frontpage.
newsnewestaskshowjobs

Open Source @Github

fp.

Open in hackernews

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

https://100toolkit.com/blog/deslop-model/
66•erikkannike•1h ago

Comments

anilgulecha•1h ago
What's with the very distracting cursors of others?
mullenba•1h ago
Exactly. I was interested in the article but I don't have time for that nonsense.
markgavalda•1h ago
Agreed. In the upper right corner there's a "hide" button that I immediately clicked. Then read (ok, skim read) the article.
HumblyTossed•57m ago
Agreed. Just because you _can_ do a thing, doesn't mean you should.
awestroke•1h ago
Nobody knows what it means, but it's provocative. It gets the people going
Almondsetat•1h ago
going... away?
maven29•1h ago
The ancients brought us "reading mode" in browsers as an escape hatch for exactly this kind of cruelty
jmusall•1h ago
There's a "hide" button in the top right corner
2bird3•56m ago
All it is missing now is a chat and a friend/follow feature.....
RonanSoleste•56m ago
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•1h ago
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•56m ago
Impressed how little hardware is needed to train this.
hashmap•51m ago
I saw those moving dots and snap-closed the page.
chmod775•48m ago
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•39m ago
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•28m ago
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•10m ago
> 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
Royce-CMR•38m ago
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•36m ago
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•36m ago
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•32m ago
It's definitely a slight improvement, but it is still rather obvious that the output is AI-written :(
mrweasel•29m ago
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•26m ago
Did you check this? https://huggingface.co/sam-paech/gemma-3-27b-it-antislop
BoredomIsFun•31m ago
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•22m ago
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•10m ago
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•6m ago
Well, imagine that 90% of the comments on HN are generated and it's very hard to spot it. Would you read them?
10111001•21m ago
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•16m ago
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•19m ago
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•14m ago

    and needs no system prompt
A human would write "and doesn't need a system prompt."
jwr•13m ago
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•11m ago
Because no one knows what those are, and they're a major indicator of AI writing.
ahmedfromtunis•9m ago
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.

•
9m ago
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.

Documents: Police officer gave her Flock login to her ICE agent husband

https://www.wjcl.com/article/savannah-police-officer-ice-agent-flock-access/74135816
496•DeepLogin•3h ago•236 comments

Linux 7.3 Released

https://lwn.net/Articles/1101143/
414•cjd8•7h ago•182 comments

Timestral: Mistral release time-series foundation models

https://mistral.ai/news/timestral/
218•groar•4h ago•43 comments

Wine Ported to OpenBSD

https://marc.info/?l=openbsd-ports&m=179236125692175
20•iamnothere•58m ago•0 comments

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

https://100toolkit.com/blog/deslop-model/
66•erikkannike•1h ago•38 comments

Alternatives to Animal Testing Are Finally Taking Off

https://spectrum.ieee.org/alternatives-to-animal-testing
98•rbanffy•3h ago•15 comments

Let's Encrypt is adding two additional remote perspectives for domain validation

https://community.letsencrypt.org/t/lets-encrypt-is-adding-two-additional-remote-perspectives-for...
98•gslin•3h ago•26 comments

Harness Engineering 101: The LLM is the brain, the harness is the body

https://github.com/IsuruMaduranga/harness-engineering-101
98•isuruwijesiri•2h ago•47 comments

ESA's Hera Will View Asteroid Scars from DART's Strike

https://spectrum.ieee.org/esa-hera-dimorphos-rendezvous
36•rbanffy•3h ago•0 comments

Gdynia and the Dream of a Maritime Poland

https://engelsbergideas.com/notebook/gdynia-and-the-dream-of-a-maritime-poland/
60•lermontov•2d ago•13 comments

Hybrid Reasoning Agents: LLM agents that think in latent space, act through text

https://arxiv.org/abs/2610.18861
464•allinonai•17h ago•224 comments

Show HN: KiviDB, a faster Redis-compatible memory database and vector cache

https://github.com/kividbio/kividb
5•murtazayusuf•40m ago•1 comments

Beyond the Glass

https://longreads.com/2026/10/15/mother-museum-skulls-grief-agni/
6•Petiver•3d ago•0 comments

There's never been a better time to go into math (by: 3blue1brown) [video]

https://www.youtube.com/watch?v=LKrwhjOI8Nc
460•mdp2021•1d ago•593 comments

William Gibson and the trademarking of cyberspace

https://colo.to/cyberspace.html
233•Eric_Gullichsen•14h ago•109 comments

If you don't run your own LLM proxy, how do you know whats going on?

https://jrdsgl.com/if-you-dont-run-your-own-llm-proxy-how-the-fuck-do-you-even-know-whats-going-on/
244•amrrs•17h ago•164 comments

How to Train Your Robot

https://www.construction-physics.com/p/how-to-train-your-robot
96•crescit_eundo•3d ago•30 comments

Tmux and Claude Code: Sessions, Send-Keys, and the Ceiling

https://crystl.dev/guides/tmux-for-ai-agents/
96•ankitg12•8h ago•57 comments

Rabbits were a medieval royal luxury in England until the Black Death

https://phys.org/news/2026-10-rabbits-medieval-royal-luxury-england.html
116•samizdis•2d ago•112 comments

Someone behind the software

https://karrisaarinen.com/posts/someone-behind-the-software/
502•tosh•22h ago•183 comments

Dogs seem to agree with humans that 'bouba' is round and 'kiki' is spiky

https://www.sciencealert.com/dogs-seem-to-agree-with-humans-that-bouba-is-round-and-kiki-is-spiky
303•gmays•1d ago•174 comments

Laptop Based on 6502 CPU

https://www.maddox.pro/?p=448
169•nanochess•3d ago•81 comments

RepRapMicron

https://github.com/VikOlliver/RepRapMicron/wiki
424•RobotToaster•1d ago•67 comments

The DNS Museum

https://dns.museum
113•zdw•2d ago•13 comments

How Quickly Can the DoD Rebuild and Recast the Munitions Industrial Base?

https://www.csis.org/analysis/how-quickly-can-dod-rebuild-and-recast-munitions-industrial-base
109•gmays•14h ago•150 comments

Harness2 by Google Research – Recursive Agent Harnessing

https://github.com/google-research/harness2
33•irthomasthomas•3h ago•5 comments

Superlite: A Faster Alternatie to Superpowers

https://contextfirst.blog/superlite-hand-an-agent-a-spec/
120•iheatuchriswogu•12h ago•57 comments

Free Pascal compiler now supports dynamic packages

https://www.getlazarus.org/dynamic/
183•sysrpl•19h ago•44 comments

Beyond These Stars

https://store.steampowered.com/app/2295060/Beyond_These_Stars/
4•doener•28m ago•1 comments

Show HN: WhatsApp MCP running free on Cloudflare

https://github.com/finereli/whatsapp-mcp
134•zoozla•16h ago•70 comments