[00:00:00] From privacy concerns to limitless potential, AI is rapidly impacting our evolving society. In this new season of the Brave Technologist podcast, we’re demystifying artificial intelligence, challenging the status quo, and empowering everyday people to embrace the digital revolution. I’m your host, Luke Malks, VP of Business Operations at Brave Software, makers of the privacy respecting Brave browser.

[00:00:24] and search engine now powering AI with the brave search API. You’re listening to a new episode of the brave technologist. And this one features Gilad Lombroso, who is squad lead at A121 labs, where he’s responsible for word tunes, AI writing features, managing engineering algorithms, and product teams.

[00:00:42] Gilad joined A121 labs almost six years ago as a sixth employee, and has since worked on many generative AI initiatives. In this episode, we discussed how to deal with hallucinations and misleading information. Ways that creators can use various AI tool to scale their work and improve efficiencies and how [00:01:00] natural language can improve usability within your products.

[00:01:03] And now for this week’s episode of the brave technologist. Why don’t we start off with kind of like setting the table a bit, you know, what’s your involvement with AI and kind of what are you building?

[00:01:16] And just kind of give the audience a bit of a little bit of background maybe on your current work. I’ve been working at AI21 for almost six years. And we’ve been working on generative AI before it became cool. It became cool recently. So yeah, IA21 is, you know, one of the leading companies that is, building both LLMs and LLM technologies from scratch and consumer products.

[00:01:44] So the flagship product is WordTune. This is what I’ve been working on for most of my time here. And, WordTune started off at 2020 already. Today it has, millions of, uh, monthly active users. This is, uh, what we’ve been working on and [00:02:00] me personally as well. What does WordTune do? WordTune is a tool that helps you basically create and consume content.

[00:02:10] So, it helps you articulate your thoughts and find the perfect phrasing for what you want to communicate. And it helps you read and scheme through content faster with helpful summaries. I would say that WordTune is just our interpretation of how AI could be harnessed. In order to help people solve problems with regards to, again, creating and consuming content.

[00:02:40] So when we started off again, it was really early days and we. I had many thoughts about how specifically how writing assistants should look like. And I would say it developed quite a lot, but the basic thing we’re trying to do was always and still is [00:03:00] how can we harness this technology? Where does it thrive and where does it fall and how we can allow users intuitively to get the benefits out of this technology?

[00:03:12] And there’s quite a lot of benefits you can get to your users. So this is working. This is what we’ve been trying to do. How does AI 21 labs kind of stand out from others in the space? I would say the fact that we build large language models from scratch. This is one thing that not a lot of companies are doing.

[00:03:33] Second is the fact that we both do. Large language model technology and building consumer applications for the end users. So having both in the same company really allows you to build the exact thing you think is best for the experience and for the product. With much more flexibility than you would otherwise, you know, have to, when you can only use available API APIs [00:04:00] and stuff like that, it’s just, you’re more limited.

[00:04:02] And when you only have the scientific technological perspective, then you’re missing the way to productize the technology. So having both in the same company really creates this synergy that I think is apparent. And honestly, the third thing I would just say that we’re being working on these problems for quite some time.

[00:04:24] So not a lot of companies have been here since, you know, I think six years I joined in the early days, but, you know, a 21 really has been working on these problems for quite some time. So we have. More context and more perspective that I think is important as well. Kind of touching on that consumer piece, because it’s super interesting.

[00:04:43] I get it. Cause I feel like we’re seeing more and more adoption and use on this broader consumer market now. Like how do you deal with hallucinations and, you know, misleading information and things like that as they come up in those consumer experiences? Are you guys having to change strategy at all as more and more [00:05:00] people are using the product?

[00:05:01] Yeah, so I’m not sure about changing strategy. I will say that this is one of the major challenges that we as an industry need to address. AI21 has been focusing on this area. We believe that this is one of the first things. We should improve and WordTune, for example, already tries to solve it by checking its generations that it provides to users, the textual generations with sources from the internet.

[00:05:30] So if it spots something that Contradicts some reliable source, then it won’t present it. And when, you know, claims in the generation requires some, uh, support, then it will present the user, the sources that support it. So this is one way of solving it, but there’s many possibilities for how you can solve this issue from the technological perspective and also from the product, making, [00:06:00] making it clear.

[00:06:01] When, uh, the models are just not sure, for example, is one way to go about it. And there’s many other ways you can try to solve it. So A21 really puts this front and center and, uh, things we want to address. That’s awesome the point you’re talking about earlier to just because we care a lot about creators of bravery we have a whole creator cohort of like you know almost two million of them that are verified with us and you know this is such a promising field for creators right how a i can impact their work and especially a lot of them are startups or self employed types of people are with small teams.

[00:06:36] You mentioned earlier that your software helps to solve problems for them with creating their work or consuming it. Can we drill down a little bit into some of those use cases? I think it might, might be interesting for folks from our creator community to hear about some of these ways that you guys help them out.

[00:06:50] Yeah, definitely. So first of all, we provide access to the large language models we build and even tasks, specific [00:07:00] APIs. Like rewriting and summarizing and you know, other capabilities that entrepreneurs and people who want to grapple with AI could use. So this is served through the studio platform, which provides us access to these technologies we’re building for organizations and entrepreneurs.

[00:07:19] Again, we have WordTune, which helps you, for example, create content from scratch by prompting it to write, for example, a press release that you want to publish and get help with finding just the right words you want to phrase your ideas with. So I would say one of the main components in WordTune that I think users love is the fact that it provides a lot of control.

[00:07:47] With which you can choose exactly what words you want to use for your sentence or paragraph. Sometimes you can even choose just a few words that you’re not sure about and it [00:08:00] will offer alternatives. Which really helps you just find this exact word you’ve been looking for. So I’ve been using it. Ever since we released it, to be honest, this feature specifically.

[00:08:13] So I really encourage people to go to virgin. com and see how it can help them. As well as, you know, studio for people who really want to have more control, um, for, you know, organizations and developers, especially that’s awesome. Yeah, I mean, because so much of this gets kind of stuck in this like prompt box, right?

[00:08:33] Of like, okay, this is what AI is, right? And these are really power tools for people that they can start to use. Like, I’ve been using them for all sorts of stuff to like with our regular work here and just knowing that those tools are available for people is awesome. Getting broader to you, like there’s always this kind of like doomerism takes on like concerns and worries around AI.

[00:08:53] Are there any like practical concerns that you have with what’s going on more broadly in the space that people might be [00:09:00] missing or just general things are either moving too fast or too slow or anything like that that you want to share? Definitely not that it’s moving too slow. That’s not a concern.

[00:09:10] Yeah, so, you know, people have been talking a lot about the. Ethical concerns and the pace of progress, which I agree entails, dangers that I think are just, something that people talk quite a lot about, I would say that something I saw as someone who’s been working on these technologies, people often don’t realize.

[00:09:33] Where it might fall and where it might offer value. People usually project human intelligence on this intelligence that emerge from the large language models. And it’s sometimes not so sometimes these models will perform amazing tasks successfully. And then you would expect them to sort of get it right with simpler [00:10:00] instructions.

[00:10:01] And they just fail miserably. And you’re like, I just saw you pull off this amazing thing. How come this is something you can’t do? And you know, the answer is that it’s not the same. It’s not the same intelligence as a human intelligence. It’s surprisingly close in many ways, but it’s still very, very far from human intelligence, obviously.

[00:10:22] And the ways in which it fails sometimes are counterintuitive. So I think this is something that I think, you know, for entrepreneurs. Might be confusing and might be misleading and how, you know, their expectations of what will work for their products or their use cases or their needs. And it’s something that you really understand once you play with it and go into the nuances.

[00:10:47] Where do you think that it’s been most successful so far and kind of what are the current gaps you see in capabilities? Yeah, I think that we’ve been seeing a lot of content creation tools, mostly for creative [00:11:00] purposes. I think it makes sense given that the technology is really good at creating something that looks good and sounds good.

[00:11:09] But once you really need to control the output, this is where you start seeing the limitations of it, right? So we all saw many examples of Tools for creating content for blog posts that you really not 100 percent care about the tiny details of it because you just want to create content that converts and works with SEO and stuff like that.

[00:11:32] Because again, like when you’re not really concerned about the actual details and every word you want to get right, this is where these like large language models sort of just. Hit the target in the first attempt, but I think that pushing the envelope here would be to make it more and more controllable, making it easier and easier to really get my thoughts and fulfilling my [00:12:00] intent, both for the entire piece I’m writing or entire research I’m, I’m now conducting and I need to scheme through these.

[00:12:08] Documents, or I want to write something from these documents, but I want to just nail that thing I want to accomplish here. So this is where I think we still have a lot of work to do. And so AI look, AI is. It’s amazing to put it simply, and we can harness it in many different ways. I just think that there’s still more work on the technology side, of course, but also on the product side to make it more accessible, make it more intuitive in the places that it’s still obscure.

[00:12:39] And to fill some of those gaps, do you see, do you see that through like more feedback loops with the product or with more training or where we’re, or a mix of both, like, is it just kind of in push all levers type of situation? That’s a very big question. We haven’t reached the ceiling of the technology.

[00:12:57] I think that’s something we can say [00:13:00] pretty positively. We don’t see any signs of slowing down with the technological. And I think on the other hand, even if you stop the technology today, there are just so many things you can do and so much things we need to figure out in terms of the product, in terms of the experience, in terms of the UI, even if you would stop the technology today, we still have a lot of unanswered questions there.

[00:13:26] And of course, as you alluded to, there’s this, you can use the user input and design the UX such that people provide feedback, and then you can. Retrain the models again and again and improve them further for specific tasks and in general. So, there’s just, so many things we can do and user feedback is one way to go about it.

[00:13:49] We are working tirelessly on just figuring out the best way to serve this technology to our users. There’s still only if you look at the product [00:14:00] side and the website, there’s a lot to figure out. And, you know, adjusting the models so they perform specific tasks that are a specific component of the larger feature.

[00:14:13] This is also something that I think we are very good at, at WordTune and at AI21. In general, I think it will be apparent in, you know, in the products we release. What’s the most impactful professional or personal example of AI you use in your day to day life or work? Like your top AI hack you would suggest to the audience or that might seed them into thinking of an interesting way of using this.

[00:14:42] Yeah, so I don’t know if you’ve heard about a product called Wartoon. I’m just kidding. Uh, I will say, I will say that, like I said earlier, I use WordTune and yeah, don’t really want to pat myself, uh, ourselves on the back here, but really I’m using WordTune often [00:15:00] and the amount of control I get with it is something that I don’t feel other products currently provide.

[00:15:06] So again, I can really just, for example, select three words. I’m not sure about a phrase. I’m not sure about, and it offers me the alternatives and it’s just, sometimes I just raise my hands and say, wow, like, it’s just like having this friend, this expert sitting next to me and say, this is what you really wanted to say.

[00:15:27] And it’s like, exactly. This is one hack I use with WordTune. Another example, I found that ChatGPT was helpful for me for learning. So I sometimes. Just, you know, when I need just, I have a few questions on a topic, I won’t start researching it on Google and I just want to figure it out with a few questions, questions I have in mind.

[00:15:52] So I found ChatGPT really helpful for that. Just, you know, I bounce off a few questions about some random [00:16:00] topic and it. Provides these answers. So for the basic sort of general knowledge level that sometimes I’m looking for this, uh, so chat GPT proved to be helpful. It’s awesome.

[00:16:13] It might be talking your book a little bit, but that’s good. One of the cool things about being early to something like this is that, you know, by the time it does kick in, like, and we kind of experienced this too, a brave a bit, like just being early on privacy and things like that, that when people start to actually care about this.

[00:16:27] You’ve got something that people can use that’s actually useful. You’re kind of like past a lot of that stage zero kind of, uh, really rough edges, right? Like, and so it’s kind of, I love that people can want people to try this out because I think, you know, the, the options, there are really cool things being built out there that people need to know about and a world of just like chat GPT or one or two things is like kind of boring.

[00:16:48] A lot of that nuance is what makes a lot of the field really cool. And so more broadly to like, where do you think AI will take us? First of all, we’re going to see more and more products that [00:17:00] provide powerful, intelligent, flexible tools, creative tools, something, a pair of words that might sound weird just a few years ago, but now it, yeah, it makes sense.

[00:17:14] You can provide tools that are creative, that are helpful. In flexible ways, I think in the near midterm, we’re going to see many more products and very different kind of products and what we’ve seen up until today, natural language is just going to play a significant part in all of this. It doesn’t mean that all UX and you, you know, all UX is going to be just a chat side panel.

[00:17:41] It’s probably going to be something more nuanced and more flexible and more integrated in the other flows you came to know. But it’s going to play a major part because I think natural language is just [00:18:00] something we all do all the time and we learn to do from an early age. And it’s just intuitive. this is the human way to specify what you want, but exactly, no, not like this, but like that you see this iterative process is exactly what.

[00:18:17] We used for, uh, as long as we existed to communicate. So the fact that we can communicate in similar ways with computers basically leads to the conclusion that products are just going to look different because you’re going to utilize natural language as the main interface. With humans, so this is definitely going to affect products in the, I would say near term because, you know, we can basically start working on it in the longer term.

[00:18:50] I really believe that eventually we’ll see robots and autonomous cars and other manifestations of AI in the physical [00:19:00] realm. And after that, you know, artificial general intelligence, which. People talk a lot about this point in time where the machines become as intelligent and more intelligent than humans.

[00:19:15] I don’t see a reason why we won’t get there, but you know, who knows? And especially who knows when. Right. If you could have a magic wand and get any like AI feature capability that you don’t have today that you really wish you had today, what would that be right now? It’s funny because basically what we’re asking ourselves every day at the office when we come to work, I don’t know something if there’s one thing that comes to mind again, it’s the question we ask every day.

[00:19:46] Uh, in order to build good products that are built on AI, but this is my, passion, I think to find ways where we harness this technology and provide an intuitive [00:20:00] and helpful way to harness these technologies for users. So this is really, like I would say, uh, the greatest passion we all have here at 21 and you know, it’s just, AI is being progressing so rapidly that, you know, the moment you wish for something, you get it and it’s like, this is the general feeling we’ve all been feeling for the last couple of months.

[00:20:27] So, yeah, I don’t have a one specific example for that. What’s the one myth about AI you wish people would stop believing? Yeah. So earlier I spoke about, uh, this policy of just projecting our intelligence on the AI, which is not the same intelligence. It’s different. It’s optimized and trained on the internet, which is some proxy for.

[00:20:52] Human communication and human intelligence, but it’s not, it’s just not the same. I would say there’s not this one [00:21:00] line that, you know, where computers are here and humans are here, and then they are advancing more and more in this one line until they reach us and then surpass us, it’s much more complicated.

[00:21:15] People don’t really know how to. Exactly define intelligence other than solving problems, but it can take many forms, so I think, you know, the Smith where there’s just intelligence and it’s one lane and someday. Machines will, will surpass us. It’s much more complicated than that. one example of it is how these models fail in unexpected and unintuitive ways, I think it’s a good example for that.

[00:21:45] A lot of the listeners here might be developers or folks that are familiar with building software, but aren’t familiar with AI and want to learn more. Are there any resources you’d recommend for somebody that has some technical background, but might be new to. [00:22:00] This field of AI, but wants to learn that you’d recommend.

[00:22:03] Yeah. So first of all, I would say that Twitter or X is a great resource. For, um, everything AI from, you know, the product, the, design, the technology as well. I would say Andrej Karpaty, who was the AI director at Tesla, is now working in OpenAI. He’s a brilliant, brilliant guy who speaks a lot about large language models and provide, uh, courses on the topic.

[00:22:34] So follow him. On Twitter or just see his videos. I would say, you know, Tesla and Elon Musk are really interesting players in that domain as well. I feel like it’s everywhere. Like if you really want to follow AI from the different perspectives. It’s really there i just read a lot of blog posts from different companies there’s not necessarily [00:23:00] one person or one company i would follow it’s just it’s all around look it up i change sources and you know speakers all the time that i listen to.

[00:23:09] Is there anything that we didn’t cover today that you want folks to know about i came from i would say the user experience. I studied computer science and I programmed a bit, but, I would say. My forte is a user experience and design. And for people who are looking to build products, I think that the human computer interaction domain is something that I’ve been reading a lot about.

[00:23:37] And I, I I’ve been working in at this and it’s just, it’s going to change from, you know, in radical ways in the next couple of years, so it’s just something that is. Other than being super fascinating to talk about in a podcast, it’s really something I think people, entrepreneurs [00:24:00] and product builders should be aware of.

[00:24:02] If natural language is the way we communicate as humans. And if now technology allows us to communicate similarly with computers, then it’s going to have an impact. So I would say that my number one recommendation for entrepreneurs and product builders and product designers. Is go learn about it and start looking for ways to figure it out.

[00:24:28] We won’t just pack everything inside. Chat interface. Dialogue is definitely something that is going to play a major, major part in products, but there will be times where you reach this point where you need to adjust things and you just need this thing you want to pull and make it work and you don’t necessarily want to ask for everything, type it in again and again, right?

[00:24:54] There’s specific things you want to surface for your users and, you know, make it easy for them to [00:25:00] choose. The exactly two or three options you want them to choose between. So there are cases where we need the classical UI. The places where you combine both and provide the flexibility and range of possibilities you can provide with a chat interface, or, you know, with the natural language more broadly and combining it and integrating it with the places where you want to limit your options for your users.

[00:25:29] You want to make something very clear. You want to make something very accessible. This is where the magic happens. And I think the product will see either, you know, one year from now or three years from now. Or tomorrow in today’s space, I would bet that these, products will be the products that combine both the classical UI and the natural language UI in the best way for the given use case.

[00:25:57] This is, I think we’re product builders [00:26:00] and designers should focus and, you know, invest time in learning and, and just grappling with everything that has been, uh, Happening for the recent years. Yeah. I continue talking about it for hours. It’s, it’s really fascinating. And again, like, I think as someone who has been watching the progress for more than, you know, like, uh, five years, it’s really, really amazing to see just how fast everything is progressing.

[00:26:29] So it’s exciting. It’s really exciting. And, and I think we’re going to see a lot of interesting stuff. In the next couple of months, years, and you know, beyond. Oh, we’re excited to see it. And, and also we hope you to have you back to, to, um, cover any updates or anything like that, that you guys have going on at a 21 to you, I think it’d be really, really great.

[00:26:49] Yeah. I’d be happy to do that. Yeah. For folks that might want to follow you, like where can people find you online and, find your guys’s products too. Mostly on LinkedIn. I think I’m more of a consumer [00:27:00] in Twitter and not really creator. So yeah, LinkedIn, I would say.

[00:27:04] Really appreciate you joining us today and, like to love to have you back to, uh, to give us some updates

[00:27:08] and, uh, thank you so much for the insight. This has been super interesting. Great. Thanks. It was a pleasure and uh, yeah, I hope to be back here soon. Thanks for listening to the Brave Technologist Podcast.

[00:27:20] To never miss an episode, make sure you hit follow in your podcast app. If you haven’t already made the switch to the Brave Browser, you can download it for free today@brave.com and start using Brave Search, which enables you to search the web privately. Brave also shields you from the ads, trackers and other creepy stuff following you across the web.

[00:27:41] ​