Zhizhuo (Z) Zhou

@zonderlabs, generative 3d @Stanford, @CarnegieMellon, @UMich

[Zhizhuo (Z) Zhou @zhizdev]

Why is AI slop so frustrating to read?

AI writing often obscures main ideas and invents awkward phrases, making sentences more complex than necessary.

Many of these slop patterns violate writing principles established in English writing books.

In our benchmark, AI writing breaks these principles far more often than human writing, with Claude Code replies by far the worst offenders.

We show examples of the most common slop patterns in this thread

[Zhizhuo (Z) Zhou @zhizdev]

@jefftan969 @StanfordSVL this is super cool!

[Zhizhuo (Z) Zhou @zhizdev]

@ND06517041 yes! passive voice can be far more natural when used correctly

[Zhizhuo (Z) Zhou @zhizdev]

Why is AI slop so frustrating to read?

AI writing often obscures main ideas and invents awkward phrases, making sentences more complex than necessary.

Many of these slop patterns violate writing principles established in English writing books.

In our benchmark, AI writing breaks these principles far more often than human writing, with Claude Code replies by far the worst offenders.

We show examples of the most common slop patterns in this thread

[Zhizhuo (Z) Zhou @zhizdev]

@ram_dw @AlexReibman @addyosmani bro this website is AI slop pro max++, there is not a single clear sentence

[Zhizhuo (Z) Zhou @zhizdev]

@WaifuverseAI this would be scary

[Zhizhuo (Z) Zhou @zhizdev]

Fable 5.2 is not better at writing

A perfect example of why we need to align AI writing for humans!

[Zhizhuo (Z) Zhou @zhizdev]

Why is AI slop so frustrating to read?

AI writing often obscures main ideas and invents awkward phrases, making sentences more complex than necessary.

Many of these slop patterns violate writing principles established in English writing books.

In our benchmark, AI writing breaks these principles far more often than human writing, with Claude Code replies by far the worst offenders.

We show examples of the most common slop patterns in this thread

[Zhizhuo (Z) Zhou @zhizdev]

@karenzzhou yes! i think there is a human mental capacity alignment here where we want AI to write in a way that maximizes information given the same amount of mental capacity (human attention and focus)!

[Zhizhuo (Z) Zhou @zhizdev]

I think the first step is to make writing not suck, ie not break very obvious writing principles in Elements of Style. This is likely solvable with a small army of high school english teachers.

Going beyond making writing not suck is much more subjective and artistic, and maybe it would be closer to hyper personalization

[Zhizhuo (Z) Zhou @zhizdev]

@rulimanurung @thomashikaru a huge fan of this paper!

[Zhizhuo (Z) Zhou @zhizdev]

@zehavoc this is perhaps something only LLMs can understand

my hypothesis is what we need for writing to be clear is actually pretty inefficient bc our memory is not as robust as giant transformers, and therefore, we have a hard time reading LLM jargon

[Zhizhuo (Z) Zhou @zhizdev]

@babaliauskas brooo

[Zhizhuo (Z) Zhou @zhizdev]

@SengirShowsU when a sentence is unnecessarily complex, it makes it harder for the reader to understand. I roughly use clarity to refer principles in writing books such as Elements of Style

[Zhizhuo (Z) Zhou @zhizdev]

@yiz_be_building yes! AI needs to take high school writing class

[Zhizhuo (Z) Zhou @zhizdev]

@illulachy each principle is scored on its own. the most identifiable principle that separates AI from human writing is just its awkward word choice

[Zhizhuo (Z) Zhou @zhizdev]

@EmeryJohnsonIV I think that day for video / image / audio surprisingly close

[Zhizhuo (Z) Zhou @zhizdev]

@PageBaltimore thanks for the submission! I'm collecting a slop dump for fun of the worst AI responses

[Zhizhuo (Z) Zhou @zhizdev]

@juliette_hiyb lol great point

[Zhizhuo (Z) Zhou @zhizdev]

@alexyrtsai one end of good writing is art form, which would be hard to imitate. the other end is just not terribly incorrect writing, which can be modeled

[Zhizhuo (Z) Zhou @zhizdev]

@thisIsSrijon AI needs to retake high school English

[Zhizhuo (Z) Zhou @zhizdev]

@GuangyuRobert slop writing seems to be mainly an artifact of post-training, @thomashikaru has a great paper comparing base models, pre-frontier models, and near-frontier models

also transformers are way more capable than our brain, so it needs less "markers" to help organize information