AI for Everyone: Staying Ahead of the Curve

Episode 198 July 18, 2024 00:28:57
AI for Everyone: Staying Ahead of the Curve
Localization Today
AI for Everyone: Staying Ahead of the Curve

Jul 18 2024 | 00:28:57

/

Hosted By

Eddie Arrieta

Show Notes

We chatted with Veronica Hylák, co-founder and co-CEO of Metalinguist, curator of the popular AI vlog, The AI Almanac.

Veronica shares her secrets for staying informed in the fast-paced world of artificial intelligence, and her thoughts on the recent Apple Intelligence launch. You'll also learn about her favorite AI tools and how the future of the translation industry is being shaped by AI advancements.

View Full Transcript

Episode Transcript

[00:00:02] Speaker A: So I started the AI almanac. Actually, because of myself, I'm a little selfish because I felt like I was falling behind on my AI research every single week or every single day. I felt like there was something new and I just could not keep up. And only one month of neglecting what was going on. I felt like my information was ancient. You know, there's so much noise. I gotta just figure out what the top four or five updates are and summarize it, right, because. Because odds are they're going to change in the next few weeks, right? Even, even the same entities change over the, over the weeks. And a lot of the updates that I give are themes, you know? So for example, Apple intelligence, when they announced their huge updates around Siri, anyone that had been following AI Almanac knew where that came from, because they published a paper about a month ago called the Realm model, which I talked about a month ago. So when, when Siri was announced with all of this crazy functionality and context and everything, I was already aware of it. And my listeners are already aware of it because we talked about that. [00:01:05] Speaker B: Hello and welcome everyone to localization today. My name is Eddie Adiata. I'm the CEO of multilingual magazine. And today I'm joined by the wonderful Veronica Heilac, co founder and co CEO at Metalinguist. We're going to be talking about amazing topics, including AI and the future of our industry. Veronica, thank you so much for joining us today. [00:01:27] Speaker A: Thank you so much for having me. [00:01:29] Speaker B: I'm so excited, actually, we are very excited to have you. I've been following your newsletter and I have to confess, I've been, as I was saying on the backstage, I was very impressed by the amount of information that you cover. And sometimes I'm very impressed by not only the details, but how surprising the details are. And these past weeks, we've been looking at your newsletter, and the topic that we're going to be covering today is feature in one of your newsletters. So of course, later on, we'll talk more about that and about how people can actually join your newsletter and follow you on social media and get all these weekly content. If I'm correct, this weekly, I think, because that's why I'm very surprised that you get all that content weekly. I don't think I can get it on a monthly basis. But back to business, Veronica, we want to know much more about you, much more about your company, and much more about the context that taking you in the path that you have taken as a professional. So to begin with, why don't you tell us a little bit more about who you are. Who's Veronica Hilac? How did you get to co found a company? [00:02:39] Speaker A: And I have been very heavily involved in the AI space way before. I actually joined the translation industry this year. Was voted Tas AI revolutionary, but came from the AIML side for years on the us government. So I actually worked on several top secret programs for the Department of Defense, one of the well known ones being a fleet of unmanned surface vessels, basically self driving robot boats for the Navy, using the same decision making software on the Mars Pathfinder rover, which is actually quite fun. And nowadays I own a growing video blog on YouTube called the AI Almanac. Before all of that, my love for data privacy and copyright. It's a theme that most people seem to take notice of in my AI blog. It began working at one of the largest music labels in the world, Sony Music in New York. I am actually a concert pianist, so that's kind of how I transitioned into that. So my path to the translation industry was long and winding. It came through a family member, my mother, who was also an LSP owner. But I am really happy to be here and I touch copyright privacy AI translation updates pretty much all the time now and have touched it all through my career. [00:03:49] Speaker B: That's fantastic. How did you then switch? I'm not sure if you switched, but how the whole story started playing out from you talking the language of music or the music language going into these other elements of your very interesting professional career. [00:04:08] Speaker A: Yep. I thought I was going to be a music copyright attorney, then learned that I cannot travel the world in America. You are stuck in one state for the rest of your life. So transitioned into software, which is why I got into the government and how I started in the language industry. I kind of argue I've been in it since birth because I have seen my mother run a company her entire life, and I've seen her what works for her and what makes what mistakes and everything but metalinguist actually came to me because of her. It started as a pro bono project for her and some of her colleagues, who were seeing this growing need for a client solution. Or I guess, overall this trend of neglected, or at best, completely fragmented client operations. So if language companies weren't using traditional email, which some of the largest entities in the world still consider to do for fielding, you know, order requests or quotes or interpreting, many companies had six, seven, eight different logins, so the client had to log in one for orders, one for quotes, one for invoices. One for interpreting, one for updates. It just got all really crazy. So me and my co founder, we built metal linguist as again, pro bono project and realized a lot of people needed it, so kind of took off from there. [00:05:25] Speaker B: And thank you for sharing that inside story. And started as a pro bono project. But going from a pro bono project, from any type of project into actually deciding to be an entrepreneur and a co founder is a huge leap. What motivated you to do that? [00:05:43] Speaker A: I think just the need we kind of got catapulted into it, to be honest. I mean, we were both working in the us government side, we were building this on our nights. And after like six, seven, eight months, the request for beta users just kind of exploded and continued to work on both for a while. And then I went full time into metalinguist once we actually did our full non beta launch as of earlier this year. So it's been really rewarding, really just fun kind of journey to watch unfold. [00:06:17] Speaker B: And thank you so much. Of course, I'm very curious about your knowledge about artificial intelligence, machine learning, machine translation, the whole data privacy conversation, and we could spend hours asking you questions about this. We thought there was a huge opportunity once we saw last week how Apple intelligence came about. And of course, before we get into that topic, I love to hear more about your process to get all this information, gather it, synthesize it, and then share it in a format that's appealing to people to actually listen to or read to. So why don't you tell us a little bit about your creative process and see if we can get some inspiration from it for ourselves. [00:07:03] Speaker A: So I started the AI almanac, actually because of myself. I'm a little selfish because I felt like I was falling, falling behind on my AI research every single week or every single day. I felt like there was something new and I just could not keep up. And only one month of neglecting what was going on. I felt like my information was ancient, so I created the AI almanac about three or four months ago as a way to stay accountable. Because I'm like, okay, Veronica, you need to siphon through the dozens and dozens or hundreds of AI articles, decide what are the top four or five updates, because really there's so much noise. I got to just figure out what the top four or five updates are and summarize it because odds are they're going to change in the next few weeks. Even the same entities change over the weeks. And a lot of the updates that I give are themes. For example, Apple Intelligence, when they announced their huge updates around Siri. Anyone that had been following AI Almanac knew where that came from because they published a paper about a month ago called the Realm model, which I talked about a month ago. So when Siri was announced with all of this crazy functionality and context and everything, I was already aware of it and my listeners were already aware of it because we talked about that. So it was really kind of just following trends and the way I go through the information. I set a timer for 2 hours and I tell myself I'm going to identify the top five, you know, trends, and I just start digging into them. But I think probably it takes me a full day of research to write the article and another half day to record. So it does take a good amount of time, but I feel on top of everything, which is good, and we. [00:08:46] Speaker B: Are grateful for your time and the fact that you take the time to do that. As a matter of fact, I use your newsletters as a repository of AI news so I know I can go back and then try to find some of the things that I've read. And then it just is just right there. And of course, it's a ton of information, tons of trends happening. There are tons of sources. What are your favorite places to stay up to date with artificial intelligence stories? [00:09:16] Speaker A: It changes every single week. So I couldn't tell you because sometimes Forbes breaks it, sometimes Washington Post breaks it, sometimes TechCrunch it kind of depends on the week. But I follow a few AI blogs, but overall I get them from raw news sites. I just find going straight to the sources is better for me. [00:09:38] Speaker B: Excellent. And of course, the story that brings us today, and the excuse to have a conversation today is Apple intelligence. And as I was telling you backstage, our production team and myself know very little about Apple intelligence. And what we only know about this is that they did an amazing marketing pr stunt by calling it Apple intelligence. That was very lucky of them, right? Amazon could do the same, but too late, I guess. Yes, yes. Maybe they can still pull it off and then turn it as a joke, make it a joke or something. I would actually do it myself. But what can you tell us about Apple intelligence? What's relevant about this story beyond the fact that it's Apple and a lot of people have iPhones and MacBook Airs and whatnot? What's relevant about this story? [00:10:33] Speaker A: Absolutely. So if anyone's not aware, Apple finally launched its own AI infrastructure called Apple Intelligence last week. And by first glance, a lot of people on the Internet were very quick to say, oh, my gosh, our data is about to be stolen. It's going to be on our devices. How do we control this? It's not even that special anyway. They're just generating images and helping us with text and stuff that we've kind of seen before. But what is so special about Apple intelligence? In my opinion, and anyone that knows me knows this is a really strong endorsement. I don't give lightly. It's by far the most advanced AI architecture for data privacy protection that exists for the everyday user. So the Veronica Hilux, just wanting to use my phone or my computer in a safe environment. Apple traditionally has had a very strict on device policy for data privacy protection. So keeping everything running locally on the device, but the full power of foundation models and LLMs and all the different reasons that we use them, they can't be harnessed on a single device. It's just impossible. The compute power is impossible, and the chips are getting more efficient, and Apple silicon is better, but they can't run a whole LLM on it on your phone. So it was an inevitable next step that they would have to go to the cloud. In the last few months, we were again on the bog speculating, are they going to abandon their on device first policy? And Apple Intelligence officially does leverage both on device and cloud computing models. So the way that they're protecting our data, which is just legitimately so incredible, they are leveraging this concept called confidential cloud compute, where basically every user, if Apple intelligence decides to either. So Apple intelligence basically decides to run your on device model, which you do have. There are some on device models that will just run on your phone that can do certain tasks, but if it determines that it needs more compute power, it will decide to go to the cloud. And leveraging this concept called confidential cloud compute, every single person has their own cloud compute instance to execute AI tasks. It's end to end encrypted. There's no permanent data storage, and it's not accessible to Apple or anyone. So it's really the extra mile in terms of data privacy to allow users to leverage AI in a way that is not almost muddying with other users, and it's safe and it's only there their cloud compute instance, and it's not being stored. It's just absolutely incredible what they've done. So I think my whole point, all of this is Apple intelligence on the surface looks like, okay, they're generating images. Siri's finally useful and we can do some things with writing now, but deeper is really this incredible AI architecture for data privacy and setting a whole new standard that just doesn't exist for us yet. Normally, we're the ants that nobody cares about, right? Our data gets used. You saw Adobe, they just changed their data privacy policy, and a lot of creators are freaking out that they're going to have access to all of their lives work. And that's typically what happens. But Apple went the opposite direction, which is really, I guess, encouraging for me. [00:13:55] Speaker B: And that's very interesting, because you, as a data privacy enthusiast, probably are rightly so, very excited. And you mentioned it on your newsletter, how you've recovered hope, and you're really excited about this one. And most of us, normal human beings, we only care about data privacy when we sit in the title of an article, and every other day our data is being taken because we want it to be taken or stolen, even if we don't want it to be taken. But we really don't really understand what's happening with our data or where is it going, or if it should actually be, if it's actually an issue, or if it's just kind of like a human gimmick to say, don't take my data from a personal perspective. I love them taking my data. If this is going to help you send me the things that I need faster, take it all, all of it. If he's going to suggest to me a pair of shoes I was thinking about, read my mind, please, and actually deduct it from my bank and have it on my doorstep in the next hour, I love that to happen. Read my mind. [00:14:59] Speaker A: Well, I'm the exact opposite. I'd be the one that will go on the farm in Scotland, some sort of cottage, and I would shut off all electricity Internet for the rest of my life if I could. But I think for me, it's control of being able to control where it's going. Right now, we're in this really bad precedent where these companies, Instagram, Facebook, their data privacy policies are absolutely horrifying. Exactly the same as adobe. They have the right to train models on your photos, on your posts, on your messages, on your interactions, like pretty much everything on Facebook. Even WhatsApp, to a degree, can be used and trained for models. And they have this kind of clause in there that says we have the unlimited right to license your content until you delete them from our systems. But the catch there is that once you feed something into an AI model, I don't personally know of a way to pull it back out. It's stuck in there forever. So for me, I think we're in a situation where we have to look at our data as an incredible asset and then we can control if we want somebody to recommend a pair of shoes for us and buy it with our credit card. But I don't like that they can just do it on our behalf, basically. So I take a very cautious stance. [00:16:19] Speaker B: Overall, of course, and without our knowledge, I guess, from my perspective. And this always happens whenever there is an interview or conversation about data privacy. Everyone at the very end, feels very naked and feels like they need to go to their phone and put it in. What is the name of the Faraday bag? Faraday. How do you pronounce it in English? The Faraday bag. Apparently, you just put it in there and it can tell where you're at in the world. But I'll send you the link before it's the one that Elon Musk was referencing on his tweet. I, of course, want to also think that despite the positive steps that Apple is taking, there are still some concerns. And beyond the more superficial concerns, is there any merit to the privacy concerns that some people have voiced over the Internet on the specific Apple Intelligence launch? That you can tell there are some. [00:17:27] Speaker A: Concerns that are valid, especially in reference to their partnership with OpenAI. So I was speaking to a friend who was the former CTO of Taos, and we both had the same conclusion. That OpenAI partnership announcement was honestly one of the smallest parts of Apple intelligence announcement. But upon further investigating, he told me that Tim Cook had said that if you're leveraging the OpenAI integration, like within your computer, it is not being held to the same Apple intelligence data privacy standards as if you were using Apple intelligence models. So that's something that is a really valid concern. I don't really see any sort of reason to leverage OpenAI within your Mac right now, because it could open up some data privacy concerns. Of course, as for the confidential compute infrastructure, you don't take everything I say with a grain of salt. Nothing's perfect. And they do have a program, I think, for hackers to try to find the vulnerabilities and actually reward them financially for that. So it's going through that process right now. But overall, I mean, there's not another infrastructure like this that exists for us. So I'm still really excited. At least I know that they're trying, which is incredible for me. [00:18:46] Speaker B: And I think you're on point with that. Can you tell us a little bit more about their translation API as well? [00:18:52] Speaker A: Yeah, so they did launch two new translation APIs at WWDC. But they're designed specifically for app developers looking to expand reach by integrating Apple's machine translation technology into their own applications. So these APIs use the same models as the Apple translate app and the iOS system wide translation, and therefore offering the same language support that Apple users are accustomed to. Now, of course, normal localization protocols are still highly encouraged, but it's really good to know overall when these types of APIs are launched. So they did provide access to that, which was awesome. But I think it's very similar to Google's API in that sense. [00:19:36] Speaker B: And you're going to forgive me, but I'm going to be very superficial with this next question. There are tons of AI tools, and I've been in the process of surveilling my kids and nieces and nephews because they're always downloading some sort of like chat GPT fake messaging system, telling them fake hallucinations all the time. And they're like, oh no, that's not true. Look what I found. And I'm like, you're referencing some system that it would be called like Sophie AI or Charlie Aihdenhe something random AI. But professionals, we tend to take these tools seriously and we use them. I use them on a daily basis, and some of us don't have the time to look into them, or we don't have the knowledge to actually look into the most useful AI tools. Could you tell us some of your favorite ones? And of course, you have some data privacy considerations, you have productivity considerations, you have sophistication considerations. So your input is going to be very valuable to all of us. Tell us a little bit more about those AI tools that are your favorites. [00:20:47] Speaker A: Well, I want to first answer it by the fact that most translators actually have been leveraging AI for a very long time because MT is kind of considered AI. So welcome to the crew. You actually have been leveraging AI for ten plus years, and NMT especially, but specifically generative AIh. For me, they're all very experimental. I experiment probably with ten or 15 different products a week, but the only one I use very, very, very consistently is chat GPT. I have it open almost all the time, and optimization in terms of the precision of the output is the most important thing. Otherwise, you're wasting my time. I've tried apps like motion AI. I don't know if you've seen that. And it's supposed to be an automatic AI scheduler. And I, at the end of the day, it caused more time for me than if I were to do it and run on apple notes to get all these tasks done so consistently. Only chat GPT to be honest, but I am experimenting constantly. So I think if I have any updates I'll let you guys know. [00:21:50] Speaker B: Thank you. And we'll talk a little bit more about where people can find your newsletter, your company and you as a professional. But now I really want to have you bring out your crystal ball and tell us about the future. Tell us everything that's going to happen. Elections and everything. I'm kidding. Could you tell us? Let's not go there yet, or at least in this conversation we'll find another podcast to talk about those things. Veronica, could you tell us a little bit more about the future of the industry? And it's great because you've seen your mother work with an LSPDE, be an entrepreneur as well. Now you are an entrepreneur, you are in the industry, and you are the very cutting edge. So you definitely understand how to harness the power of these tech tools. Others are very afraid of doing that and not only doing it. Some of them perhaps are not at the right season in their professional careers. Maybe they are ending their professional careers. If you have five to ten years to go, maybe you have no motivation to jump onto any new technologies and you just want to get it done right. How do you see the future shaping up for the industry? And what are some of the things that you think are completely inevitable for us? [00:23:05] Speaker A: Well, I don't want to spend this time saying that. I hear all the time that you're going to be competing with translators that leverage AI. And while that is true, I want to bring up the fact that translators traditionally, no matter how long you've been in this, in this profession, 510, 1520 years, we've gone through multiple technical evolutions. I mean, computer assisted translation tools were traditionally only on the, you know, on the computer. Then they moved to cloud, then we had tms, then we had NMT, then we had or MT, then NMT, and now we have generative AI. And generative AI is a whole other ballpark for a lot of different reasons. And I'd love to talk about the technical parts of that at another time. But it is sad to me when I see a lot of linguists be scared or resistant to start to experiment with AI technology because there are a lot of tools out there, especially from a data privacy perspective, that are protecting your data. And that's what I hear all the time. I'm a translator. I'm scared to use a tool like BWX because I don't know what they're doing with my data. Well, I can tell you firsthand that Gabriel told me they have a lot of data privacy protocols in place to pass the most minimal amount of data into any sort of LLM, and they're protecting it. And they have retrieval augmented generation, and they have semantic index search, and most companies are doing that. I know Meloq is doing the same thing. I know that RWS has a version. So ultimately, don't be scared. From a data privacy perspective, I really think that most of the companies in the translation industry are really focused on it. I also will see. I think there's going to be an increased awareness of the value of translation assets already in the possession of most all language companies and freelancers. So we already know that our AI output is extremely dependent on how clean your translation memories and your glossaries are, but less spoken about. Those assets also need to be protected, extremely valuable. They're special to use. So I'm seeing a huge increase in questions by language. Company is asking exactly what you're asking. Hey Veronica, what does data privacy look like for us? Are we violating NDAs? How do I ensure that these models don't fit on my data? Really all valid questions that a lot of people don't know, but thankfully are asking a lot recently. So I'm happy to hear that, and. [00:25:30] Speaker B: I'm happy to hear your insights on this. Do you think we'll ever get any money from the companies using our data for their training purposes? [00:25:40] Speaker A: So I'm going to tell you, I'm going to answer that with a response. I got a Paypal notification that betterhelp.com comma, I might have used it seven years ago. It's like an online counseling app. Miss sold my data to Facebook. They had a huge lawsuit, and I received a dollar nine payout for the misuse of my. [00:26:01] Speaker B: That's great. [00:26:03] Speaker A: I can maybe buy a latte in Washington, DC with it. Nothing, unfortunately, nothing that I feel will be worth it. But actually this week I'm going to be recording it right after I get off this call with you is the EU actually stopped meta from training on Facebook user data because of a lot of restrictions that you put up. So that was a huge announcement that came out this week. I do think that the resistance that people are pushing is effective and putting those kind of blocks up for a huge company like meta in the entire EU is shows us it's not a lost cause. [00:26:45] Speaker B: Great. So I'll stop Elon Musk from jumping on my brain to read my mind, to buy my stuff as I think about them. I'll hold on to that idea for a little while. This is a perfect segue. Veronica, could you tell us more about where we can find your newsletter, how we can sign up for it and follow these amazing trends and insights that you're putting out every week? [00:27:10] Speaker A: Yeah, so if you want to add me on LinkedIn, I'm pretty findable. But also if you search YouTube, the AI Almanac, or go to aialmanac dot substack.com, that's the written version of my blog and you can find everything there. And as Eddie mentioned, I do post weekly. So this week's will out in a few days. And every Thursday or Friday they come out. [00:27:32] Speaker B: Mila, you're going to have to edit this little rascal that just jumped in my office. That's Matteo. No AI would ever do that. [00:27:42] Speaker A: We should keep him. We should keep him. [00:27:44] Speaker B: Let's keep him then. No AI will ever jump on your goal like that. Thank you so much. So, the AI almanac, we will put, of course, all of the links on the descriptions of these recording when we put it on the newsletter as well, we'll make sure that this is perfectly linked. And Veronica, we're going to take you up on the idea of having conversations in the future whenever a major announcement comes up. And by the end of the year, I also have an idea. You will see Veronica on our magazine, physical magazine and digital version as well. But more on that later because I will let Veronica know after this recording. And that is all for today. Veronica, is there anything you'd like to tell our audience and your audience, which I assume would also, of course, listen. [00:28:30] Speaker A: To some of this learning is exhausting for all of us, but it can also be really exciting. So this is really an exciting time. And I'm fatigued. Everyone's fatigued. Don't look at AI as another thing to learn. Look at it as another thing to, I guess, explore. So it's, you know, that's. I guess that's it for me. [00:28:56] Speaker B: That was Veronica Heilac, co founder and co CEO at Metalinguist. I'm Eddie Arrieta, CEO of multilingual magazine. And this was localization today. Thank you so much for listening.

Other Episodes

Episode 27

February 09, 2022 00:03:18
Episode Cover

The Economist shares data on English's role in global pop culture

Data journalists at The Economist used information from Spotify to identify several trends in how local communities across the world listen to music —...

Listen

Episode 292

January 01, 2023 00:06:28
Episode Cover

LangOps: The Vision and The Reality | December 2022

In this piece, Nimdzi Insights co-founder Renato Beninatto dissects the concept of LangOps, identifying a handful of challenges he foresees for the term as...

Listen

Episode 25

February 15, 2023 00:04:21
Episode Cover

Oregon senate bill generates debate among interpreters, LSPs, labor organizers

A recently introduced senate bill in Oregon is drumming up a bit of controversy among healthcare interpreters and language service providers (LSPs) that work...

Listen