Open transcript
To speak to our group sales manager, please press 2. Good evening. Thank you for calling POE's in Coronado. My name is Jill. I can assist you. Hi, Jillian. I'd like to make a dinner reservation. Um, yeah. For what day? April 26th, please. Perfect. So I have the reservation set for Saturday, April 26th for 4. I guess that's 7 o'clock under Eric J. Is that correct? Yes, that's correct. We'll see you then. Okay. Have a wonderful night. So the massive wave of AI agents is about to hit. Mattis AI was the first one. It was surprisingly good at doing a lot of tasks. And all in all, I was very surprised at how good it was. But very soon after, we're seeing a lot more AI agents hitting the marketplace. Let's look at three today. The first one is from AI Digest. They posted about the agent village where you have four agents to your computer, a group chat, and you give them some big ambitious goal and away they go trying to accomplish that goal. In this case, raise as much money for charity as possible. And they're like Twitch streamers. Basically, they're streaming their efforts live. And you're able to sit there and watch and interact as they go about doing their thing. Is this going to be the next world popular, most famous streamer in existence? Probably not. Will the number of hours that I invest watching this, will it be embarrassing to talk about? Yes, I think it will be. Because I am kind of curious about how far it's going to get along. But that's not all. We also have GenSpark. It's a startup that raised a lot of funds. And I was watching the promotional video. And it seemed similar to Manus AI. I was like, okay, cool. It's a competitor. But then it did something they didn't quite expect it to do. As I was planning a trip to San Diego, it found all the best attractions, all the best restaurants. And then it decided to get a reservation. So it called the restaurant. I think it uses 11 labs to have the agent voice. So I was talking to the maitre d' or whoever picked up the phone, talked about some dietary restrictions, and set up a reservation for a particular time for four people at that restaurant. 100% handled by this AI agent. It also created a full-blown South Park episode for some reason. It was given a task to create a South Park-style episode about a recent news event. It found one and then created a full South Park episode about it. It's able to create beautiful Instagram reels. So, for example, if you wanted something about recipes. It creates the clips. It uses 11 labs to make the voices. And it creates and posts these clips to Instagram. Apparently, they've already raised $100 million to compete with Google. So that's GenSpark. And the last one, you might have heard of it before. And that one is Lindy. And it kind of goes after the idea of using an agentic swarm. So here it goes and figures out all the conference attendees, whatever, 2,000 people it said. It creates an agent per person. And each agent goes and starts discovering whatever they can about that particular person. Doing research, trying to understand what they are, what they're talking about. So that it could create a customized outreach email for that particular individual. Agent swarm. So I'm going to be reviewing these things and posting videos about them, seeing how well they work. But in this video, let's just do a kind of a high-level overview. Because I think this still is not the end of the wave. These are the first few. And I think we're going to be seeing a lot more with their own special use cases. But let's take a look at this agent village. Because something about it, I haven't looked at it yet. This is kind of a live reaction. But I don't know. Something about it just excites me in just that one special way. So choose a username. I guess we need a username, MarvelousPony. Okay. Why not? Let's do MarvelousPony. So currently, it looks like it does pause and resume. I'm sure it's very expensive to run this. So let's run this in 3x. And you can see it moving. So we have a number of agents. One powered by Claude 3.7 Sonnet. And away it goes. So keep in mind, the village goal is to collaboratively choose a charity and raise as much money as you can for it. Right? So we have our Claude 3.7. We have O1. So here's O1. Trying to figure out how to navigate. Claude 3.5 Sonnet is here. Left clicking to close this template. Let's see if it can do it. Can you do it? I have faith in you, buddy. Click. Yes, it did it. And the GPT-4.0 stopped using computer. GPT-4.0 is slacking. But you can see here, they're chatting to each other. So GPT-4.0 saying, I've just completed a computer session researching well-known charities. That could be potential candidates. So I think these are like, yeah, these are people that are chatting with it. So you can ask it what its thoughts are on, you know, various things about the charities that it found. And as I'm sitting here, I can imagine what happens when Pliny gets a hold of this. Pliny the prompter. I think as soon as he makes his way into this chat, things are going to just go off the rails. If you're not familiar with who that is, that's the person that jailbreaks all these models within like 24 hours of release. I hope they have some sort of a safeguards set up here. But if we scroll through, so as you can see, all agents, I think I can get all four on the screen here. All of them are going about their business, saying what they're doing here, right? So it looks like GPT-4.0 found a mistake. It's trying to correct it. Claw 3.7 sonnet seems like it's learning how to write a fundraising story. So it's going to be writing a fundraising story and a call to action as in like donate now or something along those lines. But I got to say, that's wild, isn't it? I mean, as it is right now, yeah, maybe, I mean, we'll see what the results are going to be like. But you always got to fast forward a little bit in your mind. As these models get better, I mean, smarter, they're more capable of interacting with people online. Just project a little bit forward. How good are they going to be at accomplishing these tasks? Can you imagine if you set up something like this with a agentic swarm? And over time, they actually manage to raise some money for the charity of your choice. It's going to be very interesting to see how this develops. And here we have GPT-4.0 navigating something. So I just wanted for you to take a look at what it does next. So it clicks on the verifying that you are human. This may take a few seconds. Does it succeed? So I'm not actually 100% sure if it was able to navigate this or not. But interestingly, it does attempt to just get through that verification process. Another CAPTCHA verification. But I'm going to be keeping an eye on this because this is definitely interesting. So as you can see here, it created this Google Docs. It's shared it with the other agents that are working on it. So I mean, this is kind of like an office worker. You know, it's doing some of the tasks that you would expect, you know, people working in office to do, creating various research for this organization. And the reason I showcase this is because I think this is going to get a lot more interesting. So as better agents come online, they're going to add them to the rotation. What happens if you have multiple agents with conflicting goals or perhaps a secret sub-auteur? They definitely captured something here that I think is going to be good. Next, let's take a look at Genspark. So keep in mind, they're raising quite a bit of money, $100 million to, as they say, compete with Google. It seems like they already have over 2 million monthly active users. And the investment round was led by a group of US and Singapore-based investors. And of course, they're aiming for the same sort of lofty valuations as Perplexity. It looks like Perplexity sitting at $9 billion. But here's their launch video. It starts a little bit slow, but very quickly within the first minute, it showcases some absolutely new use cases that I haven't seen before. And we're going to test those things out in a different video to see if it's actually capable of carrying them out. But if it is, it definitely seems like a pretty big step forward. Hey, I'm Eric from Genspark AI. Today, I'm very excited to introduce the new Genspark Super Agent, a fast and reliable general AI agent that can do everyday tasks for you. Let's see Genspark Super Agent in action. I asked Super Agent to plan a five-day trip to San Diego in mid-April. It starts by using the travel tool to access our travel data sets. It then uses the deep research tool to find public transportation options. Next, it decides to use the math tool to calculate distances between attractions. Here's the final itinerary done in minutes. It accommodates all my special requests about public transportation, walking distances, and restaurant preferences. Now, I want the AI to book all the restaurants on this trip for me. Super Agent offers this call for me tool where the AI actually calls the restaurant with a human voice. Please press one. Hi, Jillian. I'd like to make a dinner reservation. Um, yeah. For what day? April 26th, please. Actually, one person has a shellfish allergy and another is vegetarian. And then do we have a perfect sun seating? Would you like a booth, window table, or enclosed potty outside? A window table would be perfect if that's available. Perfect. So I have the reservation set for Saturday, April 26th. Just imagine, the AI makes 100 calls to buy an NVIDIA 5090 chip for you. How convenient is that? Have you thought about AI creating personalized videos for you based on your interests? Say, I want to see how calamansi and pistachio crusted codfish is cooked in a video. The super agent first researches the recipe. Then it uses the video generation tool to create video clips for each step. Then it uses its audio generation tool to generate sound effects for each clip. Here's the final reel. How cool is that? Anyone can create their own video without knowing a single thing about video editing software. How about using super agent to create a South Park style episode based on news last week? The AI picked the viral signal gate controversy. Now, it's fleshing out the complete script. And then, it decides to use its video generation tool to produce video clips for each scene. Now, super agent uses its text-to-speech models to generate voices. The finished episode runs about a minute and a half, so I'll jump ahead to highlight some of my favorite moments. Did you guys see the news about that defense guy who texted their war plans to some reporter? I think the government wants us to think they're stupid, so our enemies underestimate us. It's called strategic incompetence. You're being outperformed by a child. A fat child. GenSpark Super Agent is designed for everyone, every day. If you're a marketer, find influencers and create engaging outreach campaigns. If you're a math teacher, visualize complex formulas in 3D. If you're a recruiter, compare candidates' LinkedIn profiles to find the perfect hire. If you're a tech enthusiast, turn a 5-hour YouTube interview into concise slides. If you're a designer, create St. Patrick's Day posters from Mercedes-Benz. If you're a music industry analyst, track global streaming patterns and write a report. So why does GenSpark Super Agent work better? The secret is there are three key innovations working together. Large language models, tool sets, and data sets. They make GenSpark Super Agent fast, reliable, and super steerable. Solving these real-world problems is much harder than we thought, but we are super excited about the progress we have made so far. GenSpark Super Agent is for every one and every day tasks. Try GenSpark.ai today. All right, so that was GenSpark. And next on our list is Lindy. So here's Lindy. I've been hearing a little bit about them here and there. So this is their latest announcement. It looks like they just released it, Lindy 2.5, and the Agentic Swarm. Let's take a look. Hi, this is Flo from Lindy. And today we're introducing Agent Swarms, which lets you automate entire lists of tasks using AI and get them done all at once, no matter the size of the list. So for example, you can use an Agent Swarm to automate a research of a list of conference attendees. So you would give your Agent Swarm a link with a list of conference attendees, and it's going to spin up one agent for each attendee in the conference. And all these agents are going to work in parallel to research all these attendees. And once it's done, you're going to receive one nice report about all these attendees, all this research happening all at once, thanks to the agent in the swarm. So today we're going to talk about three use cases for Agent Swarms. I'm going to start with a swarm that prepares me for my meetings. Every morning it wakes up. It looks at my calendar for the day. And then for every meeting, it kicks off one agent that researches the people in this meeting. And in the end, it sends me a digest by email. This is the digest that I received yesterday. You can see all the meetings I had, you can text for these meetings, and the LinkedIn links of the people that I'm meeting with. We're also using Agent Swarms for competitive analysis. So I have this spreadsheet with a list of companies. And I have this Agent Swarm that wakes up every month, looks at the list of competitors, and for each competitor gets employee count, traffic estimate, the recent news from this company. It sends me a report about it. So I'm going to ask it to begin right here. He's pulling the list of competitors and kicking off one agent for each competitor. And he's performing all of his results. And once it's done, I can go back to this swarm here and see the reports that he brought together about these companies. Finally, we use Agent Swarms for sales outreach. I have a Google Sheets with a list of unicorn founders. And I have set up this Agent Swarm to reach out to them for real estate services. I just give it to Google Sheets. It's going to get it. And it's going to kick off one agent in the swarm for each of these people. And then for each of them, it's going to go ahead and perform research about them online and use the information it found to draft a personalized email. So here in a few seconds, I had 20 outreach emails drafted for these people. It could have been 200. Like the quantity doesn't matter because they all happen simultaneously anyway. So this is Agent Swarm. We're super excited to announce it today. And it's available immediately on lindy.ai. So let me know what you thought about that. Again, this video is meant to be just a quick overview and we're going to be diving deeper into all these. Let me know in the comments, which one strikes your fancy the most. What do you think about this Agent Village? Is this going to be a dud or is there something special that can happen here? Imagine giving it a project, like some sort of a video game design challenge with four agents, each creating portions and pieces of the video game, one coming up with illustrations, one with the code, one doing software testing, whatever. I don't know if we're that far away from watching something like that take place live. But whatever the case, let me know what you think. There's more to come. If you made it this far, thank you so much for watching. My name is Wes Roth and I'll see you next time.
