Yet another Hacker News

Show HN: Qwen3-4B with the AI writing tics trained out

66 points erikkannike 1 hour ago 38 comments

anilgulecha 1 hour ago | parent

What's with the very distracting cursors of others?

mullenba 1 hour ago | parent

Exactly. I was interested in the article but I don't have time for that nonsense.

markgavalda 1 hour ago | parent

Agreed. In the upper right corner there's a "hide" button that I immediately clicked. Then read (ok, skim read) the article.

HumblyTossed 57 minutes ago | parent

Agreed. Just because you _can_ do a thing, doesn't mean you should.

awestroke 1 hour ago | parent

Nobody knows what it means, but it's provocative. It gets the people going

Almondsetat 1 hour ago | parent

going... away?

maven29 1 hour ago | parent

The ancients brought us "reading mode" in browsers as an escape hatch for exactly this kind of cruelty

jmusall 1 hour ago | parent

There's a "hide" button in the top right corner

2bird3 56 minutes ago | parent

All it is missing now is a chat and a friend/follow feature.....

RonanSoleste 56 minutes ago | parent

I don't understand why this is a feature and what it brings other than shout "you are being watched by random others". Exited the site before ending the first block.

ac2u 1 hour ago | parent

Is it still capable of writing an essay about AI writing tics or has the very means of producing said tics, even when the model is asked to do so to demonstrate, now abitterated?

swiftcoder 56 minutes ago | parent

Impressed how little hardware is needed to train this.

hashmap 51 minutes ago | parent

I saw those moving dots and snap-closed the page.

chmod775 48 minutes ago | parent

Models have proclivities, and many people use the same model.

Imagine all books on earth were written by just 5 people - and worse, they could not remember what they had written yesterday. You would quickly get tired of reading the same flourishes and patterns over and over.

Whatever style models currently in use have, that's what people are going to be tired of, and it's what they're going to recognize as "AI".

sejje 39 minutes ago | parent

I predict by the end of the year they'll have a style/voice selector for responses.

You can already do that in your agents.md file, of course, or ask for a re-write and dodge the sameness.

wongarsu 28 minutes ago | parent

The most difficult thing in doing it yourself is knowing what to ask for. I can ask my model to talk like a caveman or like a Baltic pirate. But asking for something more nuanced, in the range of regular speech, requires the correct lingo

I'm sure it's trivial for somebody who has studied communication or contemporary linguistics, just as it's trivial for me to ask an LLM for tenant-isolation using row-level security in an app

CharlesW 10 minutes ago | parent

> I'm sure it's trivial for somebody who has studied communication or contemporary linguistics…

But you know what you like, and you know what bothers you, and the good news is that you don't have to one-shot a solution. You can create a skill and refine it over time. As you test and/or use it, identify copy you don't like and tell the LLM what you prefer instead, and ask it to distill and generalize that feedback into the skill.

ndriscoll 10 minutes ago | parent

You could try taking a body of your own writing and asking it to characterize it/come up with a prompt (or fine tune if you can) describing your style and quirks.

Royce-CMR 38 minutes ago | parent

Applause earned. Neat results especially out of such a small model.

Separately we have overlap between AI slop styles and right for the situation writing styles which this highlights.

Using the new CRM example, the clean version looks like a millennial student who rambled on a page, and the AI slop version looks like a draft I could send for PR to corporate managers. I buy the former looks more real, but it also looks sloppy and doesn’t work for most of my use cases.

When I was in school, the structured essay style of the “slop” examples was encouraged and I see it continue to resonate in corporate. The cleaned up versions here def have use cases - reviews for example.

Main point - different styles for different situations and our AI slop versions vs humanized versions both overlap different areas of the “ideal” we continue to look for.

trilogic 37 minutes ago | parent

The q4 loops after 16k tokens generated. Great idea though, hope to see more models like it. Thanks. https://huggingface.co/erikkannike/qwen3-4b-deslop-v4/discus...

charv 36 minutes ago | parent

Side note but the multi-user cursors was a super cool feature, not sure it was relevant to this article but would love to see it on more interactive pieces (a la The Pudding, etc.)

JustSkyfall 32 minutes ago | parent

It's definitely a slight improvement, but it is still rather obvious that the output is AI-written :(

mrweasel 29 minutes ago | parent

The tell tell sign for me is that it's still incredibly hard to read. AI tend to write in a style that I'm unable to easily read (humans does that as well sometimes). I constantly have to go back and re-read the previous sentence, because the text simply doesn't register.

A next step I'd suggest is asking the LLM to use shorter words when available.

BoredomIsFun 26 minutes ago | parent

Did you check this? https://huggingface.co/sam-paech/gemma-3-27b-it-antislop

BoredomIsFun 31 minutes ago | parent

This is not the first deslop. The true pioneer is this dude: https://huggingface.co/sam-paech/gemma-3-27b-it-antislop. This model is a perfectly usable, full size not-toyish model.

f311a 22 minutes ago | parent

Why would you support this sort of projects? LLM text should be spottable for as long as possible, otherwise human communication and expression on the internet will be killed.

squidbeak 10 minutes ago | parent

In what sense will it be killed? I don't plan to stop writing my own messages or notes, just because AI can write, and I seriously doubt many other people who prefer their own voices will change either.

f311a 6 minutes ago | parent

Well, imagine that 90% of the comments on HN are generated and it's very hard to spot it. Would you read them?

10111001 21 minutes ago | parent

Now can I have a model with the AI writing tics trained to the max?

I want my boss to know I'm a Good Boy using AI.

Izmaki 16 minutes ago | parent

You can definitely get far with skills and personal instructions. That is my estimate, not research.

(*defeated chuckles and gazes towards my Claude session*)

BSOhealth 19 minutes ago | parent

for https://slopwash.com I’ve started approaching this more as an eval problem than something to go post-training for. I’m excited to be trying these new Jev-type models in more of a harnessed approach. maybe some heavy LLM to measure preservation of intent

bschwarz 14 minutes ago | parent

    and needs no system prompt

A human would write "and doesn't need a system prompt."

jwr 13 minutes ago | parent

Why kill the em dashes? They are a legitimate mark — one that I have been using for years. On a Mac, you quickly get used to pressing shift-option-hyphen. Em dashes are typographically correct and should be used.

literalAardvark 11 minutes ago | parent

Because no one knows what those are, and they're a major indicator of AI writing.

ahmedfromtunis 9 minutes ago | parent

Or at least match the pre-llm human usage frequency.

A total, artificial lack of em dashes can in itself be perceived as another sign of AI writing.