the night sky China baby see the dog gets the violin
and the dogs that I see the new stars of Butoh in your home makes no sense
a van Ballard is be my nice guy
nice
lights got hot and being on dark gray cloud. I can forgive you our song was cold and the silence is so loud. You just stay dancing
I can find where to get sales
new stars of your home makes no sense
you be my night sky
beam Nice
nice.
Last insomma my thoughts and when we start to think, oh the closest stars and their brains are tried so hard. All the scars they cut so deep bathroom have like so long clip nine wing and I can see the crowd. Can we stop this gravity it keeps bringing me down
Thank you Bye
give your gave to me so long, sleepless nights are laid awake just begging God says suicide through you stands in an angel stand firm have shown me how to the game I can see the crowd. Can we stop this guy that sees me? It keeps banging me down. He can
say goodbye
the darkness
You All right everyone. Thanks. Thanks so much for coming in at four o'clock on a Friday. A ton Horowitz. I'm a partner operations manager working at Google. I'm really excited to be here. I'm actually from Philadelphia area I grew up here. live here now about eight miles away from here. So it's it's really exciting. To have a q&a here in my backyard and see all the energy this week. That's just been amazing. So hopefully you'll find this really interesting. I will say actually, I'm not the person that was originally supposed to do this talk Marian oniac who was a trainer for the Google News initiative was supposed to but unfortunately, she got sick. She's watching the live stream. Hi, Mary. I hope you're doing well. Thank you for putting this together. I will say this tool pinpoint is really awesome. I knew a little bit about it before I found out on Tuesday that I was going to be doing this but since diving deep into it even more in the past couple of days. It's really an amazing tool. It's free. It's the kind of thing you know, when I was a local reporter awhile ago, about 13 years ago I would have sort of killed to have so hopefully if you're not familiar with it after today, you will be and yeah, let's just dive in. So just one one quick thing. So this session here is one of two that we here at Google and YouTube did. There was another one yesterday that took place in this room on YouTube shorts for news. If you go to that link or scan the QR code. We have a microsite for you that has information from both of these sessions, as well as some of the resources that we'll mention here. I'll leave this up for another minute. And then I also have the slide with the QR code at the end of the session. So continuing on, you know, as I mentioned, this is this is brought to you by the Google News initiative. And just to give you if you're not familiar, a little bit of an overview there's there's basically three pillars of the Google News initiative, advancing the practice of quality journalism, strengthening and evolving publisher business models, cultivating a collaborative global news community. That work takes shape in a lot of different forms. We do a lot of trainings. Since you know, we launched the global news journalist training program with the news lab in 2015. We've trained about 457,000 journalists and seven country and 70 countries, excuse me, and I've done about 2.4 million online trainings. You know, we do a lot of our own combating misinformation. You know, also, we'd like to sustain essential reporting through some of our different journalism funds. We work on developing new formats for news and also advancing technology in the newsroom which the pinpoint tool that you're going to see today is a really key example of that where we've partnered with the news industry to sort of understand what what the needs are and help kind of develop tools that can really help journalists do their work in in new ways. So, so pinpoint that we're going to talk about is part of a broader suite of tools called journalist Studio. You can see here it's a collection of tools that have particular use cases for journalists, as I mentioned, some like pinpoint were developed specifically for journalists, others, like Google Trends, you know, have been around for a long time, you know, has uses beyond journalism, but certainly is also an important and useful tool for journalists. So these are just some of the tools that you'll find within journalists studio. Okay, so talking about pinpoint, you know, and there you have a short link for it, you know, if you want to get to as I said, you can also get it to it from the journalist studio page. It's one of our newest tools. As I mentioned, it was developed to really help journalists in their work the tool further and further how many people are familiar with pinpoint raise your hand couple of people have used it, okay, great. You know, it's it's one that continues to evolve. So even if you are familiar with it, you'll probably see some new features because new features are rolling out all the time, that are largely developed with feedback from journalists in mind. You know, some of the things I'm going to show you today were requested specifically by journalists who've been been using it. And then a one big new feature which we'll see at the end, called extract your data. That was actually an open beta very recently in the past couple of weeks. So you're, you're probably some of the first folks to see it. Really kind of a very cool tool. So you know, search and obviously search is core to what Google does as a company and our mission. And that's really what is sort of a big part of pinpoint. It's using the best of Google's machine learning and AI capabilities to help you make sense of your documents. Because, you know, if you're a reporter, you're going to be dealing with a lot of documents they may be board meeting agendas. They may be records that you get through FOIA requests, you know, things that you download, and you know, we've Things have certainly gotten better in terms of documents being online, but you're still you know, going to have probably PDFs and other documents that you know, on the surface may not be that easy to search through on your own. So what you'll see and again, the format here is I'm going to show you some slides that that show the the tool a little bit and then we're actually going to do a demo of it of it live so you'll kind of see how it works but just to sort of get you a little familiar. So you know, part of it is when you upload a collection of documents. And immediately upon uploading the documents, you'll see a breakdown of the most mentioned people, organizations and location in your document like the machine learning will just scan it automatically. And see okay, these are names that are mentioned and will give you you know, account of the name and then how many times it was mentioned. If you click into it, you'll see you can go right and see all those mentions. It does the same for organizations and also locations. One thing that is new, pretty recently is you can actually search date ranges. So within within all your documents you can say Hey, show me any references you know between this date and that data. And again, that was a feature that was requested from journalists who had been using pinpoint and it's now been rolled out. So we've talked you know, we've said documents, but actually it goes beyond just you know, a text document you can see actually, you know, it supports lots of things PDFs, Word documents, also, email archives, PowerPoint presentations. You know, if you cover government or military, you may have a lot of these that you get in the course of your reporting. Some other cool things which we'll see is you can do audio so one key and very easy use cases. It's a great tool for audio transcription. You know if you if you upload your interview, you know with with sources or notes you'll get automatic transcription which which we'll show you. One thing I thought was interesting since we're talking about PDFs, I don't know if anybody saw but the creator of the PDF actually passed away yesterday from Adobe, who created it in 1993. So I guess we can say sort of in in his honor, we'll be talking a lot about PDFs and really, you know, using pinpoint to to really expand the capability beyond those documents, you know, which is pretty ubiquitous in everything we do. Another you know, piece, which you'll see in here is that, you know, when you upload a collection, it's private by default just to you but you know, if you want, you can share your document sets with others, you know, very similar sort of interface and functionality to Google workspace, where you can add collaborator, collaborators, you know, that might be a fellow reporter you're working with, or maybe your editor to look things over. That's also a new feature, whereas previously, adding somebody it was view only, but now you can actually add them and assign a role as an editor or we're just a viewer. So again, another feature that was was requested by journalists using the tool.
Another thing that is also new, which you'll see is that you can make your collections public. So if you have documents that you think that the public or other journalists or researchers may find useful, you can actually publish those and they can be available and you'll see those as well. All right, so now we're gonna switch over to the demo, and I'm gonna give you a tour around the tool and how it works. Let's see. Okay, so as I said, it's, it's, um, it's part of journalists Studio, you can get to it by going down and finding pinpoint one thing to note, you know, you if you go to it now, it's possible you may have access to it. It's not open to everybody by default. If you don't have access to it, there's a button you can click and then there's a form that you can fill out and it will just ask you what your organization is. And there's a team that reviews these requests. They're generally very fast to get to get back on those usually within 24 hours. I know from Mary when she's given this presentation before she's mentioned that some people have requested access the beginning of a presentation and gotten access pretty immediately. So, you know, again, because it is a tool aimed for journalists. It's not just out there for everybody. So when you log in to pinpoint, you're going to have two tabs. Here. You have the Explorer, tab and the My Workspace tab and let me maybe just make this a little bit bigger. Hopefully that will help. Okay, so I'm going to start with Explorer. And as you can see, this is the section that has what I was mentioning the public collections that others have made available. You can you can see all in organizations are actually also pretty new. You can search and see you know from specific news organizations that have published collections. So here we have the Associated Press, you know, with with some collections, you know, 259 documents, Trump indictments, different, you know, bills. So lots of lots of news organizations have already been putting collections on pinpoint, you know, again, making them available for for people to use. You can also certainly search, you know, collections. So, you know, obviously you could probably will find a lot of stuff related to Trump on here, you know, both from news organizations as well as individual users, you know, in the Explore tab, so if you're interested in playing around with it maybe but you don't have any documents or collections yourself, this is sort of a good way to kind of get in and, you know, start playing around, you know, you can just just come in again and open up any of these, any of these collections. And yeah, and just start, you know, start kind of working with them. I'm going to jump back to my workspace and want to show you a couple of things before we jump jump into some of the collections. So you see in the little Settings icon, you can actually set the language for pinpoint which obviously if you're you know from outside the US sir in a non English publication is very useful. So in terms of you know, when you saw the, the little entities that had people's names or had locations, you can choose what language that displays in, so a number of languages in here. You can also for your audio transcription, you know, you can you can set you can set it if you work in a particular language, so that when you upload your audio, it will be transcribed in that language. One thing to note is that the language settings are global. So they apply to all your collections. You can't yet set them per collection. But, you know, again, it's probably useful depending on you know, which language you work in as a journalist. Alright, so just showing you some of the collections themselves. I'm going to jump in, we have and you can see, some are ones that I've created. Some are ones that colleagues at Google have shared with me. So these documents here these JFK files, you see, you know, there are 50 documents that have been uploaded here and pinpoint has, you know, searched through them and as identified, you know, all these different people, organizations, locations, you know, that are show up in the documents. One thing that's a little bigger, that's helpful. Yeah. One thing to note is that some of somebody like John F Kennedy, so that he is obviously a known public person to to Google and the knowledge graph. So if I click in here, the people you know, will show you like, okay 20 year documents, John F. Kennedy was mentioned, but pinpoint knows that John F. Kennedy, and let's see, President Kennedy are the same person so instead of making an entity for John F Kennedy and another one for President Kennedy, it knows to combine those together. For people that are not known. It doesn't do that. So if your documents have, you know, my name, a town Horowitz and then if a town Mr. Woods, it's going to have have them listed separately, but that's, you know, just just one thing to note and again, it's it's very useful when you upload just to be able to immediately have that rundown you know, and if you click into any of these, you know, it will take you right to where so you can see like this was originally a handwritten document you know, that at some point in time, you know, was was scanned and you know, pinpoint is able to recognize here, you know, the typing of that. Beyond that though, I will show you there's there's the searching is is very good in here. It's very, it's very powerful. So if I look at let's see, if I look at these NASA documents, and if I search for Miss State, let's see a search here for Loomis state. And I click in you know, so I see it says Loomis and state here, but if I click in so you see Loomis was from the type one but it actually even picked up the handwriting that has stayed here. So very, very cool in terms of like, what it what it can pick up. Beyond that another another example that kind of shows you about it is if I do s, t, d n in here, let's say again, you can see well, okay, that combination of letters was found in this in this document. But if we click into it, you can actually see it was picked up in this sign right here. So it can even recognize, you know, on on the desk, that good, you know, different signage, so this could be useful for campaign reporting, you know, or if you have a bunch of photos that have, you know, signs from somewhere. It's really useful. It's really powerful. A couple other things to show you just in terms of the searching. So also in these in these NASA documents, sort of similar to similar to the way that it knew to group the people together. If I searching here for moon it will know it will have both moon and lunar and you know a knows that lunar you know, is another term for for moon so there's a lot of intelligence built in for things like this that really, you know, when you get a bunch of documents, it sort of helps you, you know, know what you have, know what you're looking at, let's see some couple other searches to show you or just things that I've uploaded, you know, maybe give you some ideas of how you might, how you might use this. The people that may be familiar, Reuters, you know, does the digital news reports every year, so we like uploaded a couple of those and, you know, like, Okay, people are mentioned in here. You might like Oh, I wonder if you know, my news organization was mentioned like I used to work at CNN A while ago, you know, I can click and see like, okay, you know, CNN was mentioned, was mentioned five times in here. You know, and here's exactly where it is.
Another very cool feature of pinpoint and this is particularly helpful if you you want to make some of these files public. So if I were to go here, and say click on one of these Jack Ruby ones, and say, All right, okay, so actually that one, I think I already did, let me do a different one. Okay, so say say I wanted to call out this particular part in the document. So if I highlight it, and then I get a pop up that says highlight and get link, so I click that. And then if I just go into another browser tab, I actually have a link that highlight you know, takes me exactly there. So again, in your reporting, if you're reporting on documents, and in the course of an article you want to say, hey, you know, in the document, you know, Trump indictment obviously is a perfect example of this. You know, if you want to reference a specific part, you know, instead of necessarily having to copy and paste the whole thing, you know, for your online viewers, you can use this functionality and highlight and get the direct link to it and take it right to your readers right to there. So that is, that's a very cool feature of it. The The other thing I wanted to show I mentioned the date search is relatively new. So we also have in here, I uploaded some state of the union speeches. And if I was curious about okay in all these speeches, you know, let me see, when 2020 was mentioned, so if I put in, you know, just for the whole whole of 2020 or any date range you want, you know, in these 11 documents, it will show me like okay, well here's here's some different mentions, you know, okay, this was a speech in 2020 State of the Union, but then also, in, you know, in Biden's 2021, he referenced, you know, sort of in the notes here, he referenced the George Floyd murder that was in 2020. So that's a new feature. The you know, the ability to search by the dates mentioned in the document again, another thing that was requested from journalists and the pinpoint team was able to add anything else on do so Okay, so that's that's the basic as far as text documents, photos, very powerful search. lots you can do with it, to sort of see what you have. The other piece of it, which is, like I said, a very common use case amongst journalists, is for audio, audio recording. So one thing you know, just to sort of test this out like I was looking, I was a long time ago, I covered county commission and City Commission meetings and you know, didn't always get to go to them and, you know, like so here in like Orange County. They have all their meetings and they have some as NP threes that you know, audio recording of it so I went in and I uploaded, let's see a bunch of them, you know, and again, you can see the, let's see, you can see who's mentioned and all that. The way the audio works is if you click into it you know, it's it's it's a it's we are back and let me inquire as to where counsel is. So just the different renames at a time if you could please reference spots that are just pauses but you can you know, so it's like it's free. You know, if you have an audio recording, I know there are a lot of transcription services people use that cost money. This one's free. It's on here. You know, you can, again, it's an easy way to sort of search through things. There was an interesting eye to look at this, you know, you can also use pinpoint, frankly, for personal stuff. So I had on my computer I had actually some audio from my wife when she was at a summer camp in 1987. Her parents did. They did tape recordings, and they sent them to her. And so I had one and I uploaded it to here. And it was just sort of interesting. Well, first, it was interesting to see, you know, what was was mentioned and, you know, like, oh, wimpleton You know, so like, if you click to that and brings you right here and then you know like if I you know Martina Okay, it says Martina Nava, so the tennis is my wife's father.
The other day that Martina Nava terone.
What's what what's very cool about this record, I think it's the eighth
time in a row, she wanted Wimbledon.
So that was what's what's cool. So it's like, okay, you can see the transcription didn't get it exactly right. So but you can go in and edit so I can go in and I can say, okay, I can say Martina See, I wrote this down. Because I don't remember how to spell it myself. No room to move. And then one again, so you can just edit it yourself. And then you know, and I think
this was this record, I think is the eighth time in a row she wanted women.
Yeah, so this is also you know, one at Wimbledon. So that's very cool. You know, it's very easy to do your audio recordings there to edit them. Then you have them you know, easy to search. Similar thing if you if you if you have hours long interviews, you could upload them use the same capabilities that you can use for the documents. The other thing beyond just straight audio files that you can do is you can actually take video files and upload them and then the audio will be extracted. You can do the same thing. Since we're in Philadelphia and since I'm a sports fan, there's a very famous press conference with Allen Iverson where he talks about practice and going to practice and so I downloaded this video and I uploaded it so this this description from YouTube says the others practice 22 times but when I actually uploaded it to pinpoint let's see. So it's all has all the people, you know, and if I go in, if I go into here, you know, okay, I'm gonna say it's like, oh, Arnold Schwarzenegger, like how did he Where did that come from? You know, so if I if I click in here, you know, you can you can hear
what I gotta do. I'm gonna come in here. I'm gonna be the biggest, baddest strongest. I'm gonna come in here. I'm gonna be big.
Yeah, so you can and then actually looking at looking at let's see, practice. So that said, that said he, he. He added the 22 times but pinpoint, actually says 35 mentions of practice. So it's kind of you know, like if you had something like this, again, when this came out, somebody probably had to make the transcript sit there count how many times it said practice, but if you had a tool like this, you could just take the video, feed it right in, you know, obviously we have we have lots of video captioning tools today. But you know, this is a free one that's easy available. So lots lots you can do there. So I'm going to jump back to show you a little bit about the second part of of our demo, and then getting there and I will I'm going to leave time at the end for questions, so don't worry about that. So the newest feature is called extract structured data. And really what this what this does here is it lets you take a collection of similar documents that are in a similar structure and turn them into a spreadsheet. So for example, the one shown here, say as part of the public records request, you got a stack of W nines, you know, are some document employment applications, other things and the document, you know, each of them has fields for name, you know, date, all that kind of thing. But in order to sort of sum that up and you know, you might have to go through that by hand or do you kind of copy and pasting because again, the PDF document doesn't always you know, you get there's tools that can scrape PDFs, but what this tool does, it lets you easily turn these these documents into, like spreadsheets essentially. CSV files. And you'll see I'll show you this again, I'm going to sort of show you some slides, and then actually do do it live kind of show you how it works, where what you're you what you're going to do is you tell pinpoint, like okay, these are the fields that I care about, and you highlight them, and then it does the rest. Essentially I'm going to show you how this works. Just you know, sort of dig, let's dig a little deeper. There's a couple of different ways that you would do this. So key value pairs, you know, so, a key like name and then the value 18 Horowitz tables, repeated sections. These are common ways that documents like these have been created. And so pinpoint has a way to basically annotate each of these types of documents. Again, key value pair just you know, for an example take a drink. You see you know, your birth is a key 92 is a value all those type of things very common on on a document tables, you know, things often will come and tables as well. Repeated sections here. So these are just different types of way that the data is shown and displayed that are fairly common and that the pinpoint team has sort of identified as the most common. So again, you'll see when we do this live, but basically, once you annotate it, you'll you'll instantly kind of get to review it. Just some things to keep in mind, you know, before we jump in into the demo, the you know, obviously it's it really works best when the documents have the same structure. So on the top, those are all W nines, they're all the same on the bottom, those are all invoices, but if you look they're all different, you know, so if you had three different invoices from three different companies and the style and format of each one was different is not going to work well but if you had one company and and all of their invoices, and it's all in the same style and format like that, that will work well and that's really you know, kind of what what this tool is good for. Okay, all right demo time. So let's see. I am going to jump back into pinpoint here. Let's see. And I'm actually going to sort of do it from the upload point. So again, as mentioned before, any collection you create is as private by default. Let's see, let's see extract data. Oops, I'll call it that. You know, and you see then you get your upload screen. You know you can upload things that are already in your in your Google Drive or on your computer. I'm going to grab from my computer. Let's see. Am I just Oh in a Fact Book. So some of you use these files from the CIA World. Factbook, which is they're sort of common files that Wikipedia uses. So I'm going to select these 10 files and watch as I as I upload these you'll see on the right hand side of pinpoint, you'll see those those entities start to fill in the names and the places and the locations pretty much automatically as it's uploaded. Or actually after here and then once it goes okay so like look right here as he's processing see, it immediately adds these people in here and then that was very quick for the 10 documents. You know, that's any anything you upload it's automatically going to do that you know, obviously video or audio files might take a little bit longer, but when you upload, it's automatically processing and analyzing using machine learning to kind of identify these different pieces. So that's basically sort of the basic pinpoint, and then what we showed before, what's new, and I'm going to show you here is where you see it says extract structured data beta. So this is the new feature. Very new, as I said, I think past week or so, if you have access or will have access to pinpoint, you'll have access to this as well. So I'm going to click on this you know, again, gives you a little tutorial if you want you know, kind of explains a little bit what there's a video there that that's very helpful. And then I'm going to click get started. So it says it requires further processing of your collection. In a click this shouldn't shouldn't take too long, you know you can also watch watch the video. You can see we have 10 documents. So again, to be able to do the extract structure data that we have. There's a little bit more processing that goes on. You see it's almost almost done. It's up what's done. Okay, so now we're gonna go and say annotate collection. So here's some we're now in the extract Structured Data screen. And so what what pinpoint does is it selects one of your documents so I had 10 documents at random, and for what it calls a golden document for you. And this is basically the idea is it's a document that that is going to be like the sample document or the document that the rest of them are trained on. If for some reason you don't like the one that it picked you can you can click here three dots and change it pick a different one you know, but for our purposes like this one seems fine that it picked you know, it's again, it's one just to show you what this does. They all kind of look the same but country report, you know, you see it's got some some data chief of state government type capital, you know, land area, and it's got a table with, you know, the largest cities. So, you know, it's got some data in here. And basically, as I said, if I were to look at any of these 10 documents, they're all going to have the same structure to them. So what what I'm gonna do first with the key value pair is, is annotate the document and I'm going to select the value and you'll see kind of as I'm doing this, one thing to note so here even though this Columbia is only this many characters, I had some longer titles in there like Afghanistan. So just sort of a tip and good practice is to kind of extend it out here.
So you can see so it basically picked like I said, Hey, this is the value and pinpoint picked up that this is the key, you know, country report, again, sort of continuing through here. I'm saying okay, this is this is the value. This is the key. So and you can see at the bottom as I'm doing that, it's automatically doing it to all the other documents. So I've just done country report and Chief of State for Columbia, but it's gone into every other document and done the same thing like you know, instantly as I'm doing that, so fine. If I were to jump into you know, one of these other ones France, you can see it did it right there. As I'm going through again, you know, highlighting the value, giving a little bit of room, picks it up, you know it as you can see down here, it's sort of it's sort of showing you as you add a new key value pair that it's being added into, into your into your data. So again, it's building that spreadsheet for you based on what's in these documents. Again, you'd have to do this potentially by hand, or you know, with copying and pasting on every single document, but instead because they're all the same, you can just go and highlight here, again picked it up that it was the capital. When I've done this before, it's some you know it incorrectly grabbed this one time, and if that happens, you can actually go and you can delete it you can edit it. So there's a lot of a lot of functionality here you know, again, if I wanted to give it you know, a label or if I wanted to edit, you know I can I can always do do that. So that's that's the first one that's key value pair. You know, again, this is this is a common, a common way that a lot of a lot of documents you may get may have data. The other one I'm going to show you again, there's there's four different types out here up here. I'm going to show you let's see the tables here. So if I click on tables, so basically what I'm gonna do now is I'm going to tell pinpoint, like, Okay, this it says highlight the table area. Oops. So I'm highlighted oops, I'm highlighting my table. Okay, I'm gonna give it a name. Largest cities I'm gonna save it. If if you maybe saw quickly, but if if for some reason, it missed the column. It allows you to kind of add it in there. The other thing I'm going to do is so this table actually continues on to the second page, but so far, I've only captured the data here. I could add, you know, a column if I needed to, but what I'm going to do is I'm going to continue and I'm going to drag I'm gonna drag this down. So now I have the whole table in here and save it or just table largest cities. And again, just like with the key value pair, I click over to here, so I'm on the Colombia document. But it's done this on all of the other ones. Afghanistan, you know, Brazil. It's also interesting to note that so the one here had 1-234-567-8910 10 rows in the table, but some of these other ones have a lot more like a lot more rows and even though I didn't highlight, you know, I only highlighted 10 rows. It's smart enough to know that like when it goes to this Egypt one that it wants this whole table here, so you can see. It was able to capture everything on there. Pretty, pretty awesome, pretty powerful. Again, once you've once you've done this, so So this part of annotating and saying like, Okay, here's my document, here's the data that's in it, here's the format of it, here's what I want you to focus on, then then you can go you know, and again, it gives you this nice, it gives you this nice preview here so you can make sure you know you could look through if for some reason there was, you know, a blank whether there are some blanks in here, actually, I mean, we could see we could go in and we could see what happened here. Let's see, okay, so there wasn't anything in there so it wasn't, you know, an anomaly. You could change that if you wanted. But basically, after you've done that, if you extract it, you know, it will you can download your results. And basically I'll just show you in my computer downloads unzip it, you know you can see you know you've got a nice you've got a nice spreadsheet or a CSV here. You know, you can I sort of cooking show version, I had brought this already into into Google Sheets, you know, so again, I just upload that CSV that I just downloaded. Now you have a very workable, you know, workable data set here. The other thing that's very cool to note is that for each row, you have a validation link so you can click and it will take you exactly to the spot in the document where that piece of data was extracted from. So you know, if again, if you're looking at the spreadsheet, and something doesn't, doesn't add up or it doesn't seem right, you have that link to go right back and see it. So that's, that's our second demo. Again, brand new tool extract structured data. Very, very, very powerful. Just, um, you know, in our last, we've got about 20 minutes left. So I'm gonna show you a couple of case studies of how journalists have used this and then we'll open it up to questions. So a couple of stories. I mean, you're probably familiar with Maria Teresa. She's used pinpoint a lot in her in her reporting. The Yeah, she's used it to go through 13,000 documents. You know, for a story about history and corruption, and CIA, CIA daily briefings about martial law in Manila in 1972. So again, like 13,000 documents, a pretty powerful tool, in terms of what you can upload to it. Oh, actually, that reminds me I forgot to mention a couple of stats. So you can have up to 200,000 files in a single collection, up to one gig per individual PDF file. Each account gets 100 gigs of storage. However, if you do ever run out of storage, you can email the pinpoint team directly. They want to work with you on that. So if you if you email them and say, Hey, I'm working on this story, and we need more than you know, 100 gigs of storage, they're very likely to help you with that. But yeah, so it's the the amount of documents that it can process. is very cool. So this the sorry this Yeah, these were the Maria Raisa some of the stories again, using you can see in the screenshot, it's a briefing from October 3 1972. You know, and again, that this was all this was all work that she was able to do through pinpoint through the analysis here. One of the better known or examples of pinpoint is actually the Boston Globe to the series of stories in 2021 that won the Pulitzer, called Blind Spot about dangerous drivers and dangerous roads. So the team requested public records from all 50 states. They built a database looking at crashes, trucking mishaps, you know, sort of seeing that there wasn't really connection from one one state to the other. You know, actually, we have in the link that I shared, there's a nice blog post and video that I think one of the investigations editor from the globe wrote and basically he was saying, you know, that pinpoint really allows them to put everything in one place and kind of parse it right. So they, they know they wanted all these documents, but they weren't sure initially you know, what do they have? Where are those connections going to be made? You know, how do you how do you kind of find them and so pinpoint really allowed them to do that. The let's see,
again, some of the documents they said were, you know, we're not photocopied well or they were upside down. They even one of these stories was was written quickly on deadline. I mean, you saw how fast it identified those entities from when you upload a document. So this doesn't have to be a tool that you know, is only used for long investigations. You know, you could, you know, like I said, I remember when I was a local reporter, we would get these meeting agendas and you know, sort of look through them and like, Oh, is there anything interesting here, but, you know, imagine just putting that all into pinpoint, having it spit back and then being able to see like, oh, this controversial developer was mentioned, you know, this many times or, Oh, they're gonna you know, talk about this or just you know, interesting, interesting connections and stories that if you were to read through that all those documents on your own you really wouldn't be able to pick up on that's what pinpoint allows you to do. Another one here from Dance magazine, so it's not you know, always hard news that use pinpoint to generate transcriptions of interviews with sources conducted over zoom. You know, again, and just just pretty basic, you know, again, the reporter said, you know, transcribing interviews can be the most time consuming part of my reporting process, so it's been helpful to quickly and easily find facts and quotes I need using pinpoints timestamp feature which you saw, you know, where every time there was a pause, it had the timestamp and the embedded audio in the PDF. Washington Post has also used it as well you know, as part of their their reporting on on different topics. This one was kind of cool, actually, sort of to that point I was making about, you know, you may not know what's in documents without something like pinpoint. So, the reporter, you know, was thinking of a story that he could write for the MLK holiday approaching, and just decided to like us pinpoint and look at 15,000 documents from MLK and see like, hey, was Baltimore and MLK mentioned like, well, what like, how did Baltimore figure into the into these MLK files? You know, and and found it you know, found interesting stuff, you know, and basically wrote a story that that's here below, just talking about, you know, Baltimore's role in you know, and in this case, this federal agents blame Martin Luther King's nonviolent movement for violence in Baltimore and other cities. So you might not have known that without this right? And that maybe wasn't as big of a headline or a story that, you know, was known. So you know, it's just an example of the kind of thing that you can do when you upload another one USA Today, obviously, around COVID You know, went through hundreds of documents, looking at nursing home records, you know, just to see, you know, yeah, to see what's in there, and again, sort of state by state reporting, you know, something that USA Today is kind of known for So, lots of things there. So again, here's the QR code that has the link to resources from this presentation, as well as from the YouTube for news presentation from yesterday. I'll leave that up now. And yeah, we've got about 15 minutes so happy to take any questions. Anybody have the link doesn't work for the QR code. We it's interesting because we work for you go to this short link, because we we did have some problems with it before but I thought we fixed it but maybe maybe some folks are still not able to get it. Any questions? Yeah. Go ahead. Come with the mic. Yeah. If you can come to the microphone because we're on live streaming. So
how does it so JFK MLK, how does that how's it How good is it filtering that you mean? The person not a street type of stuff? Are you parsing through anything? Same entity different meaning?
Yeah, that's a good that's a good question. And honestly, I mean, I'm not sure like I think you know. So your question is well, like in in the in those examples, it it identifies John F Kennedy as as a person. I don't think on the location. It has it down to the street level yet, but I can take that question back to the team again, and find out but, you know, again, as I said, the team loves to hear from journalists. You know, there's there's a whole Help Center for pinpoint, there's support resources that you can reach out to and many journalists do while they're working on something if they they run into something and you know, they have that question like, you know, feel free to ask.
When you set the default is private does that means that the documents that are uploaded, Google doesn't look at those doesn't scan them doesn't link them to your Gmail account in any way? And also, my second question was, why journalists? Why not government figures or doctors or things like that?
Yeah, so your question was, as far as the security and the privacy of documents that you upload, yeah, it's, it's, you know, it's, we take privacy seriously, you know, and it's not being used, like for other purposes at at Google, you know, for like advertising or anything. Like that. Being said like obviously, it just like a tool like Google Drive, you know, if if you have things that are super sensitive secret that maybe you wouldn't feel comfortable uploading to a Google Drive, then then you might not want to do it here but yeah, it is. It is private to you. It's not going to be used kind of anywhere else in there. And then your other question was, why is it for journalists? Yeah, I mean, Google, you know, Google has long had a commitment to journalists. And, you know, we, as I said, you know, search and the mission to organize the world's information being Google's mission, obviously dovetails very nicely with the mission of journalists. So, you know, if you know, Google, a big part of the world's information is news and information. So and I think it was just identified as something that we heard you know, through gatherings that Oh, and and others that, hey, you know, if you could build a tool that did something like this, it would be useful. Academics are also another audience that can use it as well. So thanks.
Google was famous for developing awesome products, and also famous for taking them away sometimes. So what can you say about the level of commitment to Google Hands to this particular suite of stuff because it's really looks incredibly useful.
Yeah, I mean, you know, I'm not in charge of the product, but I will say like it's not meant to be a huge consumer product that, you know, is is meant to generate, you know, tons and tons of usage or revenue. I mean, I don't think there is any revenue, you know, made from from pinpoint, and, you know, it's, it's, I say new but it's been a couple of years. That it's been around and the team has been adding to it. So as far as I know, that, you know, we're committed to it and you know, we'll keep developing it and again, the the team is very excited for for us to come here and share these new features like literally, you know, somebody emailed me like, I'll make sure you mentioned this one. We just released this one. We just released that one. So it is a team that is dedicated to this. So yeah, we're definitely committed to working on it.
I use the Google Recorder app on my phone and that now has the ability to differentiate between speakers within a recording is that feature coming to this? And then just another separate question, you say can look at images can it recognize, you know, faces of public figures or anyone else?
Yeah, so I know again, because I had the similar question. That it's a requested feature of differentiating between speakers. It's not in there now. I bring that back. But I'm pretty sure that that is a known and popular request to be able to to be able to do that. And then the second was about image recognition. Yeah, right. Again, that's, I'll just have to bring that back, you know, again, to see what your question is, can it do that kind of recognize,
like, Could you ask you know, for the for JFK, and see how many of the images that were in there, had him in the image or something like that? Yeah.
To my knowledge account I got again I can double check and you know, if you want to write down my info or my twitter whatever, you know, just can get back to you on that but we'll definitely bring bring that back. You know, to the team has something to do you know, I think it's, it's it's mostly at this point, looking at text, right? So it's full text in a photo, you know, text in the document, you know, text coming from the audio transcription element to this but you know, that's not to say that, you know, again, these are, these are the kinds of things that the team working on this wants to hear and if that you know, to me, it seems like that would certainly be be a good could be a good use case as well. Thanks. Yeah. Questions?
I have two questions, actually. The first one is about the extract structured data feature I'm curious to know, if the, if it's not like a document if it's let's say you're looking online at property tax assessments or something, and you just have like the URL for each like property, but you go to those pages, and it's all like structured like the page is a standard format. Is there a way to be able to, like would you have to print those as PDFs and then like, upload them or how would you do that?
Yeah, it's a good question. I know that. You know, when you saw the, the, the uploading, and like it listed the, the type of documents that it supports, you know, so I would think if you could get from that URL into one of these supported ones here that you could probably do it a webpage Well, HTML of webpages, so okay, that you know, and there's probably somebody in this room that would maybe has tried that or done that, but that's, that seems like that could be a way to do that as well.
Okay. My other question was just about pinpoint in general like if you wanted to say like, analyze legislation that there's similar legislation that's been introduced and like multiple states or something like that, can pinpoint sort of help you figure out like, how similar the text is, like, like, are they all using the same language are the same
so I think there is something in here. Was there one? There was one I think, actually one of the AP ones, let me see if I can find
had transgender related legislation as a collection but yeah,
yeah, let's let's I mean, let's look you can see the press as the Thank you. Let's see. So okay, bills through restrict gender affirming. I don't know if anybody if an AP is in here that knows the answer to this more than I do. But
you see, they added some labels. So this might be similar to what your answer to your question was, could you bet like, Could you look at them? Like could it tell you hey, we looked at you know, all 17 of these and found common language among it. I don't, I don't know that it would surface that on its own because it's not one of these entities, but you could, you could certainly search yourself and then and then create labels for that is my again, just found out on Tuesday that I was doing this demo, so you know, I but that would be my my guess. But again, I'll take it back as well. You know, so it sounds like you're asking like, yeah, analysis of language and kind of processing it. That way. So beyond just like, hey, these same things were mentioned to hey, this same type of language was mentioned. So great, great question and good idea. Any anybody in the room ever done anything like that? No. You have
I work on Yes. Things like this thinking about language beyond entity. Okay.
Sorry, if you already I might have missed this. Does this have an embed function like if I am writing about gender affirming care, and I want to take one of these AP documents and plug it into my CMS. Can I do that?
Oh plugin Oh like to embed the actual Yeah, like put the document on my you know, the article web page like yeah, I'm not I'm not sure that it has the the Embed I don't know if you were here. Earlier when I did you see the hyperlink? Yeah, sure. Yeah. Yeah, to my knowledge, no, but again, we'll add it to the lesson. Hopefully, it doesn't have it and I missed it. But I know that I know, obviously, you see a lot documents embedded, you know, so I know that that's probably a good request.
If I had like eight hours of audio from a Senate floor debate, and I uploaded that Do you know roughly how long it might take to create that transcript like I know that probably be a massive amount.
I can just tell you based on my own experience earlier today, I uploaded I think some of these county council things were like, an hour and a half long meetings, and it maybe took 15 minutes or so. So if my experience could extrapolate to that, you know, but again, it might be the kind of thing where you might hit up against limits, but emailing the team, they could provide some help, you know, or you might have to break it up into different, you know, into multiple files to be able to get it all in there. See, we got I mean, like three minutes left, any final questions? Again, encourage y'all to check it out. Like I know, you know, obviously we're here talking about journalism and use cases for that but even you know, even for personal stuff, like you know, I did audio when I was in seventh grade with with my grandfather who was in World War Two, you know, we have the cassette tape and I'm like, oh, I should definitely just like digitize that upload it to here, you know, to to get like an easy transcripts, you know, and sort of have that and be able to kind of jump around so beyond beyond stuff for work like it's it's a very cool thing, you know, just to be able to you know, I mean some actually the Mary who does a training mentioned, you know, other use cases, like hiring managers, you know, maybe and looking at resumes or candidates, you know, just just different ways to sort through lots of documents and data and kind of, you know, make sense of it and, you know, yeah, and kind of pull out pull out different sites. All right. Well, I think with that we can wrap up and, and thanks a lot and I'll hang out afterwards if anybody has any questions.
I can't see the crowd can we stop this guy keeps bringing me down
Thank you. Bye
gave to me so long and go, sleepless nights are laid awake just begging to most gods is suicidal thoughts for you stands in Angel stand firm have shown me how to.
