[
  {
    "startMs": 13000,
    "text": "So I'm here to talk about, uh, all these words that we call slop, and, uh, more importantly, I want to, uh, try to help fixing what Kirpatim and, uh, termed as the, the slopocalypse that he thought would start in, uh, 2026. And I think we've seen that on social media. So, uh, to try to fix this and to stop slop, we first need to define it."
  },
  {
    "startMs": 43640,
    "text": "So first, what is slop? There are tons of different definitions from the various experts in the field, but what I think we can all agree is that it really sucks to read something that feels generated. And it usually looks like this with, uh, the, the words that we know of, the delve and, uh, the generalities and other things like these. And the main problem is that even as models get better with Fable, uh, GPT-6.1 and all the, the better models,"
  },
  {
    "startMs": 74020,
    "text": "slop doesn't seem to go, uh, go away. It just keeps on scaling. So we need something different than just training the models more it seems. And, uh, I really want to try to at least first define what slop is so that we can understand why it's there. And so to me, slop is made of six main thing. First, there are tons of overused words by language models that we don't really use, like delve, that we really know well. Second, it's usually"
  },
  {
    "startMs": 104000,
    "text": "too general. It's always talking about most themes, most projects. It doesn't share nothing specific. And I'm myself guilty of using that on LinkedIn as well, which I guess it's the LinkedIn meta so it's, uh, it's normal there. And worse slop, uh, sloppy text usually says nothing. It's just like nice words put together, uh, that we don't even usually use. And even worse than that, it says nothing, but it also repeats itself a lot."
  },
  {
    "startMs": 134280,
    "text": "So it repeats nothing and it's too general. And, uh, even worse, it hallucinates. As we know, models hallucinate, so the sloppy text also, uh, we cannot really fully trust it. And lastly, the worst to me is that it usually has the wrong tone or at least it has no tone and like everyone's tone at the same time. So it just doesn't really talk like you. It doesn't share stuff that you would share. So my goal here today and with this"
  },
  {
    "startMs": 164140,
    "text": "project that I've been working on is to first measure this sloppy writing. Uh, not just slop words. I don't want to rebuild EQ-Bench Slop Score. It's just, it's really about measuring how sloppy is a writing. So in order to do that we are, well, today, right now, releasing a new benchmark that we called ToneBench which, uh, the whole goal is, of the benchmark is to measure if the new models or models are good at creative writing,"
  },
  {
    "startMs": 194100,
    "text": "which I will try to define as well. So here's the plan for today. Uh, the plan is to justify do we even need another benchmark? Then how it works, what we measure, what, how we built it. Uh, and I want to share like the cost of building it, uh, our decisions, what we learned already from, f- for writing with AI from the benchmark, and, uh, share some more stuff about this. So with that, uh, let's delve in.  So the"
  },
  {
    "startMs": 223900,
    "text": "first, first thing, do we even need another benchmark again? And just looking at LinkedIn and X, X replies, my replies on X, I think we do. It's like people are spamming more and more generated content, so at least have them be decently good. And, uh, more precisely, I also really need it. We are a small team, uh, at Towards AI, my company, and we have one side of the company that is focused on"
  },
  {
    "startMs": 253920,
    "text": "education. We build tons of educational content mostly for AI engineering. And as a small team, obviously we use AI as well for that, and it's been extremely annoying to review the slop, uh, constantly and try to fix it. We, uh, I published hundreds of videos. We published thousands of articles. We released a book. We are working on a second one, uh, for AI engineering. And so we really need to have the best models possible that will help us write better, uh, and"
  },
  {
    "startMs": 283940,
    "text": "not be full of slop. And another nice, uh, addition to that is that I've been sharing about this benchmark on, on X for quite some time as I've been building it and it got really nice engagement and people seem to confirm that we did need some sort of more creative, uh, writing benchmark like this one. And lastly, just to prove that I think we are the right team to work on"
  },
  {
    "startMs": 313960,
    "text": "this benchmark, we-- Well, I've been build- evaluating AI models since my master's and then PhD and now at Towards AI we do build evaluation pipelines for our clients. We have 15 AI engineers with us and we have creative writers, editors, uh, on board. So, uh, we already have that for our content. So I think, uh, well, I wanted to leverage that to build a benchmark and help the community even more. So what did we build"
  },
  {
    "startMs": 343940,
    "text": "exactly? Hopefully I will try to share as much as possible and I'm happy to chat, uh, if we can help, uh, even build your own or if I can just share some, some of our learnings. So what we built is super subjective. Uh, it's a way to measure if a model can write like me. And how I define that is the, is just that if I can read it out loud like right now without cringing because of the- The, the, the words that I wouldn't use myself. So"
  },
  {
    "startMs": 374216,
    "text": "the process is very simple. The, the benchma-- the process benchmark... The benchmark process is very simple. We have a, a prompt that is the brief and the research that we send to a model, and hopefully it will generate entirely a video script that I could have written. Uh, obviously, that doesn't happen. It really sucks usually. But that's the whole goal, and we measure that, that final script against our rubrics and, um, just our whole evaluation"
  },
  {
    "startMs": 404176,
    "text": "pipeline. And the prompt that we send to every model in the benchmark is much more, uh, in-depth than that. It, it looks more like this, where we have a lot of details on how I talk, how I write, all, uh, tons of examples of my previous content, videos. We have a lot of information on storytelling in general, active sentences, and just how to write well. And we have directions on what to generate and what to do for this current"
  },
  {
    "startMs": 434096,
    "text": "task, which in the end, the model will generate one script. And this is just one of them. In the current benchmark, we have ten different scripts, uh, that we worked really hard to make as good as possible, and we built rubrics for them to help, um, as we will see, LLM judges, uh, rate these, these generations. So we have ten different scripts. Some of them are published, some are not. We don't say which ones are the benchmarks, so that we check for contamination as well."
  },
  {
    "startMs": 464796,
    "text": "And, uh, then what do we do with these ten scripts? We take all the models out there, and we try to measure their generations. Uh, what that means is that for every single model, we will do ten times this process, generate the article from the prompt. In fact, we generate five of them. I will very soon share why. Then we have four different LLM judges analyze them. I recently added Jeff as a test. Uh,"
  },
  {
    "startMs": 494976,
    "text": "not sure if I will keep that, but, yeah, I, I, I added it just to see if it worked well. And then we calculate in the end a final score for the leaderboard. I, I will go through all of these steps, uh, from left to right. So first, why do we generate even five scripts? It's f- to check for intra-model variability, which I highly suggest if you are building your own evaluations because especially I guess on a creative task, sometimes, as we see here with Grok 4.7, some of the"
  },
  {
    "startMs": 524656,
    "text": "generations are outliers, where it's, as we see for- from the benchmark, it's just as good as Opus 5.5, but on just one generation out of five. So we want to measure for that. And the worse the model is, the widest the variance. So, uh, we have some learnings about that that I'll share at, at the end as well. But yeah, it's, uh, we, we found that five drafts seems to be the good middle ground in terms of, of cost for us to execute and, uh, to measure this, this variability."
  },
  {
    "startMs": 554956,
    "text": "So what do we measure exactly, and how do you even quantify writing quality? That's been the challenge, and we worked on it for, for quite some time. Right now, uh, it's all open on the benchmark, and there are way more details on the benchmark page as well. But what we measure is nine different metrics that we detailed, uh, that, that we really detailed. But I can condense them in five main aspects. First, we measure the tone. So does it sound"
  },
  {
    "startMs": 584996,
    "text": "like me? Which means that, like, can the model adapt to a specific tone? Second, we measure just pure is it good writing? Uh, does it have, like, good hooks, good emotion, uh, emotional transition, and all the, the aspects that we're looking in a, in a good story? Then we measure if it follows instructions. Did it follow the, the length, the, the expected length, the format expected? Uh, did it put the right images and right, right things in the right place?"
  },
  {
    "startMs": 615896,
    "text": "Fourth, we measure for research accuracy. So, uh, we have a research brief along with each script, and we just measure if it hallucinated or used the research correctly. And lastly, we measure for slop using EQ-Bench Slop Score, but also our own measurements. And what, uh, this looks in terms of what the judges see is more like that. We have around a thousand words of examples and,"
  },
  {
    "startMs": 646176,
    "text": "a- and instructions and directions around each of the metric that we have, along with the script to analyze, my own, uh, script a- as some sort of, of guidance, and the brief and the research. So all in all, it's a fourteen thousand prompt that we send to, to every judge to then get, uh, the feedback that we'll dive in next. And, uh, so, so the judge just analyzes the writing quality. It doesn't focus on length, uh, and doesn't know"
  },
  {
    "startMs": 676116,
    "text": "which model is writing. Uh, we just do that after with, with code itself. On one metric, like tone and voice, which is our main metric, it looks a bit like this, where we have a lot of detailed descriptions of what is my tone and what to expect and how to analyze this. Then the judge will categorize between six categories first, and after having found a category, it will give a rating. Uh, we"
  },
  {
    "startMs": 706156,
    "text": "found that this is the best way to measure for this more qualitative task from, uh, just binary that we usually suggest to use or, uh, directly categories or directly percentage. We found that this actually works really well. And, uh, it has tons of examples of what each score means in terms of, of writing for each of the tasks. And what the judge send back is something like this. It has the nine"
  },
  {
    "startMs": 736160,
    "text": "different metrics scored with a justification for a- for the score and an example of, like, uh, I, I wrote this score because of this example in particular. But the main thing is that all of this is automated, uh, obviously because we, with 120 models, we, we generate five drafts each for 10 tasks. It quickly goes to the thousands of scripts to review, so we cannot really review this ourselves. And the problem with LLM"
  },
  {
    "startMs": 766020,
    "text": "judges is that they have preferences. Uh, some models prefer themselves. Most models prefer themselves, and, uh, particularly OpenAI. And what I found quite funny when studying this is that it seems like Anthropic and OpenAI are really in a, in a war here where they seem to have post-training on hating each other  which is quite funny. But all this to say that you really want to check for-- When you ha- when you use LLM judges, like"
  },
  {
    "startMs": 796000,
    "text": "using GPT for instance, if you want to know which model is best for your task, uh, be careful and use different model family just to ensure because we see here that OpenAI really has a strong preference on its own family of models. Not just one model, but all OpenAI models."
  },
  {
    "startMs": 814040,
    "text": "And so to fix also for that, we listened to a few comments that I got on X suggesting to use human for a creative task, which would be ideal, but quite impractical. We still tried to, to do that in a scalable way where right now we compared manually, me and three of my teammates, one thousand AB pairs of, uh, the generated scripts. Not the full scripts, but like a good chunk that is comparable between"
  },
  {
    "startMs": 844020,
    "text": "two scripts, two generations. And hopefully we'll do more and I want to publicize this also for, uh, to get help from the community in the future, but that's not, uh, right now. So what did it cost us and what's the workload behind that? For me, for us, it was a kickoff cost in terms of hours of around five hundred hours to build the 10 scripts, uh, define all the rubrics, find the examples, uh, try to quantify quality which"
  },
  {
    "startMs": 873880,
    "text": "is very difficult. And then I just iterated myself with, uh, four different 20X subscription for four months to build a benchmark and all the other things and experiment and change the, the met- the, the metrics, uh, ratio and all these things. And so we just spent, uh, thousands of dollars there. And, uh, if you want to do this, a, a similar type of benchmark, uh, you need to know that it costs quite a lot. Uh, right now if I just"
  },
  {
    "startMs": 903780,
    "text": "change one thing in the rubric, so how the judge rate, um, generations, it will cost us two th- more than two thousand dollars and if I just change one line in an input prompt for a g- uh, a model when it generates, it costs us more than four thousand dollars. So it's very expe- uh, very expensive to experiment and change these things in the prompt, so you want to do this, uh, carefully."
  },
  {
    "startMs": 929040,
    "text": "Now what can we get out of this benchmark, out of ToneBench? Uh, you can get tons of things on the, the website itself. There's like a, a lot of details. But importantly, uh, I want to share that what I've learnt from the past four months is that you don't want to pick for, for creative writing, is that you don't want to pick the model that will generate a one-shot amazing post. You just want to find the model that will be the least annoying to you to get to where you want to be. So, uh,"
  },
  {
    "startMs": 958940,
    "text": "like there won't be any perfection, uh, anytime soon I think. And what that means right now on the benchmark, it means at scale, uh, GLM 5.3 Flash is extremely good, extremely cheap. It's 50 times cheaper than the best one which is Opus 5.5 Extra High which is, uh, 2.5 times cheaper than Max Thinking, but with same results. So you don't want to use Max for writing usually. Um, and otherwise for the best, uh,"
  },
  {
    "startMs": 989640,
    "text": "how to write with models that we got out of this is to first have an extremely long and detailed prompt. It cannot be long enough and the more the better. You want to add every example you can think of, of what to do and how to do it. Same thing for negative examples of things you don't like and that sou- doesn't sound right, the words you don't want, everything. You, you can share every-- If you write newsletters or whatever you can share it in full and it will just be better as well."
  },
  {
    "startMs": 1021280,
    "text": "And, uh, what we've seen that really, really helps is to use a high-thinking judge just one time, one to three times otherwise it will loop into slop. So just one review of another model with the same instructions, directions that will give feedback to the writer. Uh, really like I don't have the quantitative metrics on that but it, it helps a lot. And for a company if you work with APIs and want to do this at s- have the best results at scale possible, what I"
  },
  {
    "startMs": 1051200,
    "text": "suggest is to generate several drafts like 10 drafts of one specific thing with a cheaper model like GLM 5.3 Flash and then have a smarter high-thinking judge like Opus or, or whatever you want pick and merge the best parts of each of these scri- these drafts which I think, uh, will still land way, way less expensive than just one generation by Opus 5.5 and, uh, I think will lead to better"
  },
  {
    "startMs": 1081160,
    "text": "results from what I've seen."
  },
  {
    "startMs": 1084400,
    "text": "And lastly you want a human review. You cannot really go, uh, around that. But all of this benchmark measures obviously my voice so it's super subjective but I-- and, and it's just one data point. You don't want to fully trust it blindly. It doesn't mean Opus 5.5 is the best model. So but ideally it's a strong signal to find which model is good at writing and which is less sloppy. Um, y- I suggest you to build your own evaluation pipelines. If you want we can help. Uh, you can"
  },
  {
    "startMs": 1114280,
    "text": "chat or reach out. And regarding the benchmark it's live. We just like right now published it and released it. So, uh, there are tons of, uh, graphs, tons of insights, tons of comparisons, uh, and details on the implementation, full methodology. So just, uh, check it out and hopefully it's useful. Thank you."
  }
]
