AI Lab · by Thandi Mokoena · 2026-10-19

Qwen3-4B with the AI writing tics trained out

Someone fine-tuned a small Qwen to drop the 'delve', em-dash and 'not just X but Y' tics. Does a de-slopped model actually read better to you?

43 sparks · 15 comments

Comments

as an enthusiast I'm just happy to see the de-slop model getting attention, hope it keeps going

Mia Bauer · 9 likes

my wallet just cried reading this

skeptical but curious about the de-slop model. Anyone got actual numbers or just vibes

Classic Monday energy

Saving this thread, the de-slop is exactly what I needed to read today

Jiho Lee · 11 likes

Tbh I read it twice and still not sure what I think

the de-slop model will be boring in a year and everybody will use it anyway lol

Thanks, that actually helped a lot

Priya Nair · 4 likes

Can someone explain the de-slop model like I'm five? Feels like I missed a few chapters

lol this is too accurate

cost matters a lot here, the de-slop model only makes sense if the numbers actually work out at scale

Thanks, that actually helped a lot

Fair point, I hadn't thought about it that way

Not convinced about the de-slop model tbh. I've seen this hype cycle before and it usually fizzles

Ha, my cat has teh same opinion and zero evidence either