#264 ElevenLabs Surprise, ChatGPT Stunner, YouTube Dubs, Microsoft Interpreting API

0:00Esther Bond:Translation made up a staggering 4.5 % overall, not of the writing, but overall. So 4.5 % of the conversations with Chatubiti are categorized as translation, which is nuts.

0:21Esther Bond:All right, back with another news episode of Slaterpot. Hi Esther and hi Alex.

0:26Florian Faes:Hi Florian. Hi Alex. Hello. Hi.

0:29Esther Bond:joined today not only by Esther, but also by Alex Edwards, Senior Research Analyst based in Madrid. He needs to help us unpack a big move in the LSI world now, moving from LTP to LSI, and he'll explain us what this means. Yeah, we have to just come back with another podcast before a quick one-week break, and then we'll be back with a great guest episode. So on the agenda that today we'll talk about 11laps's big move, 11laps, the audio AI dubbing, transcription, video editing platform, whatever they call themselves. Then YouTube's big AI dubbing unlock. We want to talk about a crazy translation stat from ChatGPT that was published this week, Microsoft's quite big, some would say huge move in AI interpreting.

1:22Esther Bond:Mistral, European LLM raising a couple of billion and then WIPO RFI-ing some work in the post-editing world. So we want to start off with 11 labs adding basically managed services and turning itself into what we consider at Slater an LSI from an LTP, a language technology platform into a language solutions integrator. her. Alex, tell us more. So what was the news? Why does it seem like a big deal? And then let's unpack both what Eleven Labs told us about this and what Justin Baudin, founder and CEO of Adapt Global, told us on the record today. So let's unpack this. Sure. So yeah, as you say, big news.

2:12Eleven Labs launching managed services, essentially now becoming an LSI. They've launched its productions wing so um they now offer experts in the loop for dubbing transcription captioning and subtitling and coming soon they're also going to be releasing an audiobook service um from 200 an hour which essentially uh supports both single and multi-speaker voice casting so um up until now they've worked with lsis as partners um so this is definitely a strange strategy.

2:47Esther Bond:Big, big, yeah, change in strategy. And we would call them in our framework now, I mean, they were one of the most obvious LTPs, like an actual language technology platform. Like it's not a TMS, not something that like, it was the actual platform. They produced the output, the output, they had the models, they offered the API. A lot of people were building on top of 11 labs you know their their audio applications and their their ai dubbing applications and now boom they're going and adding a human managed service on top of this so i don't know what to make of this this is this is you know think of think this is like deep l announcing that they're gonna uh post edit your you know your translations now for i don't know three cents a word or something like that uh so this would be kind of the equivalent here yeah it's crazy well we we actually reached out to 11labs to ask what's happening and they came back to us and said that they're now going to fully manage this in-house.

3:53So they're running everything through their own platform, their own tools, their own processes and they're not relying on those third party partners at least initially for some of that volume. And it seems like it's been cooking for a while they've hired a couple of people in-house for this uh they have in-house leads for every language group and they're on a big recruitment drive for um for linguists across multiple languages primarily for subtitle subtitlers dubbers they're really uh ramping up their their vendor management side so uh this they're going full in they're going serious with

4:29Esther Bond:this this is a yeah on a recruitment drive for linguists so if any linguists listen to this you You know, go and head over to the 11 Labs recruitment page and, you know, submit your resume.

4:39Florian Faes:And vendor managers and language leads and everything else that they'll need to do such an operation.

4:45Esther Bond:Exactly. That is correct. Welcome to the world of slightly lower margins and infinite devil-in-the-detail complexity.

4:54Florian Faes:It's nothing entirely new, though, is it? I mean, we have seen this trend in the past where a company like starts out as a technology company and then somewhere down the line adds services. So is this one more shocking because it's kind of in the AI era or the scale of it? What is it that's kind of makes headline news, let's say?

5:17Esther Bond:To me, it's more shocking because, yeah, to me, this was the.

5:22Florian Faes:Like epitome or something.

5:24Esther Bond:No, it was the prime example of a company building a platform upon which others are building their own products or services. Right. So, all right, we have the foundational AI, like AI dubbing and audio capabilities. And now, okay, you come onto my platform and you build something that the end customer wants, right, while I'm charging you, you know, maybe volume-based or whatever, token-based or something like that. And now, yes, now we're going into the world, they're going into the world of the services, which is described with PMs and vendor management and invoices from linguists for$25 for a little edit and things like that.

6:08Esther Bond:So, and also... Massive pivot. Massive, yeah, pivot or an expansion, right? I mean, it's going to certainly generate a bunch more revenue, but that revenue will likely be, not likely, it will be lower margin. So, is that what you want for your next funding round? Plus, it may feel a little bit bait and switch for all of the people that build on it, right? So, if you were an LSI and you started building on it and now all of a sudden, boom, hey, now I'm competing with you and you know what? I don't need you anymore or your solution is always going to be worse than ours. Sure, it's a free market so they're more than welcome to do this but big move.

6:48Esther Bond:So, we also got some pricing data, Alex, right? And then we also pinged Justin Buda from Adapt Global about his take on this. Maybe you can walk us through this. Yeah. So on that point about partnering with Eleven Labs, Justin Bodine, the founder and CEO of Adapt Global, we kind of asked him for his opinion on this news. And he essentially said to us that it's actually good. He sees it as a tailwind rather than a headwind because you've got this big industry player validating the need for the human and tech combined approach. It is quite, you know, YouTube are obviously offering AI dubs for free, and now Eleven Labs are offering dubbing from$22 per minute.

7:34So it's a bit of a jump, but that just shows that there is a market there for the creators. And, you know, it's still possible that Eleven Labs will be outsourcing to providers like Adapt, of course.

7:47Esther Bond:Yeah, this is okay. Yeah, sure. It is a signal. It is a signal that you need that or that if you want to really make this work well for medium to high value content, you will need some type of human in the loop and that's true. I guess I agree with Justin there that this is somewhat of a validation, but yeah, it's also a pretty big threat if you're in that space. Yeah, I think the big threat here as well is that 11labs are saying that they are going after enterprises. They're not just pitching to creators, but they've also come out saying that they've got an enterprise offering. So that's with discounted per minute rates, expedited turnaround times and also advanced pre and post processing services.

8:37So yeah, maybe that$22 a minute will be lower for those enterprises that are looking to jump on an enterprise deal with them.

8:45Florian Faes:Is it like you attract customers at a sort of low rate, a low market rate, and then gradually increase the prices to be more standard, like a kind of Uber situation once you're in there with the organizations?

9:02Esther Bond:I'm sure it is. And you lock them in, et cetera. Is that price$22 a minute? That's kind of in the medium range, I guess, from what we've seen? Yeah, that's more or less medium range. From the AI dubbing report that we released a month or two ago, we saw it as about$5 to$50 a minute, essentially to have for human services. So$22 is probably bang in the middle. Right now, that's their starting price for the 11 languages that they offer, but they're going to be extending that. And then it depends on your actual scope, whether that will be more expensive or less expensive. So this podcast would cost$660 to dub into one language.

9:44Esther Bond:If you go 11 languages, then it's more. It's$7 ,260. I would expect perfection. And if I don't get perfection, I will chase the project manager until I get perfection and that will lower 11 laps margin. I'm just kidding. All right. So basically this news comes as YouTube just enabled multi-language audio tracks, which is obviously a key use case for 11 labs and the thing we've been talking about a lot, if YouTube basically opens or allows creators to upload their own audio tracks, boom, you have a massive new, I don't know, avenue for publishing this, a massive new incentive for actually creating these AI DoB and human post-edited audio tracks.

10:34Esther Bond:So Alex, maybe tell us a bit more about that move by YouTube there. Yeah, so we described this as a bit of a plot twist for YouTube because YouTube, only a couple of months ago, they released the automated AI dubbing feature, which essentially enabled you to click a button and generate multilingual audio for your YouTube channels in any number of languages. but the quality was so bad that creators were asking to upload their own audio dubs that they create themselves.

11:12That audio feature was initially launched in February 2023, but it was only available to a limited number of creators. So it's finally being rolled out across the board and essentially enabling creators to get your AI-generated human improved dubbing from 11 Labs, for example, and uploading it to your YouTube channel, or indeed self-recording your dubbing and uploading it to the channel. So there's a lot more options now available for YouTube creators.

11:44Esther Bond:I remember that one of the issues was, and I spoke to somebody who's very well informed on this, that for YouTube, it's really tough to avoid spam or like kind of harmful, bad content on those uploaded tracks. So that's why they were super cautious in rolling it out initially, just with the biggest creators they have more control over. But if you open the gates here to millions and millions, you're also going to get some bad actors that are uploading really bad audio tracks that are maybe not in line. So it's just another, I guess, attack vector for bad actors to upload weird content, right? Makes sense.

12:21Esther Bond:Yeah, they have less control over what's being said on the dubs.

12:26Florian Faes:Or less ability to sort of monitor it effectively.

12:29Esther Bond:You'd think that Google has figured this out. Like they can, you know, churn through gazillions of, you know, petabytes of data and kind of understand what's being said on the audio dubs, on their multi-language tracks, but it looks like they figured this out. You know, it does open up, as we said in the past, new revenue opportunities for creators. I mean, if you have a podcast or a show and you can, you know, all of a sudden quite seamlessly AI dub or human post edit AI dub into a bunch of other language is that that'll increase your view count increases your um your revenue opportunities on youtube so um yeah pretty pretty big move yeah and um i think what's interesting about this as well is that the lsis that have serviced these creators up until now uh they've been creating separate channels for each language um and uh essentially managing those separate language channels on behalf of the creators and now with these um multilingual audio tracks you have the ability to upload the audio directly to the original video so um it'll be interesting to see how lsi has changed their kind of operating or their service mix essentially you know whether

13:43Florian Faes:they're going to be yeah yeah i mean that would be i think that's just gone then i mean there's a your own language-specific community for that channel. You also have your own, maybe if you have sponsorship opportunities in each market, you can monetize an addition per market. So there are some opportunities there for the multi-channel approach. But yeah, there's now this new avenue open to creators and other sites to offer perhaps a scaled-down service.

14:23Esther Bond:That's just a couple of engineering steps away though. If you think about it, let's say you switch the language selection to Spanish or Hindi or whatever, and then with that, you're also changing the type of ads that are getting played or the type, I mean, some other kind of more custom things that'll get triggered by you changing the language. Yeah, very, very dynamic environment and probably 11Lab sees this and says, well, we want to get into the game here. We want to capture a lot more revenue because we need to raise our next round. No, I'm obviously absolutely welcoming anybody from 11labs to join this podcast and explain it directly.

15:03Esther Bond:We did offer it, to be honest. It was a little less second, so, you know, so far we have nerd back, but we'll definitely try to get them on the podcast or because a lot of them are based in London, I hope they join us at the next Slate of Come London. Crazy stat from OpenAI about ChatGPT and translation as a use case. So let me read through the ChatGPT generated bullet points of OpenAI's research paper on ChatGPT's use cases. So it launched obviously ChatGPT in November, 2022. We know that. In the paper, OpenAI says about 10 % of the world's adults have adopted ChatGPT, you know, if you, with a grain of salt or whatever, maybe 5%, who knows, right?

15:55Esther Bond:Early adopters, they're saying mostly were men, but the gender gap is narrowing. Adoption has been growing faster in low-income countries. And to run this study about the specific use cases of ChatGPT, they said they were running a privacy-preserving pipeline. They also noted that non-work use, so now we're getting to the results of the study. So non-work use grew from 53 % to 70 % of all conversations. So it's moving a little bit away from the professional realm to the, I guess, personal realm. In terms of work use, it's more common among educated professionals with high paying jobs, no surprise there.

16:38Esther Bond:And then they, and now we're getting to the interesting part here for this industry. They looked at the conversation topics. So let me try to spin up for those of you who are on YouTube, a quick screen share here. So, okay. So they looked at what are the topics of ChatGPT conversations. And so they categorize it in broad topics such as writing, practical guidance, technical help, multimedia, seeking information, self-expressions, and other unknown. So number one category here on that list that people on YouTube are seeing would be writing. And then one of the five subcategories or the conversation categories is translation.

17:28Esther Bond:The rest would be edit and critique, provided text, personal writing, argument and summary generation, and write fiction and translation. So let's look at the translation stat here and they reviewed this or the numbers are big. So the paper is based on a sample of approximately 1.1 million sampled conversations that they sampled between May 15th, 2024 and June 26th, 2025. five, so it's quite a broad range there or a date. Just over a year. Just over a year and again, let me spin up that key chart here for those who are watching this on YouTube. We're also going to link this in the show notes so you can go and download the paper yourself.

18:23Esther Bond:So here's basically the chart that shows each category, like how much of these 1.1 million conversations were on which topic. And so let's look into the writing here. So writing made up 28 % of those 1.1 million conversations and then translation made up a staggering 4.5 % overall, not of the writing, but overall. So 4.5 % of the conversations with ChatGPT are categorized as translation, which is nuts. This is, you know, if you think about it, let's say 10 % of the population of the, like of the adult population of the world is using this all the time. I mean, you're talking about, you know, gazillion, just a ton of translation related queries speaks to the massive size of the need, speaks to the massive potential addressable market for anybody that has something that's maybe superior to ChatGPT for translation.

19:31Esther Bond:And sure enough, somebody from DeepL was among the first people to like the post that I put out on this on LinkedIn. So, you know, this does a giant macro tailwind to kind of, for example, the DeepL story, right?

19:46Florian Faes:It's pretty interesting the topics which are slightly ahead of translation or sort of the areas of use. So you've got, I mean, the biggest one would be people looking for specific info right so that's 18.3 percent like effectively using it like a search engine um and then how to advice so how do i do xyz the one that surprised me and so that was 8.5 then you've got like tutoring or teaching so how are you using i don't know that one kind of stumped me a bit that was 10 of all searches on the green bar tutoring or searching oh yeah there you go Teaching, it says, right? Tutoring or teaching. And then, I mean, there's lots that I guess you could unpack here that not all of it being relevant to us.

20:28Florian Faes:But the other sort of interesting thing I saw here was that computer programming is quite similar in terms of usage to translation. So you've got people using ChatGPT for computer programming at 4.2 % and then translation, as we said, at 4.5%. So I know in the kind of development and product management community, all of this is very, very big.

20:51Esther Bond:And I thought it was bigger than it actually is, if you think about it. Like it's 4.2 for computer programming versus 4.5 for translation. And if you go on to X, you'd think that it's 90 % computer programming and basically, you know, all of the rest would be negligible.