Welcome to another NachoTuesday. Today, I'm joined by Krish from fireflies.ai, and we will discuss Chat GPT, GPT-3, and AI's role in the future. We'll explore its impact on the market and the workplace. So, without further ado, I'll let Chris introduce his company and provide some background. Hi Andy, thank you for having me. Given the current prominence of AI in the market, this is an exciting topic. AI is permeating every aspect of the tech industry and extending to other industries as well. Let me start with a brief background. I'm Chris Romanini, co-founder and CEO at fireflies.ai. We've developed an AI assistant that attends your meetings, takes notes, provides summaries, and analyzes conversations. Recently, we've been incorporating Chad GPT into our core technology and products. It's fascinating. Could you tell us more about the use cases you have for Chat GPT with the fireflies.ai platform? Absolutely. One of the things we introduced is called Ask Fret. We wanted our AI assistant to be interactive rather than just silently joining meetings. Ask Fret allows users to retrieve knowledge and information quickly after a call. Instead of searching through lengthy transcripts or skimming keywords, users can simply ask Fred questions about the meeting. For example, they can inquire about the number of seats a customer wanted to purchase, objections raised during the call, strengths mentioned by a candidate, weaknesses, or areas of limited experience. Fred pays attention to these meetings, not only transcribing and summarizing but also comprehending the content. Just like you would ask any business admin or teammate, Fred understands your questions and provides answers. It significantly reduces the time it takes to find relevant information from minutes or hours to just a few seconds. That's impressive. How did you come up with the name Fred? It's an interesting choice that stands out. When we looked at the market of AI assistants, we noticed a stereotype that they typically had female names. We wanted to break away from those gendered norms and do something different. We didn't want our assistant to be labeled as a female assistant or secretary, as that perpetuates historical workplace context. We believe women are doing amazing things in the workplace. Coincidentally, most of our investors are female, which is interesting. So, we decided to choose a name starting with "F" that aligns with Fireflies. It was that simple, and Fred stuck with us. That's a fun backstory. Many companies are hesitant about implementing AI strategies. Although everyone talks about AI, few know how to put it into practice effectively. What would you say to those who are skeptical about the concept? Implementing AI does come with several challenges, depending on the focus area and whether it's an enterprise, mid-market, or consumer product. The potential achievements with AI vary significantly. We're still in the early stages, where challenges include context accuracy, data quality, data volume, and maintaining training within specific boundaries. With large language models, there are also copyright concerns. Companies must consider who owns the generated content when using AI to write code or utilizing generative AI. Controlling the output of AI systems, especially in chat GPT and generative AI, is a challenge because of its black box nature. There's a risk of generating misleading or incorrect information. These are aspects that companies need to think about, and we're still in the very early stages of AI development. It's similar to the early days of cloud software or iPhone apps, where gimmicky applications eventually evolved into solid use cases leveraging the technology. Instagram and Snapchat are great examples of successful implementation. The camera really serves as a vantage point, especially when using GPS-based services like Uber and DoorDash. We are still in the early stages of utilizing this technology, and my suggestion is to choose a specific problem to solve. We are far from achieving General AGI (Artificial General Intelligence) technology, so it makes sense for enterprises to focus on fixing specific problems and enhancing their solutions with AI. In my experience, it's important to select a problem you want to solve and recognize that General AGI is still a distant goal. Enterprises should concentrate on addressing specific issues and leveraging AI to improve their solutions. These are the considerations people need to make. Recently, I interviewed Dave, the CEO of Hawk AI. He emphasized that AI tools primarily serve as efficiency tools. While they can't be entirely relied upon, they can assist in handling more complex tasks. The 80/20 rule applies here, where humans need to be involved to ensure quality control. AI can take on some of the heavier work, especially with the emergence of new and capable tools. However, human involvement remains crucial to ensure the content is of high quality and original, rather than copied from elsewhere. There are tools available that can identify whether content has been generated with Chat GPT-3 and detect plagiarism. OpenAI has introduced their own watermark technique and tools to address this issue. Other tools like QuillBot also offer plagiarism detection specifically. This is particularly relevant in educational institutions where students may attempt to generate papers using AI. However, it is still uncertain how search engines like Google will handle generative content. Will they penalize it or treat it similarly to normal content? The middle ground might involve utilizing AI to aid in brainstorming and idea generation, followed by human curation and editing. Therefore, viewing AI as a productivity tool is a more appropriate perspective. AI tools, including chat GPT-3, are predominantly used as efficiency tools rather than standalone solutions. They can assist with tasks but should not be solely relied upon. This applies to writing content as well. While there have been notable advancements in content generation, particularly by companies like Jasper, there are still gray areas and concerns related to plagiarism and originality. At Fireflies, we have chosen to focus on solving two significant problems: knowledge retrieval and meeting efficiency. These are areas where AI can greatly enhance productivity, even before the emergence of regenerative AI and similar technologies. In the context of academic settings, code checks, and other applications, the issue of plagiarism and content generation is significant. Students in the past may have used tools like TI-83 calculators or hidden notes to cheat. Similarly, using tools like Turnitin.com for citation and source-checking has been a common practice. The rise of AI tools brings new challenges in detecting plagiarism, especially when it comes to code checks and computer science classes. False positives and discussions surrounding this topic need to be addressed separately. Moving forward, there are several AI tools that offer reliable uses. One notable application is the ability to complete actions, extending beyond generating text output. These actions can include performing Robotic Process Automation (RPA) tasks or even making purchases on platforms like Amazon through voice commands translated into code. AI can generate not only words but also tables, code snippets, and even bridge the gap between text and images through multimodal capabilities. Customization is also an exciting prospect, such as allowing users to personalize their note-taking style using generative AI. The current state of technology allows for various reliable uses of AI. However, it's important not to get too far ahead of ourselves and remain mindful of the technology's limitations. AI has the potential to automate You will get auto suggestions, right? That's a great example of generative AI, and that's been happening for a few years now. So, yeah, it'll be hard to see, but we'll see where it goes. Well, the use of Grammarly, of course, it's always good for double-checking your work. Future of GPT4: Um, but yeah, sometimes it misses the context, so you can't take it 100% for everything it recommends. So, what's next for Chad, for say GPT4? Um, what do you think is going to be in that and what's beyond that? So, I have been looking into it, and the rumor mill is endless, so I don't want to add to the rumor mill. One thing that I would love to know and see is how much new data is going to be utilized for GPT4 and what are the things that GPT4 can do in terms of computations, in terms of intelligence. One of my biggest beliefs, and this might be counter-intuitive to what a lot of people talk about, but this phrase "prompt engineering" that has come up and how people have said it's going to be its own field, and we're going to have these prompt Whisperers who know how to say the right things to get the right output out of AI. Um, to be honest, I think that we may not even need prompt engineering in a few years, and I think that is just right now a void to fill while the systems and technology get better. And eventually, the system will understand right that it can't generate a great output, and it's not just you phrasing a word in a certain way or adding a certain word and taking out another word. I don't really buy into the whole prompt engineering narrative as a career field, at least maybe I could be wrong, but that's something I don't quite buy into purely on the basis that the technology is going to get incredibly better. We'll have to see what's gonna happen, right? This stuff, which GPT3 was just in a one-year, two-year timeframe, now it's hard to predict what's going to happen in five years. But if some technology like GPT4 is trained on way more parameters and it can do more things on top of that, and the responses get better and it's able to also fact-check itself better, provide citations, from a user experience point of view, all of those things will make it less likely for someone to have to worry about the prompt engineering aspect. There's also this uncertainty part, like there's a temperature feel that you can change to see how much randomness do you want in the output, how much entropy, right? And even there, I think with the way the systems are being built and how we're getting to the outputs that we want, it'll be very customizable. And I don't think that we're very far away from something that just gets progressively better. Now, is GPT4 going to be like 10x better than GPT3? Probably not. So, we should taper our expectations. But if we see an incremental shift in the value of the outputs, that's still a good starting point. That's a cool point. Yeah, I think AI, GPT3, and AI, was kind of a buzzword up until Chat GPT3, and I think that's why it's so profound. Because there's a lot of companies bullshitting themselves that they're doing AI. But until the last year, you know, companies like yours and others that really dialed in the technology, and I think have really improved upon it, and ChatGPT3, I think it Um, cumbersome and this could save time. Yeah, I mean communication is the core for business, right? And it's a lot of communication. People aren't very good at it, and especially between meetings, emails, a lot gets lost between the cracks. You know, I think that's what you guys are working on there, a perfect solution that could help businesses become a lot more efficient. Ultimately, the goal at the end of the day is efficiency, which is also directly correlated to breaking down silos. A lot of times, miscommunication and lack of communication are the biggest barriers in a large organization. And we're still a small startup, about 100 people, so it's not massive like a company that has thousands or tens of thousands of people. But even here, it's the communication gaps that happen, why are things that should be happening in a matter of minutes and hours taking weeks? Why are people scheduling follow-up calls for certain things that don't need to be follow-ups, that could have been an email, a message, or just a quick look at Fireflies' recap to confirm a decision? So the amount of time people waste in meetings and this "let's get everyone on the same page" part, we've automated that. I can have a meeting and just share the recap with people. Oh, you want to get on the same page? Go use Fireflies. We don't need to have another meeting, bring in people, and repeat the same stuff over again. So we have a very similar culture in terms of how we operate, and we're fully remote. So context switching and silos are even harder to address in these circumstances because everything is remote. You're not seeing people in person, and there's not a lot of tribal knowledge that's passed around in person. So we have to document everything, we have to use Fireflies to store everything. When a new person comes in, let's say they're in sales, for example, we can say, 'Hey, here are the last 10 best conversations we've had with customers. Sales calls, go ahead and review them. I've also left comments in there and marked important parts of the calls. There are tips on what you can do.' Now I have automated training for all future employees. If we have a new engineer coming in and they want to know about our infrastructure, our stack, and why we made certain technology decisions over others, well, they can look into some of our planning meetings, design things, or engineering discussions and leverage that information. I don't want to have to repeat myself a hundred times. Does it matter where I'm at? How can I just say it once and then have it propagated and shared to a thousand different places and get everyone on the same page? So whether we've solved everything we needed to or not, I think one mission that we've set is, it's a mission you can build a lifetime of businesses around, and that is how can you improve the trade-off between communication and actually doing the work. This is one of the reasons where I look at organizations that bring in consultants and people to strategize and talk. Really, action is based on context, and context is king. But you don't need to just have one person at the top or a manager have all the context. You can give every person in the organization the context, and for us, that's the way to get everyone on the same page. If you want to get on the same page, just use Fireflies. Look at the notes, review everything, and you will be 10x smarter after going through that, then having to spend six weeks scheduling follow-up meetings with each and every person. Thanks for having us on today, Andy, it was my pleasure.