النص 🔗
Lam Thuy Vo: API access to all of the social media accounts. Craig and I, at some point we're able to get four million posts from Facebook groups using the API. I wrote a script. I walked away overnight and we had four million posts. Give us back our API access. For free.
Craig Silverman: Yeah. Hi folks, welcome to Show and Tell, the podcast from Indicator that takes you inside digital investigations by leading journalists, OSINT investigators, threat analysts, and researchers. I'm Craig Silverman.
Alexios Mantzarlis: And I am Alexis Mansalis. Indicator is a resource for professionals and citizens who want to investigate and understand digital deceptions. We have reporting, tools, training. If you've never heard of us before and you're here only because of our very, very, very special guests, please make sure to check out our investigations and workshops at indicator.media.
Craig Silverman: And we'd like to thank our founding sponsor, Lead Stories. Lead Stories is a global fact-checking organization with journalists working in a dozen languages, doing great work. They combine old-fashioned journalism skills with technology to find and debunk false claims on social media. Lead Stories believes fact-checking is about adding information, not censorship. And you can sign up for free weekly updates from them to learn about the latest viral hoaxes and misinformation. at leadstories dot com slash newsletter.
Alexios Mantzarlis: Today we are joined by two reporters from Documented, an independent nonprofit newsroom dedicated to reporting with and for immigrant communities in New York City, a city I love and miss very much. Lam Tuivot is an investigative reporter with Documented and an associate professor of data journalism at the Newmark Graduate School of Journalism at CUNY. Sam Rabia is an investigative technologies fellow at Documented, as well as a contributing editor at the city and an adjunct professor at NYU, where he teaches cartography, design, and urban studies. Sam and Lam recently published together with some colleagues at Documented an investigative series called Decoding Immigrant TikTok. It's made up of two stories, one focusing on deep-faked immigration lawyers and the other on What they call news fluencers pushing scaremongering stories, that we'll talk about in a moment. And perhaps most exciting for indicator followers, subscribers, readers, was that y'all published also the tools that you use to analyze your big TikTok data set, which is something we will chat about. So welcome to show and tell both.
Sam Rabiyah: Thanks so much for having us, guys.
Lam Thuy Vo: Thanks for having us.
Craig Silverman: Amazing. So thank you and welcome. And I should mention that Lav and I worked together at BuzzFeed News years ago. So it's wonderful to be
Lam Thuy Vo: Ha ha
Craig Silverman: reunited here, having a chance to chat. So welcome to both of you. Why don't we start with the data and then get into sort of the findings that came from that after?
Load full transcript
So for both articles, you were given access to what you called a trove of TikTok data from NYU's Center for Social Media, AI, and Politics. can you talk a little bit about the data set, what you know about sort of how it was collected and why did you choose TikTok to be the focus?
Sam Rabiyah: Absolutely, I can kick us off with that. I will just say, just setting the stage, I came to this project in the wake of legacies like Lon's of reporters who have been on the ground examining TikTok and its use and misuse within immigrant communities. So, in some sense, that really was the point of entry for this series of stories. And to combine that with this particular offer we got from the these folks at NYU. Particularly this group called the Center for Social Media AI and Politics, they essentially have been working for a couple of years trying to create a representative estimate of the entire TikTok universe. And I'm not particularly positioned or at liberty to describe the inner workings of their methodology, although it was something we were able to kind of review. But essentially, they have gotten pretty close to finding a way to catalog essentially every TikTok link since about two years ago up until the current day. And they have like sort of they're maintaining an ongoing database of all of this information and all the metadata attached to each of these links. And it was through that partnership that they were basically like, hey, we have this, we think it's probably something that could lead to a lot of really impactful investigations. We want to kind of put it in the right hands. Do you guys want to have access to this in certain particular ways? Could this be fodder for particular stories for you? And we were like, yeah.
Alexios Mantzarlis: And you were like, no, no, thank you. And that's the pass.
Sam Rabiyah: Yeah, exactly. You're like, no, that's that doesn't really interest me.
Lam Thuy Vo: Ha ha.
Sam Rabiyah: you know, and and just to be totally clear with folks, there was a bit of a back and forth in terms of access, just because they weren't at a position where they could give us full query access to this database of tens of millions of TikTok video metadata. So we had to kind of negotiate particular queries and really rely on the technical staff members at that center to kind of respond to our emails and answer our Deepest wishes in terms of the particular subsets of data that we wanted from this data set. And so we crafted a very particular set of queries, mainly through SQL, but you know, they kind of had their own special version of a query language that was used particularly for this data set, to essentially create a representative sample for our own analysis, which is how we arrived at this. I like to call it the immigrant TikTok leaderboard because what we ultimately landed on,
Alexios Mantzarlis: Yeah.
Sam Rabiyah: and the the intention here is like we want to see what the immigrant communities in New York City are actually watching every month for the last couple years. what are the most viral videos that are catching people's attention? So ultimately we created this kind of keyword targeting system that was that was pretty simplistic, honestly, that tried to rank up, okay, what are videos that are specific. to issues involving immigration and are specific to New York City, and let's just grab the 500 most viewed videos every single month for as long as we can. that was kind of the playground that we started with to then start playing around and see, okay, what's actually going?
Alexios Mantzarlis: So let's kind of dig into one of the stories, the one that I was immediately drawn to. which was about the the sixty or so TikTok accounts. You found more, but I think you mentioned sixty in particular, impersonating New York based immigration attorneys. and the scam was asking for an upfront fee for a legal consultation, but they obv obviously never provided and it was particularly galling in your story to see one of the scam accounts using an AI deepfake of Mohammed Salim who was Legitimate lawyer based in Brooklyn, quote unquote warning users to avoid scams and to contact him only via WhatsApp, which is of course a classic scheme So besides the pure evil of this setup, which is a scammer warning about scams, I was curious, because you say that Selim had been trying to get these accounts shut down for years, and that even after your story, TikTok did take down some but not all. did get a sense about where actually TikTok draws the line on scams and on harmful impersonation.
Sam Rabiyah: Absolutely. I think what this story is really pointing out is just the ways that TikToks Reinvention of all of these social media algorithms that we take for granted at this point has created a very particular environment for scams to thrive and scams to take advantage of people at their most vulnerable. And that also this moment we're in where AI is kind of taking over everything in these digital public spaces. How is that moment also exacerbating this issue? the example that you pointed out, Alexios, was pretty shocking, right? Where there was someone intentionally trying to create deep fakes of an attorney using AI to steer people away from their direct communications to steal money from them. There's also dozens of accounts that are simply just reposting old content to gain a following, to gain credibility, waiting for their next move once they kind of reach a certain level of authenticity or perceived authenticity. if you're someone who's desperate for l a legal consult, you're not necessarily gonna be checking for these things. And at the end of the day, TikTok is protected by all of this federal legislation that basically means they're not liable for the third party content on the platform. and I think that's telling and how we kind of played out the story, right? We're hearing from Mohammed Salim, we're hearing from other users that they are. Constantly reporting these scam
accounts for years. And TikTok ignored that. And it was not until we sent a formal press inquiry that one of the main accounts we discussed was shut down. And I will say I did check in the wake of our publication of the series all of the different accounts we're featuring. And I've actually noticed that the majority of them have since been deactivated since the story came out. There's only a handful
Alexios Mantzarlis: Hm, that's great.
Sam Rabiyah: that TikTok says was were specifically deactivated in response to us, but I think the timing there is also pretty notable. But at the end of the day, TikTok, can do whatever they want here because they they're protected from any kind of liability.
Lam Thuy Vo: and just just to zoom out for a second, all of this videos that we're talking about are largely targeting Spanish speakers. And so What we're seeing are a lot of these kinds of accounts that actually target niche audiences, so to speak, in a large place like the US. the Vietnamese speaking population, for example, is fairly small in the US, but you can get a lot more out of each person. Because a lot of the people who are immigrants in the US don't have alternative news sources in their preferred language. So for them, getting information on TikTok or on YouTube is a lifeline rather than an additional in piece of information that they can get while watching, the news in English and while reading a newspaper or in in their own language. And so they can thrive with impunity because there all the mechanisms that are there to fact check to find scammers and so on are trained in English and not in languages like Wolof. or Vietnamese or even Spanish So I think knowing that there are many micro communities that they can target that are very hyper dependent on this news ecosystem also makes the payoff for scamming someone in those communities much higher. Because no one else is there to fact check them.
Craig Silverman: Yeah. it's it's a really dangerous combination when you think about it because the combination of the language and then they don't actually have to become really big accounts because they're running a scam where if you impersonate somebody who you know charges a fee and you know, and and they could go and if they do a little due diligence, they're like, Well, this is a real lawyer. it's a really, really dangerous kind of combination. you you mentioned the kind of news ecosystem and how these accounts on TikTok for some of these communities are very important news and information providers. and so that's kind of the other story we wanted to talk about, where you know, looked at news influencers on TikTok pushing inflammatory anti-immigration content. you ended up reviewing about 10,000 videos. so this is like you got it, you had access to the big data set. you shrunk down to a still pretty good size data set of ten thousand videos. So how did you how did you get to that that place of defining that sample, and then, you know, deciding which of these influencers to look at and obviously to identify the content.
Lam Thuy Vo: I think this actually dates back to our work together, Craig. I think one of the things that I learned from working with you back in 2016. my goodness. is to really go away from the notion that the internet is the world and the internet represents at large like what people in general look at, and instead to define a small ecosystem that is particular to a user persona. so in this instance, I think one of the things that we wanted to better understand is what does a Spanish speaking immigrant in New York City consume when it comes to news around immigration, right? And news in Spanish. And so with that in mind, that's how we already narrowed down the ecosystem that we wanted to look at using hashtags and using search queries. And then from there on, I Sam and I were very interested in who are the biggest players in this ecosystem. And we started just doing a very simple aggregation analysis of looking into the most popular accounts. And they were news influencers and those lawyers. And that's why the two stories that we did were representatives of those two branches. And so from there on, we wanted to see who is the most exemplary, sort of like emblematic character of that. And we found this one particular news influencer that we then looked into a little bit more deeply. And we did some quantitative analyses, but then also some literally like reading through transcripts of these videos and watching some of these videos to get a better feel for that. I think this approach of defining editorially what you're trying to look at. That's the art of the part. And then using some quantitative methods to narrow it down, to do some natural language processing and so on. That's sort of the quantitative methods. The two go hand in hand and the editorial approach combined with the technical approach allows you to take a look at both like the bird's eye view and also the intimate understanding and feeling and experience of that kind of of those kinds of videos. And so
Craig Silverman: Yeah, you obviously wanna watch them and get a feel of what's going on there. You can't just pull the data and stick with that, right?
Lam Thuy Vo: Right. And so there's this fascinating thing where you get these sixty seconds clips of like there's a man who has sexually harassed another man on the subway. It's a video of this guy just walking towards the camera of the subway and then he says, Police has arrested him. This is news, done. Right? And so these kind of small incendiary, inflammatory sort of like portrayals of immigrants. By themselves are factual and stem from a kind of news, but their selection and their repetitiveness makes it into something that almost like pushes you to believe that immigrants are all, especially newer immigrants, are all criminals. And so understanding both the texture of these videos, the ways in which you feel when you watch them, in quick succession,
each one being 30 s to sixty seconds long, right? that combined with understanding the scale of things, the quantitative way of understanding that, that can give you both an understanding of how intimately it feels, how it feels for the person watching it, but also how they get inundated on a larger scale with that.
Craig Silverman: And and the count you were talking about as well. you would do a count level analysis, you would see his best performing stuff is the over the top anti immigrant violent kind of stuff. And so, it becomes a self reinforcing thing where he's like, that video did really well. I gotta do another one like that and it just starts this perpetuating cycle.
Lam Thuy Vo: Right. And now you add to that all of these AI generated generated videos, these accounts that are basically content farms that go out and take the same approach, see a crime. Ooh, I can take this particular crime, take a photo, make a bunch of AI generated photos, have it be stitched together by an automated editing software, and then put it out there, right? And that then takes a human made narrative strain. That is fed through and then amplified through an algorithm and then amplifies it even more by having a bunch of copycats bring about the same tone, bring about the same type of content to the same feeds. And DDIA is an institution that cares a lot about how Latino viewers understand information online. Roberta Braga, who's the ED of that institution, executive director of that
Craig Silverman: We love them. They're great people over there.
Lam Thuy Vo: She though she said something really helpful. She says that we know as humans when we watch AI generated content that it's not real and that we should be suspicious of it. But then a lot of immigrants lack better sources. We don't have as many Spanish speaking sources that give you a counter argument. And on top of that, now you're being inundated with that bad information. And there's this thing that happens where you might accidentally adopt beliefs that come to you at such a high level, even if you know you're suspicious of the individual videos.
Alexios Mantzarlis: Yeah, we've covered operations that deep fake Spanish speaking anchors the scale of the Spanish language disinformation content farms on TikTok is huge. as you sh as usual, of course. All right, you know, y'all put both a guide on the website but also published your full code on Git the evidence that underpinned your reporting.
Sam Rabiyah: Yeah, so I will say the this whole project really was a perfect example of how this kind of OSINT work interacts with like regular on-the-ground reporting. Just because I think a lot of our ambitions with the technical side of this project. Came directly from some observation we had, or a series of observations we had through the whole history and process of actually reporting this out. I think the the main thing here was. Just like actually just with our human eyes noticing the levels in which videos could hop around between disparate accounts without any way to really follow where it came from and the order in which it's being reproduced. And I think there was something also really specific about the TikTok algorithm that allowed this to happen, right? You know, I think what's really exciting and a big part of TikTok's, you know. Business model and marketability is the fact that you can be anyone and just post a video and go viral that day. Right. That's super different from how Facebook worked, where you had to spend years cultivating the social network to gain credibility on the platform. And what we noticed is that in the sample, there would be particular videos and we see it pop up in a dozen other accounts. And we couldn't really figure out who the source was. And I think one of the most notable things is we
Alexios Mantzarlis: And you mean exactly the same video or clips of the same video? Yeah.
Sam Rabiyah: Exactly, exactly. It would it would vary with different watermarks or different subclips of the same clip, but it was generally the same exact video. And I think the perfect example of just how intense this phenomenon is is the fact that we did a kind of an initial pivot table, like initial ranking of which accounts got the most videos into our sample, and one of Mohammed Salim's scam accounts ranked higher than him. They got more videos in this sample of most watched videos than his official account. So we're like, okay, there's clearly something going on here. And that's when we built this particular tool to help us out. And so what we realized we needed to do is if we wanted to match videos that essentially were ripping each other off, we had to get our hands on the actual video files themselves. At this point we only had Particular pieces of metadata from the NYU database. And so we built first this scraper, which essentially downloads all of the TikTok videos and records additional metadata alongside the download, saves the video files in some kind of like extracted cache.
Craig Silverman: one thing I I don't think we established was what kind of metadata was in the original NYU data set that you got? 'Cause obviously you didn't have the videos themselves, but
Sam Rabiyah: Right.
Craig Silverman: you had the metadata.
Sam Rabiyah: Yeah, and it would actually their the richness of their metadata was pretty extensive. I there was probably about 30 different fields in there. The ones that were most helpful to us at the time were obviously the TikTok URL, the author profile, like the author's handle, view count, share count, stuff like that. and we were mainly using view count as our metric for virality, right? That's what we're using to rank the top 500 every month. so it was pretty extensive. So I will say just to add context, this step of our script was more as a verification and an update to the kind of metadata that we wanted rather than adding something we didn't have. I I wouldn't be surprised.
Alexios Mantzarlis: But also
for people following, this is the first step for them 'cause they won't have the NYU stuff, so this is great.
Sam Rabiyah: Exactly. And I that's a that's a good point. That was another consideration we had in mind with this. so once we had the videos downloaded, we had to kind of transcribe the audio, and that's why we used faster whisper, which particularly allowed transcription across different languages. I would say roughly 40% of the videos in our sample were Spanish language videos. fifty percent were English and then the other ten percent roughly were a variety of other languages. So it was really important that we had some kind of translation system that could work with all those languages and also could be swapped out for you know, like could be swapped out in case there's some issues with that. And that kind of course.
Alexios Mantzarlis: Do you remember sorry to jump in and again a ballpark of like what this cost, the the whisper part of it? Do you have any sense? whisper is free. nice.
Sam Rabiyah: So this is a fully open source library. N none of none of the technical infrastructure that we used or built cost us any money. It was all open source tech, and I think that was
Alexios Mantzarlis: wow. I didn't realize the whisper was all open up.
Sam Rabiyah: yes.
Lam Thuy Vo: I will say Whisper is an open AI library. So just so in
Sam Rabiyah: Yes, it is.
Lam Thuy Vo: case you it is one of the few ones that they actually open source, so that was really fascinating. It does perform very differently in different languages. So I've tried it out in Vietnamese
Sam Rabiyah: Yeah.
Lam Thuy Vo: before, and a lot of translators hated it. I think that's generally
Sam Rabiyah: Ha ha ha.
Lam Thuy Vo: like an issue with generative AI. and I think Whisper uses Transformers, so I think it would qualify as generative to some degree. But like it's a small enough model that you can download and use on your code. a local machine.
Sam Rabiyah: Yeah.
Lam Thuy Vo: But yeah.
Sam Rabiyah: I think this is an important moment to just emphasize that, you know, there's a very particular intention in creating this tool to use one, sources that were free and two, sources that you could run on a local system, right? So there's no reliance even on like the free tier version of these proprietary data center reliant, you know, technologies like ChatGPT or stuff like that. and I think that was something that was very important for both of us to maintain when producing this. And so just to wrap us up here, once we had everything transcribed, that's when we would translate them all into English, one, to allow us to actually read through these transcripts if we don't speak all of the different languages in our sample, but also to create make this sort of text-based clustering algorithm work a little bit better. And we used this library called Deep Translator, and particularly like we started just with Google Translate from varia a whole set of different translation services that they use. And this is also important because and and we note this in our in our documentation. Google Translate is one of these volatile technologies. Sometimes you get rate limited, sometimes they exclude your IP address from making any more requests. So what's great about this library is that you can swap out a variety of different free translation services into the systems in case you kind of run into one of those roadblocks.
Craig Silverman: Nice.
Sam Rabiyah: And so and then at this point you have an English script of all videos in whatever sample you're working with, and that's when we're able to run this clustering algorithm. And you
Craig Silverman: Yeah, and that's like we'd love to hear more about that 'cause you sort of had a couple of approaches. One you called containment, one you called semantic. so we'd love to see like, you know, what videos sort of got kicked up through this, what it looked like as you're reviewing the data, 'cause it seemed like this is the part where it sort of showed you some that sort of became more interesting to you, I suppose, right?
Sam Rabiyah: Definitely, yes. And just to kind of like define our terms, we have these two different configuration knobs, is we're calling them, that sort of describe to the technology how to group things together, how to kind of measure that two scripts match when others don't. And one has to do with okay, is this exact line of text found as a subset of this other line of text? And that's the containment similarity. And then semantic similarity is more like, are they explaining the same things, right? does does it convey the same meaning even if words differ slightly? And in our particular use case, this was honestly more of an effort to account for the variability of the transcription and translation services. Because sometimes we'd find, you know, they would hear things a little bit differently. You know, these these technologies are imprecise sometimes. So if there are slight variations in language. solely dependent on the translation, but it was ultimately the exact same video. We wanted to still link those together. And so that's why we played around with these two ways. I think what's interesting about this model is it opens up a whole universe of other use cases, right? You can really boost, you could you can really change the levels here to make it much more liberal in its grouping algorithm. And maybe you want to just group together videos that are generally discussing the same topics, even if they're totally different videos. But for our particular use case, we were really trying to find those copycat accounts, particularly the ones that were copying the immigration attorneys. And so that was the sweet spot we were trying to get here.
Alexios Mantzarlis: you mentioned somewhere, I think it's in the piece, that scraping TikTok can be quote unquote temperamental. I'm a big believer in running my own burner accounts, but I was interested in what you were saying
about the getting people to donate their data because it seemed to be pretty easy via the TikTok interface. So I wonder if you could both show on TikTok how people can do but also give us some advice on your end on like what is the right way to ask people for this data, how do you collect in a way that is ethically appropriate, responsible, etcetera,
Sam Rabiyah: Absolutely. I'm really glad you brought this up because I think that one of the one of the things we realized producing this was, if we're just here to show our work, we had access to something that not everyone is going to have access to. And so this is going to fall short of allowing people to recreate this for whatever project that they're working on. and so I'm just gonna borrow from our technical guide, which was the third piece in the series and was ultimately a write-up explaining how to kind of create your own sample in service of using this tool to analyze and find these different points of connection. And, you know, we offer a couple different options here, right? And this is very much based in the work Lom does, a lot of which is Community building through data sharing and education around these digital spaces. And I, you know, just to get into the technicality of it, TikTok does allow you to request. your user data in full. And we include a screenshot of this and also link to a particular guide that TikTok offers to show you how to do this. But I've done this with my own data. I know Mom has facilitated this kind of data access request with a lot of other folks that she works with. And they make it pretty easy. Unfortunately I have to wait 12 to 24 hours for them to actually send it to you. But it essentially offers you a machine readable data set of your entire view history and you can even add on a bunch of other parameters here.
Lam Thuy Vo: I call this the quantified selfie, figuring out how you can
Alexios Mantzarlis: Yeah.
Lam Thuy Vo: prove harm through one person. You can prove how the algorithm works through one person. It is not going to prove harm on a systemic level, but it is good enough to showcase how it impacts individuals. back in 2017, when I was working with Craig and Buzzfeed, for example. We asked a conservative mother and her liberal daughter to share their Facebook algorithms and they allowed us to look at their feeds and we were able to show them side by side what kind of content was driving them apart even though they loved each other. We did a similar project with the Vietnamese community to better understand the media ecosystem that people inhabited and downloaded a YouTube archive. I have a scraper for that that I can also share with you guys. But basically, Figuring out how people are experiencing the web is also a way for them to better understand the algorithm. And a lot of the work that I've done for the past decade in mis and disinformation has pivoted from just pointing out the problem to figuring out what the problem looks like through this
data and then coming back with guides and resources that empower people to protect themselves in this moment. Because one of the things that's really problematic with mis and disinformation reporting or media ecosystem reporting for a long time that people have pointed out to me over and over again is that it leaves people with an understanding of the problem but with no solutions. And I think in a a moment of news avoidance, in a moment of people capitulating, due to doom scrolling, I think figuring out how to bring back this engagement from the community can be helpful for them to then say, this is my problem, let me figure out how to solve it for myself, of my community as well. and you mentioned a really important part, the ethical aspect of this, right? what we're talking about is a very vulnerable. group of people who are already targeted for scams, targeted for all kinds of horrible exploitative campaigns including like human trafficking, smuggling on these platforms and to come to them and say, hey, I'm gonna get your data, is not the best way of doing that. I think framing it as a I will help you diagnose what you're looking at is a community-based collaborative process. That then helps people understand what do I get out of talking to you and giving you this very personal information. And then the other thing, just practical things that you want to do is you wanna have a data retention policy that you keep to yourself, like figure out how long you keep the data, where you keep the data, protect the data. You basically have to be very proactive in this because most of the people who you work with won't know. what it entails for you to have the data. Then what I also tend to do is I write a script
Alexios Mantzarlis: And sorry, Lam, do you have on that, do you have like advice for people? Like would you say kind of sixty days post publication and then you just keep kind of aggregates or like do you have any like the general lines that you suggest
Lam Thuy Vo: I would say whatever data retention you have at the institution is ideally what you also keep with
Alexios Mantzarlis: Mm.
Lam Thuy Vo: that. That's also just from a legal stance, probably
Alexios Mantzarlis: Mm-hmm.
Lam Thuy Vo: the better way of doing this in journalism. You want to keep to a particular time frame. Could be three months, could be six months, but keep it, set that t timeline and then keep it consistent, right? What I sometimes also do is as you mentioned, I pre-write the script, I do my analyses on my own data set. Then once I get this, I do the aggregates and then when I'm done and read the I keep the aggregates, I delete the original data. That way I can discuss the outcomes with things without ever having to touch, read, or n acknowledge private messages on like some platforms. and like the reason why I mentioned this is like TikTok only gives you a archive of your links and what when you watched it. some messaging. But like If you work with someone to give you your Facebook archive and I did that in twenty thirteen with some people who were organizing during the umbrella revolution in Hong Kong, they give you their entire messaging archive. They give you all kinds of personal information, how many times they've blocked people, who what they've searched for online and so on. And so you gotta really be careful about like how you handle that. So pre-write the script, use yourself as a guinea pig. run the code, discard the information. And especially when you're looking at people who have a vulnerable immigration status or might be subject to other kinds of like political violence or domestic violence, these kinds of things, you might even pull up and do a risk assessment for what it would mean if this data got leaked. And so I would even for some of the particularly vulnerable ones, talk to my editor and say, Hey, for this particular group, let's delete it straight away and just keep the aggregates. And so having your data retention policy pre writing a script so that you have it done, you just need to run it once and you're done with that. And doing it with continuous checking in for consent with people is very important. Are you still okay with me publishing this? Here's what we're doing. And I'm always on the side of If someone says no and I have to sacrifice a story, I would rather do that than just bear someone's like personal life to the world, for personal gain, so to speak, for a story. Yeah.
Craig Silverman: Yeah. I think those are great, great points to kinda end that that piece on about making people feel empowered, giving them some some tools and knowledge that they don't feel just overwhelmed and and powerless in this. And then the ethics of how you interact with people and their data. and you know, not keeping data you don't need. Because stuff that invades people's privacy can be useful for OSIN. And if you are practicing this ethically, you should be minimizing your data trail and minimizing the data trail of the people that you interact with and and the data you collect from them. And we should note that this stuff is up on GitHub. You guys have very generously made it available. We will obviously put that in the show notes for people so folks will have the ability to go and and run this, you know, this great kind of pipeline of TikTok stuff for themselves. So now that we've done some show and some tell, we want to sort of finish off with some rapid fire questions for the two of you. So this first one and Sam, why don't you respond first? You get to keep only one tool that you can use for your research. what is it? Sam and then Lam, please.
Sam Rabiyah: man, I think it would be the Wayback Machine. Or just generally Internet Archive. I think a lot of OSENT
Craig Silverman: Right.
Sam Rabiyah: tools rely on it in kind of a less transparent kind of way. And I I just think it's like a I think it's also shifted our perception of what digital ephemera is and how we really need to hold on to
it. And so I would keep that.
Craig Silverman: Mm. Yeah.
Lam Thuy Vo: the library YT DLP, which allows you to download
Craig Silverman: Mm.
Lam Thuy Vo: and get metadata from so many videos around there. It's you can use it in your command line, you can also write it into script, it's very, very versatile. Has been upkept as well for many years. it's it like many of the libraries I've used have deprecated, but this one is like, I'm here.
Craig Silverman: Yeah, it's amazing also because it works on so many different platforms for grabbing videos. I personally use the it has a really nice UI. It's called Stasher, S T A C H E R, which if you are not super, you know, code friendly, command line friendly, Stasher is just you just pop the URL in, boom, the video is downloaded to to your system. So excellent choices. So Lam, going back to you, are you using any coding assistance? Like cloud code, codecs. If you are, why? And then Sam interested to hear if these are in your your workflow at all at this point.
Lam Thuy Vo: I use Claude only for debugging and I go back and forth on this all the time. I just came back from MozFest and I've been thinking about because first of all, Cloud is trained on mid code. So what you're gonna get is mid code, right? Like in journalism,
Alexios Mantzarlis: Ha ha ha.
Lam Thuy Vo: mid code is just good enough to do what you need to do. But I'm wondering whether I can just use LM studio at some point and like run my local L L there 'cause like I'm I'm assuming most things won't change drastically. So yeah. Sam, what are you using?
Sam Rabiyah: Yeah, I feel like I'm in a similar boat to you, Lom. I try and avoid them as much as possible for two reasons. I think one, it's the way you don't wanna pay for Apple storage. I feel like they are designed to get you reliant on this digital service and then keep upping the ante. And so I never want to pay for any service like this. I think sometimes when I'm like coding or like doing web development work, there'll be this code autocomplete function that just comes built in with a tool like Visual Studio Code. I I know that that relies on some form of Claude and sometimes I do make use of that, but I I o you know, I think there's a lot to be concerned about of becoming over reliant on them.
Craig Silverman: Sam, are you saying you don't pay for storage with iCloud? Wow. Okay.
Sam Rabiyah: no, I pay for iCloud storage and that's why I can't afford cloud code.
Alexios Mantzarlis: He he's out there deleting every photo. Nephe, nieces, doesn't matter. They're all out.
Craig Silverman: Yeah, that would have been incredible digital hygiene.
Sam Rabiyah: No, that's I can't I can't have two I can't have two subscriptions. No, I just can't do it.
Craig Silverman: Okay. so what about a favorite search operator? do you have one that that is your go to for for you folks?
Sam Rabiyah: Yeah, so I love site like site colon. I think that's really helpful with finding links. I I this is not so OSINT related, but I use the file type Google search operator a lot, particularly when I'm
Craig Silverman: Uh-huh.
Sam Rabiyah: designing something. I need someone's like, our logo doesn't exist with a transparent background, and then I'll be like, No, actually it does, and I I found it for you.
Alexios Mantzarlis: Yeah.
Sam Rabiyah: It's like digital sleuthing for a very different purpose, but I I end up using that a lot.
Craig Silverman: I s lamb
Lam Thuy Vo: Google Lens. Just like oftentimes I just want to see the origin of images. I it's not as helpful sometimes with like AI deepfakes, but Google Lens is really helpful. Very simple.
Craig Silverman: if you could bring back a dead tool, what would it be, Lom?
Lam Thuy Vo: what is dead? I'm like a lot of things I did. boy. Honestly, all of the news articles that have died, I wish I could find them and archive them somewhere. I don't know. All of our clips, all of this, all of this knowledge is just gone. It's so sad. no. Remember actually, man, API access to all of the social media accounts.
Alexios Mantzarlis: Yeah.
Lam Thuy Vo: Craig and I, just as a small aside, at some point we're able to get four million posts from Facebook groups using the API. I wrote a script. I walked away overnight and we had four million posts. Give us back our API access. For free.
Craig Silverman: Yeah. Yeah. good luck with that, but a great choice. Sam, what about you?
Sam Rabiyah: there was a time where different government agency websites were less technically savvy when it came to scraping and web security and you could scrape their whole database very easily without running into different security blocks. obviously it's a double edged start because I'm generally happy that the world is becoming more sophisticated with digital
Alexios Mantzarlis: Ha ha ha.
Sam Rabiyah: security, but I think it it used to be a lot easier to do things
Craig Silverman: Right. Yeah. Okay, last thing for the two of you. what is a favorite thing that you love to do away from any screens? Sam.
Sam Rabiyah: I spend way too much time on screens. I'm really currently trying to evaluate that in my life right now. But I
Alexios Mantzarlis: Hm.
Sam Rabiyah: love cooking and I love eating. Maybe together, maybe separately. I'm also a big fan of the New York City ferry. I think it's a very underutilized form of transport, especially on like a beautiful
Alexios Mantzarlis: Yes.
Sam Rabiyah: fall day. Yeah, I'll I'll leave it at those things.
Craig Silverman: Nice slam.
Lam Thuy Vo: to touch rocks and soil meaning I love to go rock climbing
Alexios Mantzarlis: Ha ha ha.
Lam Thuy Vo: and I love to I'm a ceramics instructor. So I make not only do I not touch my phone when I do ceramics, I make twelve students stop touching their phones when they play with clay. Yeah.
Sam Rabiyah: I'll just say Lam is like one of those people with twelve hobbies and is like amazing at all of them.
Alexios Mantzarlis: That's awesome. I swear I think I saw you some of your pottery because I was looking for the bio Lama. So do you ship to Italy is a question I will ask you later. also, you will, there we go. Also,
Lam Thuy Vo: It will. Yeah.
Sam Rabiyah: Mm-hmm.
Alexios Mantzarlis: Ray the fairies, you folks will know, but the New York City Fairies, Craig, who isn't a New Yorker or a former New Yorker. they're all named by second graders in public schools in New York City. So they have all of these like fun, whimsical, cute names.
Sam Rabiyah: So cute. Yeah.
Alexios Mantzarlis: But this was great, folks. Thank you so much for joining us. where can people keep up with your work or most of your work?
Sam Rabiyah: For me, Best Place is just my personal website, Samrabea dot com, just my name. I'm like sort of online but sort of not, so that's probably the most reliable source.
Lam Thuy Vo: Yeah, DM me. I would like come come find me at the ceramic
Alexios Mantzarlis: Ha ha
Lam Thuy Vo: studio. But actually weirdly enough, LinkedIn these days, I think it's the least horrible or algorithm out there these days that I don't have to chase to get attention from.
Alexios Mantzarlis: would you have imagined ten years ago to utter these words? But yes, I agree actually.
Craig Silverman: Ha ha. Yeah, well, and that's it. You can follow Indicator as well. We are on LinkedIn and other platforms. You can get more episodes and other video content from us by hitting subscribe on this channel or on whatever podcast platform you might be listening to right now. And you should definitely come to our website, indicator.media, and sign up for our free newsletter. So thank you to everybody who watched or listened. Thanks again to our founding sponsor, Lead Stories, and especially thank you to Sam and Lom for joining us today.