Representative appearances

• AI Summit - Interview Dr. Richard Socher, You.com | #MTM25

  First of all to tell you that what we've been talking about for the last couple of years has finally arrived and you have seen it in your news feeds probably or you have listened to synthetic voices a lot of AI generated content starts filtering into our media diet. The main source of it is still the American president. Um he posted on the weekend, maybe you've seen the video. I'll give you a short summary. He's in a jet. He's wearing a crown. He's dropping some stuff on protesters and it's all fake obviously. So, these are great times if you're a fan of propaganda. But what does it mean for journalism? That's what we're going to talk about today in our AI summit. And I want to start uh with someone who is uh right at the heart of the AI revolution in Silicon Valley. Um he's also from Germany and um he's going to be on this screen and we have to start early because he's in California and it's very late there. So he's the founder of u.com which used to be an AI powered search engine and they pivoted and now they are uh offering AI to companies to use their own data and also external data and um they work among others with VA and build falak and as uh we shall hear later with dodge press au as well. So here with me is the founder and CEO of you.com Rishad Zoha. Hello Rishad. He should be on the screen. There he is. Welcome. Now, uh, one of the reasons we're here, is that, uh, some journalists are insecure about, uh, what AI means for them. So, maybe at the beginning, can you tell us about your vision for AI media in the future? Yeah, it's a tough one. I don't know if I have all the answers. Um, can you hear me? Okay. Yeah. Yeah. Okay. Great. Very good. Um, yeah. I don't I don't know if I have all the answers, but I do think, uh, you know, never waste a good crisis. Every crisis is an opportunity. Um, I do think AI has a potential to make journalism better. Uh, I don't think it'll make journalism obsolete. Uh, I think there are a couple of different uh, opportunities. I think the most obvious one in the beginning is that AI can become a co-pilot uh if you will uh not not a replacement right uh research that used to take you many hours or even weeks can now be done in minutes or or a few hours um uh factchecking uh various things and articles can be automated. uh sticking to a style guide uh can be helped uh and done by a style guide uh agent. Uh searching through decades of archives uh instantly uh can be done uh with some of our technology that's what we do with the DPA for instance the doagen tour um the telegraph in the UK we work with like fact-checking agents uh and so on. I think there are you know a lot of ways that uh you can make journalism more efficient with AI. Uh then of course there are uh new use cases uh that can get unlocked when you use AI uh that were just impossible before. I call it kind of the village journalist uh opportunity. Uh you right now no one can really afford to have a journalist in every village of like 200 500 people or so. But with AI uh and the usage of for instance public data, voting records, government budgets, infrastructure projects and so on, you could actually write uh articles about how did this zip code vote even if there are very few people. So you can kind of scale and stay hyper local. Um I I do think uh obviously journalism is an extremely important part of of democracy and uh while I think there's a lot of pressure um on it and I don't think uh the content that is uh the least unique there will probably be fewer folks working in that but it's also clear to me that AI will not take away a lot of the important parts of journalism which is which are like editorial judgment uh you know knowing what stories actually matter and having a sense for that. Doing actual interviews with people and building human connections, doing investigative work and original reporting, building trust, having a brand. I think a lot of uh companies now need to lean into their brands. Um, and then thinking about just even asking the right questions. And so I could talk for a long time, but those are a couple of ideas of how AI and journalism uh can I think very mutually beneficially coexist. Mhm. Now you mentioned trust. Generative AI systems they work probabilistically which means they can not be as exact as someone who is actually checking a fact and we are in the fact business. Some might say we are in the truth business. So what do you think about can AI ever tell us the truth? I do I do think that yes uh in fact we can study this uh and there are many research papers also that go through uh and you know evaluate how often uh facts have correct citations, how often the citations really are real citations not made up um and uh how many unsupported statements there are in outputs and so on. We work with uh organizations like Benfalak and Apple Pharmacy, health magazines, work with uh insurance companies, with banks, uh with the NIH, the Institute for Advanced Study in Princeton, uh you the National Health Institute, NIH, and a lot of other organizations that really deeply care about accuracy. And so it is possible, but it's also effort. Uh and I think a lot of companies would rather give you a halftruth that is very fast and creates more engagement uh than to tell you like I don't know. Uh you.com we actually trained for years uh to be willing to say I don't know if there are no good search results. A lot of people underestimate, they think about AI and they think about large language models, but large language models are to a large degree garbage in, garbage out. And it matters a lot what you put into their prompt. If you put nothing into their prompt, they're more likely to hallucinate something. Uh if you tell them in the prompt, it's okay if you don't find the facts here to this question. It's okay to say you don't know. Um, if you take for instance a very simple article, let's say you want to ask like what happened last quarter in the financial markets in Southeast Asia. That's your question. Now, if your search back end finds this nice document that talks about last quarter's results in Southeast Asia, but that article happened to be from three years ago, and it takes that article and puts it into the prompt, and then the LLM very neatly summarizes it, it will be wrong. Uh, and so we've all been there. the search back end is very crucial infrastructure part that a lot of people um have uh underestimated the importance of uh but if you don't uh I think you can make this technology very accurate and and that's what we've been working on now at the virtual companies another subject when it comes to trust is copyright uh many among us in the media feel kind of robbed by the early land grabs uh that were done by uh AI companies do you think that After all this criticism and also the lawsuits, there has actually been uh a serious change of mind in the valley or among your co-founders. Yeah. So, um I think it's it's kind of interesting. There's there's copyright and then I think there's what I would call learn. Um and a lot of folks try to sort of expand copyright into learn. Um right. So if you say for instance you're copying this exact painting um then uh that's that's copyright but if you get inspired by the style of that painting and you paint a completely new subject in that style I don't think that's necessarily copyright right and sometimes uh I wonder there are a lot of different aspects of this I do think there are some tech companies that took articles from the press basically rewrote or copied large chunks of it and then did their own SEO trying to really actually steal content. One of our competitors uh had had done this and that really erodess trust and that should just be actually uh litigated. Um and and that that's just not right. Um I do think convenience wins and so if you build a platform that allows people to just get an answer uh right away then that will continue to just have a lot of draw for users. But uh I do think uh there are a lot of partnerships and then when we actually partner with companies like the DPA or Pilfalak and the Telegraph and others, we just do AI for them on their data and we don't train anything uh for anyone else. So uh there are lots of different partners and different ways of making it work. I do think there's undeniably a lot of pressure and I think there's a short window right now where the brands that have a strong the outlets that have a strong brand um can teach users that if you have questions related to your brand could be like you know hyper local things could be specific for finance where you have trust if you're hund for instance uh or um financial times or can be you healthcare uh types of subjects. If you're taking home show and things like that, if you have a strong brand, now is the time to teach your readers that they can come to you for answers. They can come to your platform to chat and it doesn't just have to happen on Chat Gvt. Thank you, Richard. Um I have one last question for you. As a Silicon Valley tech founder, do you think the press has been too nice to Silicon Valley tech founders? I don't know. Uh I think it depends on who you ask. Uh I think uh it's quite high variance I hear. I see. Thank you so much for taking the time to talk to us and thank you and we say hello to California. Thanks for having me. Have a great rest of your day. Thank you. [Music]