would have written anyway (or close enough to not care).

The phenomenon was observed in religious philosophy over a millennium ago (https://terebess.hu/zen/qingyuan.html).

Re:Don't post generated/AI-edited comments. HN is for conver... by abustamam on Wednesday March 11, 2026 @05:35PM (#47342321)
Tab completion was so novel back when full e2e AI tooling was not really effective.

Now that it is, I just turn tab completion off totally when I write code by hand. It's almost never right.

Re:Don't post generated/AI-edited comments. HN is for conver... by skydhash on Wednesday March 11, 2026 @06:54PM (#47343505)
Emacs has completion (but you can bind it to tab). The nice thing is that you can change the algorithm to select what options come up. I’ve not set it to auto, but by the time I press the shortcut, it’s either only one option or a small sets.
Re:Don't post generated/AI-edited comments. HN is for conver... by bruckie on Wednesday March 11, 2026 @05:26PM (#47342187)
From his description, it sounded like this was more of #1. He cared a lot about the topic he was writing about, and has high standards for himself, so it's very likely that he would have considered and rejected poor suggestions.

I have mixed feeling about it. On the one hand, you're right: carefully considering suggestions can be a learning opportunity. On the other hand, approval is easier than generation, and I suspect that without flexing the "come up with it from scratch" muscle frequently, that his mind won't develop as much.

Re:Don't post generated/AI-edited comments. HN is for conver... by yellowapple on Wednesday March 11, 2026 @08:48PM (#47344774)
#1 would be a net improvement over the status quo IMO. Seems like a great way for people to expand their vocabularies organically.
Re:Don't post generated/AI-edited comments. HN is for conver... by lossyalgo on Wednesday March 11, 2026 @09:04PM (#47344925)
That reminds me of one of the biggest IMO missing feature of Wordle: They never give a definition of the word after the game is finished! I usually do end up googling words I don't know (which is quite often) but I'm guessing I'm one of the few who goes to the trouble. I've even written to The New York Times a couple times to suggest adding a short definition at the end as I honestly feel like a ton of people could totally up their vocabulary game and it surely could be added with minimum effort (considering they even added a Discord multiplayer mode).
Re:Don't post generated/AI-edited comments. HN is for conver... by yellowapple on Thursday March 12, 2026 @12:32AM (#47346473)
That's a brilliant idea and now that you've mentioned it it seems like a rather glaring omission.
Re:Don't post generated/AI-edited comments. HN is for conver... by lossyalgo on Thursday March 12, 2026 @05:56AM (#47348505)
Please write to the NY Times and suggest it! I still play and it still irks me when I have to go google a word.
Re:Don't post generated/AI-edited comments. HN is for conver... by Terr_ on Thursday March 12, 2026 @01:36AM (#47346900)
Is Wordle really the best vehicle for that, though? I mean, it tends towards a subset of 5-letters words the audience is more likely to know in advance, excluding a lot of the more-surprising words.

A "click to see more about why this answer fits" crossword, on the other hand...

Re:Don't post generated/AI-edited comments. HN is for conver... by lossyalgo on Thursday March 12, 2026 @05:55AM (#47348498)
How often have you played Wordle? I've played well over 1000 games, and at lesat 1/5th of those were words I had to look up. They seem to enjoy picking obscure words in order to make the game more challenging.
Re:Don't post generated/AI-edited comments. HN is for conver... by Terr_ on Thursday March 12, 2026 @12:59PM (#47353855)
Perhaps the unusual outcomes are just more memorable, and so seem more frequent? Here's a representative sample of 30 that were used very recently.

    Shoal, Hasty, Lobby, Vogue, Gunky, 
    Sheep, Theft, Linen, Slime, Fluke, 
    Hydra, Dizzy, Lance, Shred, Buyer, 
    Attic, Guava, Awake, Stank, Hoist,
    Mogul, Squad, Roost, Skull, Bloom,
    Mooch, Surge, Vegan, Scene, Cello,
None of those stand out as "WTF does that even mean", but maybe I'm the weird one if we adjust for age-demographics or book-reading.

If I had to guess at a riskier 20%... Guava, a fruit some people may not have had; Gunky because it's slang; Mogul, Vogue, and Mooch were borrowed from other languages; Cello is something people may have heard more than read; Hoist.

Re:Don't post generated/AI-edited comments. HN is for conver... by lossyalgo on Thursday March 12, 2026 @01:17PM (#47354154)
> Perhaps the unusual outcomes are just more memorable, and so seem more frequent?

That's a good point and could very well be true. I just know I've played plenty of games where I was mad that they didn't show the meaning. So let's say its 5% for native speakers, and up to 20% for non-native speakers - that's still a golden opportunity to expand vocabularies. And honestly it can't be a lot of work to add a couple lines of static text. At worst it would be ignored, and at most, help people learn more interesting words.

Re:Don't post generated/AI-edited comments. HN is for conver... by yellowapple on Monday March 16, 2026 @05:32PM (#47405213)
Even as someone who already knows the definitions of 100% of those words, it'd still be neat to see those definitions, and maybe even some etymologies or other interesting trivia. “Cello” would be a great opportunity for a factoid about the oldest known cello, “guava” would be a great opportunity for a quick recipe using guavas, etc.
Re:Don't post generated/AI-edited comments. HN is for conver... by comboy on Wednesday March 11, 2026 @04:42PM (#47341356)
Oh how I despise these suggestions. You sometimes look for a way to express something and you are on the verge of giving the world something truly original, but as soon as your brain sees the suggestion it goes "oh yeah that fits"
Re:Don't post generated/AI-edited comments. HN is for conver... by Terr_ on Wednesday March 11, 2026 @04:53PM (#47341578)
True! There's an important cybernetic aspect to all this, where an automatic suggestion can be an interruption, sometimes worse if the suggestion is decent.

A certain amount of friction is necessary, at least if the goal is to help the person learn or make something original.

Re:Don't post generated/AI-edited comments. HN is for conver... by SchemaLoad on Wednesday March 11, 2026 @06:12PM (#47342872)
I disabled them immediately, it feels like the tech version of the ADHD person who keeps interrupting you with what they think you are trying to say. Even if the suggestion is correct, it saves you at most 2 seconds at the cost of interrupting you constantly.
Re:Don't post generated/AI-edited comments. HN is for conver... by TimTheTinker on Wednesday March 11, 2026 @04:46PM (#47341428)
GK Chesterton would have something brilliant to say about the inauthenticity of it all or something.
Re:Don't post generated/AI-edited comments. HN is for conver... by jrockway on Wednesday March 11, 2026 @04:46PM (#47341429)
I see the suggestions and then choose something different anyway. I don't want to use one of the top 3 most popular responses to an email from a friend. Even if it's something transactional.
Re:Don't post generated/AI-edited comments. HN is for conver... by JumpCrisscross on Wednesday March 11, 2026 @04:44PM (#47341400)
> I despise these suggestions

As an adult, I do too. As a middle schooler, we absolutely used word processors’ thesaurus features to add big words to our essays because the teachers liked them.

Re:Don't post generated/AI-edited comments. HN is for conver... by Gibbon1 on Wednesday March 11, 2026 @05:02PM (#47341775)
Friend of mine was a English teacher. She quit because she's not going to waste her time 'grading' 30 essays written by AI.

Anyway before that she HATED the thesaurus. And she could tell when students were using it to make their writing more fancy pants.

Re:Don't post generated/AI-edited comments. HN is for conver... by zahlman on Wednesday March 11, 2026 @05:30PM (#47342244)
One problem I see is that LLMs have a more nuanced... well, model of how words and their meanings relate to each other than a dead-tree thesaurus could ever present, what with its simplified "synonym" and "antonym" categories. Online versions try to give some similarity metrics, but don't get into the nuance. (It's not as if someone who takes either approach would want to spend the time reading and understanding that, anyway.)
Re:Don't post generated/AI-edited comments. HN is for conver... by JumpCrisscross on Wednesday March 11, 2026 @05:10PM (#47341919)
> she could tell when students were using it to make their writing more fancy pants

I had two teachers who called us out on this, and actually coached us on our writing, and I remember them fondly. (They were also fans of in-class essaying.)

The others wanted to count big words.

Re:Don't post generated/AI-edited comments. HN is for conver... by tigen on Wednesday March 11, 2026 @11:15PM (#47345974)
In-class essays impossible? Pencil to paper?
Re:Don't post generated/AI-edited comments. HN is for conver... by lossyalgo on Wednesday March 11, 2026 @09:09PM (#47344964)
I look forward to reading studies in 10 years how we all became stupider thanks to this "feature". One step closer to the movie Idiocracy.