Adei @AdeiWeb2

HOW ANTISLOP REWRITES AI SLOP WITHOUT AI So the running theme of the industry is to solve AI slop in two ways, and we’ve combined both and also invented our own. I know founders, content creators, developers and all of them, including anyone who has interacted with AI ever, knows that AI has patterns, often referred to in general as “AI slop” is actually a combination of burstiness, patterns, robotic sentences, and phrasing. HOW AI REWRITERS WORK: Let’s take a standard boilerplate AI rewriter. The approach of any AI software company or AI user is to solve the problem using AI. Why not right? It’s a large language model perfectly capable of doing one thing extremely well: Generating text. The thing is, no matter how you prompt it, the model is always prone to a certain set of tendencies, without going too technical, frontier in AI in general, general purpose models in particular, are engineered incredibly well for their use case. Yet, whether you use Astra 6, Sol, Opus 5.5 or Sonnet or Gemini or Grok there are certain generic patterns and tendencies that you will encounter. Particularly painful for anyone who uses the actual writing (from text generation) for some form of profession. Additionally, society is adapting by using AI detection methods. An AI rewriter or ‘’humaniser’’ exists to evade detection. Currently the marketplace has both, 1/ AI detection and 2/ humanisers or rewriters with some software focused on writing styles or ‘’brand-aligned content’’. The issue with AI rewriters in general is that while this solves one problem A/ writing or text being detected and flagged as AI, it doesn’t actually resolve the bigger one B/ text sounding like AI, and it contends with yet another C/ systems where AI is used to cheat (education, contests, originality). This is because AI rewriters will typically use a frontier model anyway, rarely do they have their own in-house AI systems and even if you did avoiding the generic patterns and tendencies requires a significant amount of training effort. Large language models generate sentences that cluster around a remarkably uniform length They pad out thoughts with introductory clauses, passive voice, or standard filler words to maintain a mathematically safe, "optimized" flow Let’s say you solve that too, then you have to deal with another issue, “burstiness”. Regardless of how you set your LLM up, fundamentally it’s still predicting text word by word. Or token by token. Burstiness measures the variation in sentence length, rhythm, and structure across a piece of writing. Humans write in bursts, often varying massively. For example we may write dense longer sentences full of commas, clauses, parentheticals. Factor in vocabulary too. And then they will follow up with a short one and write a coordinating conjunction at the start of a new sentence. While AI tends to produce sentences of a relatively similar length and grammatical complexity one after the other. The industry solutions at present only focus on rewriting in a more ‘’human way’’ by generating new text that is more varied in burstiness. Software for writers focuses on telling an LLM to follow a certain style, which it can do reasonably well, yet fails burstiness and introduces new issues, grammar, generic patterns etc… Templated systems of the 2021 Era that writers still pay for are failing them in multiple ways. Solving “generic sounding AI” has by user feedback and obvious investigation NOT been achieved. Using AI to rewrite AI either evades detection while still sounding robotic or copies a certain style reasonably well but still sounds robotic with uniform sentences. WHAT ANTISLOP DOES: So we do none of that. X evading detection X rewriting whole text using another AI X humanizing We just… deslop it. Actually, from the ground up, Antislop is not a traditional rewriting engine because it uses linguistics not tokens. Existing solutions rewrite in tokens while we rewrite in syntax. As a result there are two things we do (at present). 1- (Antislop deslopper) The first thing we do is isolate generic AI patterns, known as slop, and structurally rewrite it using linguistic mathematics. Basically think of a phrase or a pattern that you just KNOW is AI. We take that, deconstruct it, preserve the meaning or fact within it, and then restructure it. Because all of our ‘’reconstruction’’ is specifically designed to be a solution to slop, and NOT a simple rewrite, Antislop rewrites fundamentally do not re-introduce new slop. We also realised that if we built this system solely on using AI for rewrites, two things would be impossible: Achieving consistent rewrites and avoiding re-introduction of slop. 2- (Antislop naturaliser) The second thing we do is take natural human writing quirks, tendencies and patterns that AI does not use and focus on overall writing structure to make it sound natural. So sentences, word choices, language as a whole is pointedly restructured rather than rewritten wholesale. So one could ask if we do this to avoid detection, we don’t. Not once have we or will we ever test whether our system can evade detection, it’s simply not even a consideration in our mission to provide professionals, writers, AI users and the general population with less slop and a more human, more natural and more opinionated, experience driven internet. If we provide AI users with a way to easily, quickly and reliably remove those patterns from their text we improve the experience for them, their readers, and the wider internet who are sick of constantly reading those same patterns. AI is a beautiful tool for individual innovation, yet that innovation is overshadowed by its generic patterns and tendencies that readers avoid and professionals spend hours fixing, painfully. No one thought we could do it but we can. HOW ANTISLOP WORKS: Our core product is a deterministic, source-bound linguistic rewrite engine with variance capture, a constrained Formula VM, parser-backed evidence, typed edit programs, unified multipass composition, and Ledger-backed publication. In other words, it doesn’t use an LLM for rewrites. Instead LLMs, or AI, is used in our systems for what it is genuinely brilliant at. Antislop is currently engineering its AI proofer for increased reliability and accuracy. We want to provide the best possible rewrite experience for users. Using AI to catch errors, broken meaning, or broken grammar is a great way to deliver reliability, since our system is deterministic at its core such errors across all users varying inputs are unpreventable to 100% reliability. AI can and will also play an influential role in our next product iteration, which we can’t reveal yet. Either way the distinction will be clear and optional via the UI. Antislop started as an AI first project with innovative prompting systems. It’s now a combination of AI and non-AI tooling, with rewrites focused on the latter. Antislop is product first above all. Our core product is AI first, and foremost simple input/output frontend available at app.antislop.pro

@alexgg_ai Big dreams

The internet is Alt + F4-ing itself Being born right at the turn of the century in 2000 let me try @YouTube for the first time when I was 7 years old. @Snapchat a few years later. The first iPod, socials, @facebook before most even knew what social media was. But it's changing:

antislop @antisloppers

2 weeks ago

We’re hyped to introduce Antislop to the world! Finally a way to reliably and easily remove slop from all of your AI generated outputs with one click. Save the internet. 0_o/

13 9 25 3K 3

Antislop is basically Ctrl + Z for the internet! Check it out: @antisloppers Together, we can save the internet! Web: app.antislop.pro

Adei @AdeiWeb2

2 weeks ago

The internet is Alt + F4-ing itself

1 1 14 535 0

@yubbos @antisloppers Less slop more you <3

The internet is Alt + F4-ing itself

Adei @AdeiWeb2

2 weeks ago

HOW ANTISLOP REWRITES AI SLOP WITHOUT AI So the running theme of the industry is to solve AI slop in two ways, and we’ve combined both and also invented our own. I know founders, content creators, developers and all of them, including anyone who has interacted with AI ever,

3 8 23 1K 0

@GLaDOUGHSandHiv @YouTube @Snapchat @facebook Both! Bifurcation bby

@S1Y4HS4NC4KS Awesome and thanks!! It’s at about 5% of where it will be soon. Definitely expect Beta to be a bit broken. Always open to feedback as well 🙏

@piselliii @YouTube @Snapchat @facebook In every possible way a great question and in every possible way this is the opposite of what Antislop is designed to do! So yay for us <3 x.com/AdeiWeb2/statu…

Adei @AdeiWeb2

2 weeks ago

HOW ANTISLOP REWRITES AI SLOP WITHOUT AI So the running theme of the industry is to solve AI slop in two ways, and we’ve combined both and also invented our own. I know founders, content creators, developers and all of them, including anyone who has interacted with AI ever,

3 8 23 1K 0

@ben_sage capitalism

@ananyairl 🔧

@aixarizzo praise tha lord

this will be needed. because internet is gradually becoming one massive ai agent, and without even realizing it, we are all positioning ourselves as sub agents within this massive network. that’s why ai detection, labeling, and similar tools will become increasingly useful and important, especially for video and visual content, both technically and ethically. i think projects like this will play a huge role in the future. good luck!

@AdeiWeb2 @YouTube @Snapchat @facebook Kudos for this innovative tooling. You don’t know what you’ve done for the internet. With Antislop, we’ll finally be able to read content that actually sounds different instead of the same flat tone everywhere. Reading the same robotic style over and over has always been boring.

Personally I think the concerning part is a future where we can’t even tell when original from fake and it gets crazier each passing day Thinking has been replaced by recycled ideas, overused phrases, and the same content wearing a different outfit each time Although i still don't think ai should be eliminated tho but used as a proper tool not the other way around That’s one reason I’ve been keeping an eye on @antisloppers. Worth to keep an eye on.

Adei @AdeiWeb2

2 weeks ago

The internet is Alt + F4-ing itself Being born right at the turn of the century in 2000 let me try @YouTube for the first time when I was 7 years old. @Snapchat a few years later. The first iPod, socials, @facebook before most even knew what social media was. But it's changing:

15 12 36 3K 1

We're losing the ability to tell real from fake, and it's getting crazier by the day. Everything is just recycled ideas in a new outfit. I don't think AI needs to go away, but we have to use it right, not let it use us. That's why `@antisloppers` is a must-follow.

Adei @AdeiWeb2

2 weeks ago

The internet is Alt + F4-ing itself Being born right at the turn of the century in 2000 let me try @YouTube for the first time when I was 7 years old. @Snapchat a few years later. The first iPod, socials, @facebook before most even knew what social media was. But it's changing:

15 12 36 3K 1

Raw unpolished content will win in our world of AI slop

@wujir0 Exactly

how easy is to switch and how many users are unwilling to put up with arbitrary pricing or model degradation? since their workflows depend on capable models and reasonable pricing, they are very opinionated and ready to move if necessary. that said, there's a point where large model providers can increase prices and maintain the same revenue if not more, even IF x amount of users leave. > revenue up while users down is real. > also simply consider hardware inflation. unfortunately in my opinion open source is not keeping up fast enough with frontier models, so when @wujir0 says the industry getting more expensive is a natural path, it's true, since I believe the pricing power is still centralised. I also think API pricing is the wrong measurement, in fact I think worse, it just won't stop. but it will also be slow, user pushback will slow it further, and pricing is ultimately capped at how useful AI actually is, so same output at higher prices may be a losing battle for companies - of course, they will still try to win it, since they don't really have a choice. > I'm still net bullish I think users, smart users, will adapt by using older cheaper models with more extensive, step by step reliable workflows.

coops • b300 maxi @wujir0

3 weeks ago

Are you ready to pay $500 a month for AI? there's numerous levers closed source AI pulls to increase margin. - new models that use more credits in/out - embedded bloat (yappy replies, inefficient thinking) - reduce the number of credits allocated to each plan in hourly /

36 11 52 3K 1

@wujir0 @YouTube @Snapchat @facebook Stop it. Deslop it 0_o/

@Chaos_dev_ @NhungNaive_ Ok we’ll onboard her once we’re ready to expand 0_o/