Speaker 1  0:43  
Hello and welcome to today's autumn webinar, telling your AI tools, protect, transfer, transformation. My name is Don vo young, and I'm a member of Autumn's educate team and today's staff host. All lines have been muted to ensure high quality audio, and today's session is being recorded. If you have a question for the presenters, we encourage you to use the Q and A feature on your zoom toolbar. Should you need closed captioning during today's session, the Zoom live transcript feature is turned on and available on your toolbar. Before we begin, I would like to acknowledge and thanks Autumn's online professional development sponsor, Marshall Gerstein, we appreciate your ongoing support. And now I would like to welcome today's presenters, Devorah gracer And Peter Bittner. We're excited from to learn from you today you Well, I can go ahead and take it away, yeah,

Unknown Speaker  1:47  
thanks for having us.

Speaker 2  1:55  
Would you like us to take it away, Don, V, L and get get it away? Great. Yeah. Happy to introduce myself really briefly. I have a quick slider to hear and can just jump in and get started. It's wonderful to have you all with us here. Thank you for your interest in this. You know, there's been such a frenzy with all these new AI tools, and it's really a lot to process, even for those of us who are really, you know, daily in sort of the trenches with this. My background is as an instructor in new media at UC Berkeley's Graduate School of Journalism. Originally, I was a multimedia journalist and a correspondent overseas. My background has sort of morphed and taken me into marketing and communications as a consultant since then, and I've been teaching AI at UC Berkeley for the last three or four years. I've been a lecturer there since 2017 and you know, the last couple years in particular, there's been just a flood of interest in my courses, as well as consulting. And so the upgrade is my brand that I've begun to really accommodate all this, you know, outside work, and really pivot towards, you know, helping people bridge the digital skills gap when it comes to these, you know, amazing new tools that are coming out and and Devorah, I want to make sure you have a chance to introduce yourself as well before we jump into the content itself.

Speaker 3  3:29  
Sure. I'll introduce myself briefly. I have a slide farther down. We'll get to it later. So I'm Devorah grazer, founder and CEO of rocket smart rocketing your IP out of the university and into a great licensing deal. And I have been interested in programming since I was 16, but my PhD is in pharmacology, so I run the gamut from hardware software all the way through to the life sciences. And I'm really excited to be here today. So with you, Peter, and also with everyone at autumn, it's looks like we have a great audience, so I'll be excited to get your questions.

Speaker 2  3:59  
Yeah, absolutely please. Do you know? Let us know in the chat. If you have questions. We have a dedicated time for that at the end the last 15 to 20 minutes or so. But as they come up, feel free to pop them into the Questions portion here on Zoom, and we'll be able to address them and make sure that, yeah, they get into the queue here. So what is generative AI, right? Generative AI is really, you know, branch of AI that has come into its own in the last 18 to 24 months, right? This is the type of, you know, specific sub branch of machine learning that deals with neural networks, typically, which are kind of the backbone for all of these new products, like chat, GPT, like Cloud, you know, there's metas llama, right? There's so many different AI models that are coming out that are useful for all types of different tasks, right? And now. And quantitative reasoning, as well as generating, you know, not just text blog posts, for example, but image generation, video right? This multi modal sort of ability to transfer content from one format to another really seamlessly, is something that's super exciting and really on the cutting edge of a lot of these models here. They're capable of producing a huge diversity of outputs, from creating, as I mentioned, images to music and voice. There's so many different applications for these tools and so many different companies springing up that are often, you know, based on these core models. You know, there's additional products that are built using them to solve specific types of, you know, business solutions for individual professionals. We'll go into a few of those later here. But just to kind of provide a bit of a background, you know, AI has been around for decades, right? We've been using AI in our, you know, consumer products for well over a decade or two, right? In many cases, without really knowing about it. But it's this generative AI revolution that we're in the middle of here that is, you know, really drawing all the attention, and it represents a big change. Is all kind of mentioned here, right? This is a really remarkable period of disruption that we're in. You know, tech love sees the word disruption, but this is truly what's happening right in the job market, in the workplace, in sort of the entire economy here. And with that, you know, there's a lot of not only just media that are being impacted, as I alluded to earlier, right work on the content creation side, but there's also, you know, all of these important considerations that are being really, sort of, you know, teased out and real time. So there's a lot of issues when it comes to things like ethics of using these you know, things like legal questions. There's a lot of ongoing litigation, right? Just the other day, you know, eight major newspapers joined the New York Times in, you know, a copyright lawsuit against open AI, the maker of chatgpt, right? And so this is an ongoing, evolving, new realm that we're, you know, the tech is moving faster than society and sort of the legal context and policy can keep up with, right? And the market for a lot of these types of roles that involve any sort of digital what we call white collar jobs and the information economy are changing dramatically. I try to think of this as an era of opportunity as well, to sort of reinvigorate the work that we're doing. I don't personally like to think of it as something that's going to suddenly put everyone out of work. I don't think that's the case. I think what it will allow people to do is actually spend more time on the meaningful aspects of their job and automate part of the drudgery and daily sort of grind work that they're not super excited about, and also enhance the parts of their jobs, right through AI assistance and, you know, strategy coaches, etc, that are, you know, allowing them to do that higher level work even better, too. So, you know, there's a lot going on here. Just a few trends before we jump into some of the tools here. There's a really big shift towards, you know, this multimodal media. I mentioned that earlier, but I just want to highlight that this is, you know, a really exciting development, and a lot of these tools, for example, Sora, which will be launched later this year publicly, it's the open AI's video generator that simply takes text, turns it into video, right? This is something that is was unimaginable, really, even two to three years ago, in a way that would be ready for public consumption. And a lot of times they're not fully ready for public consumption, right? These tools also, just want to stress, are being pumped out at a rate in which, you know, there's Yeah, kind of caution to the winds when it comes to things like policy, legal ethics. So we need to be really sort of educated consumers and do our best to make sure that we are, you know, representing our universities, our you know, yeah, researchers to the highest level of professionalism, and it's tough to sort through all of this, right, but we are essentially being inundated with all these new tools, and this online media environment is changing so rapidly that it's really critical to kind of take a deep breath pause and think. Critically hear about some of these tools and their uses. A few other quick trends involve some of the abilities to have personalized media and advertising campaigns and marketing, being able to have tailored content and being able to take types of for example, you know, patents or other sort of documentation, and quickly turn that into, you know, a personalized product, whether that is, you know, a media campaign, it might be, you know, a specific type of document or grant proposal, even, or things of that nature, that is specifically, you know, created an automated way to address one parties, right, or target audiences, needs, values, perspectives, priorities, and that's pretty cool with this is also deep fake technology, right? That we're running into. You probably have seen this, you know, early 2023, viral image of Pope Francis in a puffer jacket. Well, as we know, it wasn't actually a real photo, but it took the web by storm. And you know, we've seen a variety of more nefarious use cases since then, in particular, deep fake porn and things of that nature, right? Taylor Swift being the highest profile victim of this targeted online harassment. So this is there is a darker side to this, right? There's a number of things that are changing that we really need to be aware of. Search itself with this rise of content that's AI generated is going to cause a lot of issues in terms of trying to surface and sort through to find those real, human generated articles now, right, that aren't just sort of AI clones of other websites packed with keywords, right? It's a it's a real challenge to cut through the noise as a marketer,

Speaker 2  12:02  
AI and journalism and communications is changing dramatically, right? There's, you know, possibilities of consolidation, and you know, some integrations with AI tools to serve these targeted results you know, that are built on reporting and real human labor, right? And and really automate that. And there's also, you know, some other risks here. We're not going to get into a lot of those today, but I just want to really acknowledge those as you know, part of this really complex environment that we're all you know, in 2024 as professionals here having to navigate right? So a few quick things, privacy and data security. Are some things in your daily work that you really need to be thinking about and you know, really being educated on your university's policies and making sure that you understand some of the ways in which you know these AI tools may put you or your university employer at risk, right? Also thinking about, yeah, ethical Best Practices Are there guidelines and norms that your team, your organization has making sure, also in using these tools, you don't lose out on that authenticity, right? Trying to think about how to be transparent in the use of these is important being accountability, right? There's Yeah, ultimately, the buck stops with the person using those tools, whether it's, you know, a marketing agency, a PR firm, an individual team at your organization, right? You need to make sure to own this and understand that, you know, there may be some impacts here and fairness and bias. We also need to realize, really importantly, that a lot of these models are trained on massive data sets and troves of, you know, scraped articles from or, you know, forums from every corner of the internet, and it's a reflection of society and all of its biases, stereotypes and, you know, inequity. So it's really important to think about representation and about, you know, some of the products of these AI models and tools, taking that with a big grain of salt and thinking about, how do we make sure that we do reflect, you know, equity in our work, great. So just really generally, a few ways that marketers are using AI Well, I'm curious to hear if any of you are using chat GPT on a daily basis, using sort of these raw AI models. Is kind of part of the first phase of AI adoption that you know began early last year, even a little bit earlier, with these chat bots, right? Like chat G, B, T, Claude. There's many, many others out there now, and you know, it requires some prompt engineering skills, right? It requires kind of a different skill set than we're used to typically. And sort of as part of this, there's been a variety of user interfaces, you know, companies that have built these specific basically on products that are on top of the raw and models to help solve specific challenges related to things like copywriting or marketing, right? Jasper is a very popular one for marketing. Grammarly, many of us probably use right is wonderful for, you know, checking our work, making sure we don't have too many typos going on. There's also a couple products, specifically in this space and sector scale, IP and first, Ignite. I'm not an expert in tech transfer specifically, Devorah can speak to this a lot more than I can, but there's a variety of new products that are being built that will save you the work of learning the prompt engineering and having to be an expert on this. And there's even further, sort of the third wave of this product evolution that we're seeing in this market is rag enabled, AI assistance. What? What is rag? Rag stands for retrieval, augmented generation. It's a fancy term to describe, you know, how you can use your own proprietary data to generate better quality, customized outputs, right? I have a quick demo and prepared for what that looks like here that I'll show later, but you're able to essentially input your own library of resources from your team. Maybe it's, for example, an entire database of previous marketing materials that align with your university's brand standards. Maybe it's the brand guidelines themselves. Using these as part of, you know, the actual output process is something that's now possible. And what are the advantages of using these type of techniques, really having, you know, much better, enhanced understanding, you know, the AI models that you're using to be able to have better, more accurate context, aware and information rich responses, right? Also able to, you know, make sure that you have, for example, yeah, better applications across these diverse domains, you can, for example, have much better tools at your fingertips to summarize and integrate massive troves of information and even create custom chat bot assistance trained on your own proprietary data to be able to serve you and your team better than even something like a jasper or something that is a bit of a more generic product. You know, having these specialized assistance is really the power of using these rag type applications and techniques, right, and being able to scale the knowledge integration having, you know, live continuously updated database that sort of powers these rag assistants. This is part of what powers the next generation of AI agents, which is part of this evolution we're seeing in the marketplace. Right? So, what is an AI agent? It's a system that is capable of making decisions and acting on them to achieve specific goals. Sounds a bit scary at first, right? But it's also incredibly powerful. And designing these isn't as hard as you may think. To be able to have them built upon or built into your own particular you know, data sets, your own software stack, being able to integrate with your own suite of tools. And you know, whatever type of software you're using as your CRM, as your CMS, it's possible now to build in AI agents and assistants and automations that fit seamlessly within your current daily workflows and those of your team which is unlocking just a entirely different order of magnitude of things like productivity, accuracy, streamlining. It's just really the frontier of what's happening. This is the year where we're seeing enterprise AI with these AI agents, with, you know, these retrieval, augmented generation techniques starting to be used in real life, in enterprises and organizations of all sides sizes. Excuse me, there's a number of different platforms that have become quite popular to build these. And there's a huge variety of AI agencies that are now popping up to help organizations create these. So mind Studio is a great product for building these types of you know, rag powered, customizable AI agents, even the ability to have. Multiple AI agents working powered on multiple data sets, potentially built on numerous AI models at once, having the ability to, you know, have GPT four and cloud three and metas code llama, for example, all working for you in the background. To, for example, stand up a new website just for your you know, project and and specific goals seamlessly. These are the things that are suddenly in the realm of possibility that can save a lot of time and, you know, generate more better, higher quality results with the same type of you know, team you already have, right? Okay, so I'm going to send it over to Devorah. Great.

Speaker 3  20:48  
Thanks, Peter. Thanks for the introduction. That was really interesting. So I'm going to be talking about generative AI, could it be your personal assistant? Now, I asked the dolly, which is the image generator from chatgpt to make an image of someone who, where someone who's getting help from a cute robot for doing business. Now you can see we have a bit of the stereotype going. Who is the person doing the business? I thought the robot was cute, however generative. AI isn't quite there yet. So while it can help you, it's not going to be a cute robot sitting on your desk and helping you out. Next slide please. So this is me, as I already mentioned before, I've been programming since I was 16, founder and CEO of rocket smart. We do AI powered IP licensing. We also do AI consultancies. So you can reach out to us if you have aI questions. Let's go on to the next one. Alright, so I'm sorry, folks, see three Po, it's not available. So I'm dating myself. I was quite young when the first Star Wars movies came out, and see three Pro was great, and I wanted to see three Po. It's actually one of the things that got me interested in programming, but generative AI, isn't there yet. So while it can help you quite a bit, you're not going to get the kind of real human interactions that you got with this character, which is, after all, powered by a human in a suit wandering around in the desert sun and actually speaking the lines. Let's go on to the next one. All right. So some tips to remember generative AI is a prediction machine. What does that mean? You put in your question, your request, whatever it is you want, and it predicts each word of the answer in order, as a result, because it's going on statistics, sometimes you get really wildly wrong results. But even if the results aren't wildly wrong, it doesn't understand the difference between true and false. It doesn't understand the difference between fact and fiction. In fact, I think it's already was a couple years ago. Now, there was the case of the Microsoft AI that tried to convince a New York reporter to lead his wife and to be in love with the AI. Now you could say, How is, how did they program the generative AI to do that they didn't it learned from everything scraped on the internet. And if you remember movies like her and other movies that feature a sentient robot that has a romantic relationship with a human being, that is where it learned those tropes from, and it kind of went off the deep end. Now, our particular company policy you of course, should develop your own, but our particular company policy is we only use these tools with publicly available information. Here, I'm talking about the public generative AIS, the kind I'll be demoing today. Chat GPT from open AI cloud from anthropic, Google, Gemini, etc. This is because you don't want your data to end up being used to train someone else's tool. All right. Personalization, generative AI tools. There are some issues. You can do the personalization, but it's not built in. These tools were trained on generic text and generic information, so they can struggle with specialty areas. So for example, I once gave one of the was, I think it was chat GPT, a patent on quantum computing, and it actually did okay, but then Claude struggled. Why is that? Because chat G, P, T and Google Gemini were both trained on Google patents, and so they have a lot of patent input, so they were able to handle the specialty areas. However, not all chat, not all generative AIS, are going to be able to do that. They also tend to get kind of over excited, if you're talking about social media. So they'll put a lot of emojis in for LinkedIn, and you'll have this sort of bland corporate speak for everything else that's their default. That is because if they don't know what it is that you want, they're going to go for what they think the average is, what is the average expectation. So when we're talking about personalizing, what we're trying to do with these tools is to use them in a way that we make it clear to the generative AI itself. The kind of output that you're expecting, right? It's gone. Or here's an example you want to create a LinkedIn post. It's about one of your published invention disclosures. So this is the text. I actually took it from an actual university invention disclosure that was published, publicly available on the internet. If you search this text, you'll probably find the disclosure itself, but I will not name who did it. And it's interesting that it's very technical. So we find out that reactive oxygen species and oxidative stress can cause IBD inflammatory bowel disease. And they have discovered we that is lactic acid bacteria that can neutralize these art West, and it can help to facilitate maintenance of healthy oxidative levels in the intestine and modulate inflammatory conditions. So this is actually written by a tech transfer office published on the internet to interest life potential licensees in licensing this product. So I want to use chat, G, P, T, to write the LinkedIn post. So if we go on to the next slide, my here's my prompt. Please write a short LinkedIn post about this new invention, and then the previous text that I gave you I would copy and paste in. So first of all. Why did I write this? Well, I want to make it clear it's LinkedIn. Twitter is different. Facebook is different. If I say social media, I'm not entirely certain what I'd end up with, but it might not be suitable for LinkedIn. Why am I saying short? Because these generative AIS tend to go on at length. This is supposed to be short. It's actually kind of long. You can ask it to shorten it, but if you don't ask for short, it's going to end up being much, much longer. And why did I say please? There has been some research that shows that saying please and thank you, and in some cases, even offering to pay the generative AI tool like chat GPT, will enable you to get a better output. Now, joking aside, no one's quite certain why that would be. It may be that in trying to mimic human language and human conversation, it's also mimicking human motivation. But for whatever reason, I had this prompt followed by the invention disclosure. And here's my result. I have two emojis. I think they're supposed to be microscopes. We're thrilled to share a groundbreaking discovery in the battle against oxidative stress. All right, you can already see it's very technical. Our research has unveiled a lactic acid bacteria. The bacteria not only helps modulate inflammatory conditions, but also helps aids in maintaining healthy oxidative levels. So this is almost like a copy of what I put in. Stay tuned for more updates. I'm not sure why it wrote that, but that's what it thought we should end with. And then about a gazillion hashtags. You should not have this many hashtags and LinkedIn posts. You should have, you know, three to five at the most. But chat GPT always goes overboard on the hashtags. All right, so let's go on to the next slide. Let's suppose that was boring. So I say, please add more emojis and make it catch here. That is actually the, the the prompt that I wrote. Now you need to be careful that it understands add more emojis to what make what catch here. So in the full prompt I wrote, please add more made emojis and make it catch here for the LinkedIn post that you just wrote. Otherwise it may get confused and may try to add emojis to my original disclosure or to something else. So here's now. What I have is, you can see there are just emojis galore. It got very excited. It is a bit punchier in terms of the language. It's a bit shorter. It's a little bit less technical because I asked for the emojis, and also because I asked to make it catch here, exciting breakthrough, discovering new frontiers. It still takes a little while to figure out what the invention is actually about, and it's still a bit generic. And of course, you still have too many hashtags that's gone so then I said, Please remove the emojis, add bullet points and make it more formal. I also asked it to add bold for the first word or phrase of each bullet point. And so now I have a much calmer post. There are no emojis. There are some nice bullet points, innovative discovery, key development, health implications, not necessarily exciting as a LinkedIn post. It's still kind of generic, like, why would people be excited about this, I would ask, but it is certainly calmer, still, too many hashtags, but it's also gotten a little shorter. So this is something you could consider posting if you want to, but, and there's a big but for all of this, let's go on to the next slide.

Speaker 3  29:45  
Ah, before we get to the big but, sorry, I forgot the problems with formatting. When you copy paste something from chat, G, P, T, they have a little copy button at the bottom. It's at the bottom of the prompt, um, you'll see it during the demo. Little hard to see all of the gender. Of AIS have this. Claude has it. Google Gemini has it. The problem can be when you copy that text and you paste it onto LinkedIn, you paste it into a Word document, you paste it into a PowerPoint, you can end up with problems with the formatting. So instead of having bold, it'll have the two double asterisks on either side of the word that is to be bolded. Instead of having a larger title size font, it'll have multiple hash hash marks to indicate that. So some ways to solve that are just simply to fix it after it's in, which is usually what I do. You can use something called a Markdown editor. There are Chrome extensions that that support proper copying. So how does this work? You have chat GPT in your Chrome web browser, you have one of these chrome extensions activate that maintain the formatting you do your prompt with chat GPT. That response then goes through the Chrome extension and has nice formatting that you can then copy paste. Just remember, though, that everything that's kind of being put into and out of chatgpt is now going through, not only chatgpt, but also this extra extension. You have to decide if you trust that. I've had the formatting problem with all of these different ones, but I've only seen a dedicated solution for chatgpt. Now, Peter, you had one that you said worked for you, that didn't work for me. What is it that worked for you? Yeah,

Speaker 2  31:22  
so these things are so finicky, right? If you're using just kind of the the raw chat, G, P, T, occasionally, you can coax it, you know, sometimes through emotional blackmail, or, you know, yeah, word wizardry to to do this and actually format it itself in Markdown in a way that, you know, makes it really easily copyable over into Google Doc or Word doc. But you know, this is also part of some of the cool part of these AI agents, right? Is the actual ability to potentially just have the result piped over in the background to directly into your Google Docs, directly into, you know, your team's Yeah, software that they're using, but, but anyway, yeah, this is certainly a challenge, right, to actually make sure these results are usable.

Speaker 3  32:14  
Great. Alright, let's go into the next slide. Okay, so all of this is not the main problem. These are tips and things that you need to remember. They're very helpful, but the main problem is that you need to actually take very specific actions to get the output you want. Now I hinted at it a bit, but I want to show you a few slides that actually have me going through this process, and you'll see how the result is refined. Let's go on to the next slide. So first of all, why are you posting on LinkedIn? There should be a why? There needs to be a focus for your message. And you have to let chat, G, P, T, or clog, or whatever generative AI or AI you're using. You have to let that software know what your goal is. For example, is your goal to attract potential licensees? Do you want to raise general awareness? Maybe you want to get your professors to give you more disclosures to your office? You want to bring more students, or if you're looking for employees, the point is, the LinkedIn post, every message, every content piece that is generated through generative AI, needs to have a focus and a goal. So in and you have to tell that to the generative AI. So in this case, I decided to go with attracting potential licensees. And what I did then in the next slide is I said, Okay, I want to attract potential licensees. I'm going to tell this to chat, GPT, but what other information is helpful? Well, who is the potential licensee, industry, market within the industry, type of company, location, size, person within that company. What are their main challenges, and how are those challenges solved by your invention? You may know, if you're used to doing marketing, you may note that this actually looks an awful lot like a buyer persona, and there's a reason for that. If you can feed the information that is along the lines of of a buyer persona, who the details, the industry, the market, the company, the challenges, how you solve them, you will get a much better content output. And it and it doesn't matter if it's a LinkedIn post, a blog post, or a specific message that you want to send to a potential licensee. Now let's say you don't know that information. You have no idea what the industry could be. You don't know about the market, and you certainly don't know a company could be interested. You can actually input your industry disclosure into chat GPT and ask these questions again, I recommend asking them an order. First of all, don't ask who the potential licensee could be in my in my experience, that's too general. Ask what industry could benefit from this invention. What is the market within that industry? What about the type of company us? Company? Maybe you want to limit on size. Maybe ask chat GPT to do that, etc, etc. Each of these small bullet points, I see someone asking a question, can you share a tech transfer office as an example using this and getting really good results? Well, you know, I'm going to do a demo that hopefully will help answer that question, and you can actually get really good results. The point, though, is, if you ask these questions one by one industry, market in the industry, etc, if you break it down, you can actually get this information from chat GPT itself, even if you don't know it, but you do need this information to get a good result. This is your buyer persona. Chat GPT needs it as much as you as a tech transfer professional needed to write a good message. Okay, it's gone all right. So here's my detailed prompt, please rewrite the LinkedIn post again. Please. I'm saying rewrite. That's a specific cue, so it's going to kind of check out what it did before and rethink it the LinkedIn post. You got to tell it what you want it to rewrite, to appeal to a director of r, d at a company which develops probiotics for over the counter sale. Now I got this, this information through using chat, G, P, T. You could also get it from your researcher, or you might have industry knowledge that would enable you to know that this is a good target. I've specified this particular company. I want it to be in the US. That's because of Bay dole, with more than 10 employees, fewer than 100 you could make it smaller or larger, depend. You could also try different variations on this. The company is focused on probiotics for managing IBD and other inflammatory gut conditions. Why? Because this is the bacteria which we're going to stick into the gut, and we want to do something there. The company is active. Development with news announcements about its new products. Why did I specify that? How is chat GPT going to know if the company has an active R and D effort in your area, or any active R D effort? Maybe they just sell their distributor. They sell stuff from other people, you need to target the folks who are actually doing R D. So you need to specify that. How will it know about R D? News announcement. News announcements about new products. That is the key I'm giving to chat, G, P, T, to show it how to get the information that I need for it to use to write the good message. Now, if this sounds a bit confusing, first of all, I believe the PowerPoint will be shared afterward in a copy, paste, double version. You can also hit me up after to get, like a more detailed demo. I can show this to you, but it's really building this up step by step within chat, GPT or another generative AI, or using other sources for your information. Okay, that's gone all right. I'm going to compare the new response to the old. Now I've only copied the bullet points, not the whole response, because, again, it's a bit wordy, and those hashtags are making me crazy. But let's just let me start by looking at the bullet points in the new response. We have a novel lactic acid bacterium, enhanced gut health management. That is what our buyer persona cares about gut health management, because they make probiotics for gut health potential for new probiotic products. Aha. This is hitting all their keywords. Now you can decide if you want to go in and add hashtags here and there to this, but the point is, this is much more likely to catch the attention of someone into someone who is interested in your invention for their market. What was the old response? Innovative discovery, key development, health implications, very bland, very generic. This is what chat GPT defaults to, because that's what it thinks you want. If you don't give it a lot of detailed information, you're going to get the bland instead of the super exciting. Now, on the left, it also says, we're open to collaborations, promising avenue for differentiation and growth. Again, hitting those aspects of interest, making your target more interested, making your buyer persona, the person reading this thing, wow, I want to talk to this university. This is actually pretty cool, whereas in the old response, it's just stay tuned for updates. Okay, there's no call to action. Do you want a collaboration or not? I don't know. You can't tell from this message. So you need to be super specific about what you want your bio persona to do to get there. You have to be super specific with chat GPT. All right, let's go on. We're getting down to the end. We're almost to the demos. Folks. I then asked chatgpt, how could this invention help such a company? And I asked that after I actually wrote the whole LinkedIn series of requests. Why did I do that? Because sometimes it can be help. It can help you to start generating content and then backtrack. You know what? I want to check to see if chatgpt really has a firm grasp and wedding what I'm asking it to do. So I'll ask it a question that I could have asked earlier, but if I didn't let me ask it now to make sure we're all on the same page. Well, it could help with product differentiation, scientific credibility, new partnerships and collaborations. Okay? So. Chatgpt did understand what I wanted. I did this kind of backtrack question to check, and now I can go forward with more confidence. Let's continue on now. This is what the actual screenshot looks like, of just the part with those bullet points. You can see there's, like, a lot more words. This is the last question I asked. How could the invention help the company. This is the response I get. You can ask for links, which it won't always do, but sometimes it will. Otherwise you might have to go check yourself. Sometimes Google Gemini is better about giving links. Sometimes Claude feels like giving links and sometimes not. Or you can use a tool called perplexity.ai which is actually designed for search and giving links. Why do you want links? Because you got to go check all this yourself. Remember the whole hallucination, the AI falling in love with the human kind of a business. That is because the ad doesn't really know truth from fiction. It's just doing a statistical prediction, even this, which looks like really, a really good response, statistics, all statistics, and then someone suggested it can also be helpful to acid generate. Do you understand what I said to make sure? Yes, do I do you understand what I said? Or what do you think I'm asking for? What do you think I will This is an indirect way of doing it, but totally love that point. You can also ask directly. And I think let's see if anyone has any questions, because I believe after this we finish the slide, maybe you want to go on to like, the last slide for just one second. Peter also has his contact details in, and when we send it out, everyone can see it. Okay, the reason, yep, there's like, there's also your stuff as well. Very few models actually provide for users ownership of the data non ingestion. So there's chat, GPT Pro, there's Claude claims it doesn't ingest the data, and that's in their terms of service. It depends if you trust or believe them or not. This is part of the issue, right? And Peter's actually going to type an answer. Ah, that's totally awesome. Now another point that was Peter pointed out is you can actually and he's going to show you how to do this in a moment after I do my personalization of general chat GPT,

Speaker 3  42:10  
there are ways that you can actually use these tools to make them more personalized, but if you're using something like mind Studio, you do have more control over what The models are doing and also which models you're using, you can use open source models that are not presumably going to pump data back anywhere. You do again. Have to check where did it come from and do you trust them. This is what it kind of boils down to. I mean, look, if you think about this, in the internet world, everyone is being spied on all the time. Our phones can tell everyone where we were at all times, and our information is kind of smeared all over the place. So this is actually a problem that has been asked before, but now we have to deal with it on steroids.

Unknown Speaker  42:53  
Yep, exactly.

Speaker 3  42:55  
And, oh, you've been answering all the question. Wow. Really great. Is there anything else you want to hit on the questions are on this one, Peter, before we go into the demo,

Speaker 2  43:04  
yeah, I just wanted to say, you know, that I think it probably is the most secure way currently to use these tools for things like data analysis, right regarding sensitive information or confidential, you know, disclosures, etc, to use the API directly using, yeah, that sounds very technical, but basically to plug in and patch in the power of an AI model into, you know, another type of product, like a mind studio, etc, that allows you to do a couple things. Right mind, studio servers, for example, or, you know, I'm not sure about other products I help with the education on that front and certification for that specific, you know, AI agent and assistant creation platform. And the way they work is they have, you know, separate enterprise level accounts with secure, encrypted cloud, you know, for all of those uploaded, uploaded, proprietary data and documents, and they're not stored on open AI servers, right? They're, you know, in the AWS, you know, cloud, for example, or Azure, wherever at, you know, a more reputable, like established company in their Yeah, servers. So that's something to consider. Sounds very complicated. It's not necessarily. It's just another layer of security. And I do recommend, yeah, looking into that specifically, and into your universities specific policies and vendors that they prefer to use, right? You

Speaker 3  44:44  
can also set it up with the private cloud. So again, that sounds super technical, but it's not. What this means is you would have specific processors in memory which are dedicated for you, which only you can access. It could be through AWS, Amazon, Google, Microsoft, whatever, and then you. Actually have the open source models being run on your own processors with your own data, and then you control everything, but you need to have a lot more technical help. You need a really good IT department, and it takes a lot more work. So the more control you want, the more you have to have technical expertise to bring that control to life,

Speaker 2  45:17  
right? Yeah, these are all great questions, yeah, trying to do our best to answer them here in the chat as well as in the Q. A Devorah, do you have a demo that you wanted to do

Speaker 3  45:27  
here? Yeah, I have a demo. So why don't you stop sharing your screen, and I will share mine. Let me just, first of all make sure we're looking at the same thing. By the way, folks, you can keep on asking questions as we go, and we can always stop the demo or stop something please do, because we love answering questions. All right, so I wanted an invention to be able to put in as a demo. I wanted a description of invention that been published. And I asked, what is the wildest new invention that was developed by University of the past year in the US? I'm using perplexity.ai. I have a free account. You can also use it anonymously, and it gets into the spirit of it the wildest invention. And, oh, look, here's a one. This brings you to a link which I have to stop and reshare, sadly, each time because zoom and I don't always see eye to eye. And this is the link when I clicked on that number one in perplexity. And the link, I came to this 2023 the brink, pioneering research from Boston University. And I scrolled down and I saw a bionic pancreas first developed into bu lab. And it's a nice paragraph. It's really interesting. So I said, Okay, what I can literally do is copy paste this. So I'm going to stop sharing for a moment, because I have the prompt ready to go. Figure out where I put it. There we go. All right, I'm going to copy and paste the prompt I already put in. Okay, let me just grab this really quickly. I'm

Speaker 3  47:04  
all right, now we'll put this in, get over here, and I will reshare my screen. All right. So we're in chat, GPT four. This is the paid version. This is not, however, the pro where they keep everything super secret. You, in theory, can train can turn off the data ingestion, but, you know, have it enter by everywhere. I'm not quite sure what happens then. Here's my prompt. Let me just scroll up a bit so you can see it. You are an R D director for a life science company that manufactures implanted devices for chronic disease treatment. I am asking chat, G, P, T, to step into the shoes of my buyer persona. This is the licenses. This licensee I want to interest. How could I interest you? Right? So now chappy T is the you, the you, the buyer persona in the invention described in the following text, what problems would it solve for you? Now this text, I just will be copied and pasted that whole paragraph from the Boston University article. You could also copy and paste in an abstract of a patent. You could upload an entire patent. You could copy and paste the paper or or upload a paper. You can basically put in whatever information that you want. So I've asked my question. I've asked chatgpt to become the buyer persona. That's part of the personalization. This is personalization on the fly within the prompt engineering, and I'm going to send the message, all right, let's see what chat GPT thinks as the R and D director. Ah, very interesting strategic collaboration or licensing expansion into adjacent markets. Now this is already interesting. If the company is already focused on other kinds of chronic diseases, but not diabetes, maybe you could interest them in going into diabetes, because there are various insulin pumps that have been developed. Obviously they're different from this invention. There's already competition. Maybe you can interest the competition in doing this. So we give some good things here, and then I have to actually see if I can scroll back up. Goodness. It's gotten very excited. All right. Okay, so first of all, if you want to copy, do you see down here? You can copy. If you click on this, it'll read aloud. If you don't like it, you can ask it to regenerate on the flight. If you don't like it there. You can also say bad response. Now this is the fun part. I always have to find where the heck the Okay, good. I found the cursor there at the right, all right, so you're an R and D director. Wow. This is a compelling here's some key areas, strategic collaboration, expansion into adjacent markets, technical advancement, competitive differentiation, improvement of patient compliance, very good. Now, in terms of problems, this could solve your company, it addresses several critical challenges, reducing complexity of self management, enhancing precision and treatment, continuous monitoring. This already looks pretty good, so I'm. Saying, Okay, this is a good start. I have some of the information that I need. What next would I ask it to do? Well, I pre typed this just simply, so you don't have to sit there and watch me type, which is about as exciting as watching grass grow. All right. So then I said, Okay, chat. GPT, this looks very good. Here's my next question, which US companies fit this profile that are interested in already in treating diabetes. Now, you may remember, it had pointed out that if they weren't already in diabetes, and if there are other adjacent areas, they might want to move into it, but I want to know about companies that are actually interested in treating diabetes. Now, by the way, this process is fairly genetic, generic. Which US companies fit this profile. They're interested already in treating name your disease. They're interested already in solving problem that you have discovered your invention could solve, that are interested in wastewater, treatment that, whatever it is that your invention can help do. This is kind of a generic question. You can actually have that as a fill in the blank as part of your prompt process. Develop a series of prompts with blanks and just fill in the blank with the appropriate information for each one. All right, so it's named a bunch of companies here, and I am going to try to get this to go up. All right, there we go. Medtronic. That actually makes sense. If you know this area, you know this makes sense insulate. I've actually never heard of this company, but I would probably want to see a link to make sure that chatgpt did not hallucinate this one. Dexcom Abbott, I've actually heard of Eli Lilly Sanofi, okay, so we have a bunch of really good answers to the question, and this is already helpful. Oh, and I see that someone has a question. Yeah, details on using Reg, okay, you're answering that. Perfect anything else in the chat, yep, you're handling that. Okay, great. So now what is my next question? Well, I want remember, I want to write a LinkedIn profile. That's your original goal. So now I'm going to actually ask it to write a LinkedIn profile, LinkedIn profile, and I'm going to go back now I've pre typed this, but this is the kind of thing that you could do. So it would appeal to such a person at one of these companies, without mentioning either the person of the company by name. If you don't do that, it tends to, if you say, appeal to a person, it tends to want to name a person or name a company. So if you don't want to do that, you got to be specific. Please include how this new invention could help these companies develop new All right, fill in the blank. In this case, it's diabetes therapeutics, but it could be anything. Please focus on collaboration and bringing the new product to market and why this invention is commercially viable. Is commercially viable, except for the diabetes therapeutics. This could be used as a generic prompt. You could then fill in what your goal is of your invention. What problem does it solve? How do you know what problem it solves? Remember, you can always ask chatgpt to solve that problem for you, but in this case, we know it's for diabetes management, because that's what the in invention disclosure, that's what the publication said. So again, a generic prompt with one one small change, new diabetes therapeutics, the rest of it would work for multiple different inventions. All right, let's let it rip and see what chat GPT has to say to us, okay, we got the emojis. It got excited.

Speaker 3  53:33  
So this is pretty good. I personally would find it a bit long. So what I would say, in this case the hashtags. It's gone crazy with the hashtags. All right, your contact, but this is still nice. You include that in. It's got, it's got some emojis. So I could say, Please reform it for bullet points with bold for the introductory words, phrases for each bullet point, bullet point, please shorten the non bullet point. Text pre shorten or reduce. And that's just because I happen to like my messages to be really tight. So we've asked it about formatting. We've asked it about putting it together in a certain way, and I've asked it to shorten or reduce. Now, if you don't like emojis, you can also ask it to get rid of the emojis. I'm kind of an emoji person when it comes to LinkedIn. Those of you who've seen my posts on LinkedIn will know that the emojis tend to show up. This is actually less good. So then what I might have to do is actually ask it to do is actually ask it to blend different things. So I'm not going to do that because Peter actually has some demos that he wants to get to, and we want to get to your questions, but you can see how it made it different each time, and how the personalization works, and how sometimes the personalization can actually be quite generic and can be used over and over again just swapping in. Uh, the therapeutic area, the invention area, the problems that that is solved. All right, I will now stop sharing. Let's see if there's any questions specific for this. Um, to share them on the chat. Okay, so the the prompts, um, I'm actually going to be sharing a link to the chat GPT that I did. Um, so when you share a link with chat GPT, you do a share. And then if someone has a chat G, P, T account, if not, you can make one. You will see everything, all the prompts, exactly how they're put in, what would happen. So yes, you can see that. And you'll also see it from the PowerPoint, which we'll be sharing.

Speaker 2  55:35  
All right, in the interest of time, Divor I think I would just like to open it up to questions on last few minutes. Here, I'm happy to direct people to you. Don't you want to show the rag? Show the rag like, really quick. We can show the rag real quick. Okay, real quick rag. We will do the Yeah, custom GPT demo.

Speaker 3  55:54  
Custom GPT demo. That's, I think that's an easier one to get than mine studio, which is awesome, but complicated, right,

Speaker 2  55:59  
right? So, all right. So here's what a custom GPT looks like here, and it basically, if we go behind the scenes here, you know initially, it looks just like an A typical chat session with chat GPT, except as a title on it, right? And if I simply use the default here button and say, you know, write a speech. There we go. It's already doing it. Here's a draft. It pumps it out. And, you know, it's going to generate something, in this case, that is actually not just using its default training data, right? That is kind of all of these indiscriminate troves of corners of the internet. We probably don't want it to be pulling from and referring to to actually generate this output, right? But this is actually, you know, fairly decent. And the reason why, right, still, like everything, we need human review, but the reason why it actually, you know, generates something fairly usable here is behind the scenes. This is using rag here, and let me edit this. So this is how you, quote, unquote, configure, basically create a custom GPT, and you'll see that you actually, you know, can put on your own name description, but the instructions here are to follow brand guidelines, and these are kind of baked in to the behavior of that individual, you know, custom GPT. But the power here is actually I've uploaded a number of, you know, already publicly available, in this case, previous speeches, right? That are, it's using as reference, right? And this is pretty powerful. And, you know, this is a way to generate far better results. You can make your own custom GPT. I believe you need to have a, you know, paid account to do this, it's about $20 a month, right for GPT chat, GPT plus, definitely worth it here. But you can upload a number of different files, and so this is, you know, probably the quickest and easiest way to to see how rag can actually improve the results when you provide things that you know, again, I wouldn't put in sensitive materials or upload them directly into this, necessarily. But if you have, you know public university like UCLA, public grant brand guidelines, for example, or things of this nature to help you, this is a good way to to play around and use the power of rag, which is just referring specifically to these uploaded PDF documents. And you can upload other file types too. But, you know, there's other products where you can do much more sophisticated, detailed and intentional sort of querying of these and have fine tuning abilities that are far more powerful than this. But yeah, let's we have about a minute left. Love to take a question or so here, I think you're muted. Laura, so rag, yeah, rag stands for retrieval, augmented generation. And what that means is basically it's going to be the results of that conversation and prompt augmenting the outputs through referring to a third party uploaded document. In this case, it was the PDF, right? And so retrieval, augmented generation is like the Yeah, the full term rag is kind of the buzzword, because it's less of a mouthful. Just means you're essentially using the power of, you know, the AI model, plus, you know, running it through as a reference other proprietary data. And that's how you can really just. Generate, you know, save so much time generating much better, more accurate results. I encourage you to experiment with that, you know, but be careful about what you're using right in terms of sharing data directly with your chat. GPT, in this case, I know we're about a time I see Don V L here. I also think you're still muted. Devorah,

Speaker 3  1:00:24  
I'm so sorry the mute button and I was having trouble today, so I put in a quick answer about the creating the marketing summary, the patent, you need to specify what it is that you want. If you just ask for a summary, it's going to sound like it was written by a lawyer, which is quite annoying to say the least. So be specific, The more details you give, the better answer you'll get.

Speaker 1  1:00:50  
All right, on behalf of autumn, I would like to thank Devorah and Peter for their informative presentation today, and thank you again to our sponsor, Marshall Gerstein. A recording of this webinar will be available for viewing in the autumn Learning Center within a week of this event as and is included in your registration as a reminder, please complete the webinar evaluation following this session, it will open when you sign off the session, and will also be sent in a follow up email on tomorrow. Please join us for our remaining may webinars and thank you for being a part of today's presentation. Thank you again to our presenters and have a great afternoon. Everyone.

Unknown Speaker  1:01:30  
Bye, bye. Thanks. Everyone. Thanks. Everyone. You.

Transcribed by https://otter.ai