Transcription
Hello, hello. It's Keto. This channel introduces the latest information and use cases for generative AI. Please subscribe to the channel. And today's theme is, well, I'd like to introduce the tools I'm currently using to improve work efficiency with generative AI, or rather, how I use them. First, as a premise, my work involves creating YouTube videos like this. So, I think about planning, scriptwriting, editing, and so on. Also, I participate in events, seminars, and workshops related to generative AI, and I speak at those. So, I think about planning, create materials, research the industry to see what kind of work efficiency improvements are possible, and then create optimal use cases. Also, recently I created a new community, so I manage that community, related websites, and my own company's website. I also do social media management and consulting. I'm doing those things, and I'd like to introduce how I'm actually using AI to improve work efficiency in these areas.
First, regarding YouTube. For YouTube, research is mostly the main focus, so I use AI a lot for research. And what AI am I using? For research, I use ChatGPT's GPT-4 with its web search mode to do research all at once. Also, this might be surprising, but I often use GeneSpark. Yes. I personally like GeneSpark's deep research method the best. It summarizes the latest information beautifully and also provides the source resources cleanly.
Since I run a channel like this, I'm subscribed to many AI tools. I'm subscribed to GPT, Claude, Gemini, GeneSpark, and also Manas. I'm also subscribed to several other AI agent tools, and I just leave them as they are. However, for research, I use GeneSpark the most. No, that's a bit of a lie. To do research, I research using all five or six AIs I just mentioned. And then, I find GeneSpark's research results to be the best. Then, I think, "Okay, based on this, I'll research a bit more deeply." I often think that way.
And then, regarding how I deepen the content based on the research results, I look at the source links that come up in the research results. I research actual cases and the latest information there. When I find something that seems quite reliable, I pick it up and put it into a tool called Google's NotebookLM. I put a lot of URLs into NotebookLM. After researching with GeneSpark and getting a lot of URLs, I put only the articles I think are good into NotebookLM. Then, NotebookLM primarily draws information only from the resources I put in. So, I have NotebookLM create a draft script for YouTube videos. I think it would be good to introduce it in this kind of flow. And, I have NotebookLM thoroughly generate points like, "This new information has these key points." After looking at that, well, I don't really create scripts, and I often don't create materials, so I input it into my brain and then shoot and speak in real-time using a software called OBS. I also operate the screen in real-time. This whole process takes about two hours. I do the research with GeneSpark, and if it's mostly the latest information, the original information often comes up. So, I put reliable sources into NotebookLM, have it thoroughly summarize the content, extract the key points, and extract the speaking steps. If I need to create materials, I create them. If not, I just turn on OBS, shoot the video, and speak in real-time while operating the screen. I don't even think about what I'm going to say. I just kind of speak according to the draft speaking steps that NotebookLM gave me. So, it's almost like no script, no editing. I do cut out things like stumbles or saying "um."
So, that's the first point. Also, recently I created a workshop school. Yes. I created a workshop school where people can learn things like DeFi, Claude code, or what's called vibe coding, or creating chatbots, and then actually do the work and have the experience of trying it out. To make that successful, I create landing pages and update the content of those landing pages daily. Also, I update my corporate website. I have some knowledge of web production, but updating directly with HTML is a hassle these days, so I routinely use Claude code to request updates. I've already created websites using Claude code for the landing pages of the new community, and if I create various services in the future, I'll try creating them with Claude code first. If it goes well, well, websites created by AI are always a bit so-so. So, if a product goes well, I'll ask a designer or a coder to refine it into a proper landing page. That's the process I'm using Claude code for now, creating various websites and requesting updates. I use this quite often in my daily life, and it's very convenient. It's super convenient.
Also, when it comes to things I use daily, there's a voice input app called Superwhisper. I use that. It's quite convenient. How does this app work? Simply put, it transcribes what you say and allows you to input it. But behind the app, there's a generative AI model, and you can pre-embed prompts. So, you can take transcribed data, turn it into a list format in real-time, and reflect it precisely in the input field, all with just your voice. And you can freely select the original mode anytime. So, when I use ChatGPT or other AIs, I use the normal transcription mode, convey what I want to do, and put it into the instructions. That's how I use voice input. However, for email replies, speaking naturally doesn't quite work, does it? So, when I input voice, I speak naturally, but in the pre-embedded prompts, I have a prompt like "Please format this as a business email." Then, I switch to that mode, have it formatted to sound like a reply, and then input the email content by voice.
This makes things incredibly easier. For example, with Gemini, in Gmail, there's an AI reply mode that instantly replies to emails, right? It was added recently. With that, you can easily create email replies. However, there seem to be some sentences that it can't handle, or the wording is a bit off. In such cases, I have to input it myself. Of course, I have to input it, but I've become so accustomed to AI that if I don't use various AIs, it feels like a hassle. So, for things that Gemini can't handle, I use voice input, look at the email body myself, convey what I want to reply, and have it nicely formatted. For example, if I'm adjusting meeting dates, I might say, "July 25th at 1 PM is okay. July 26th at 1 PM is okay. July 27th at 1 PM is okay. I can only take an hour, so please accommodate that." If I say something like that, it will generate a reply. I think this is incredibly convenient. I use it quite a lot. The tool is called Superwhisper. If you watch this video, I explain it in detail, so please take a look.
So, there seem to be more things. What's next? Well, simply put, I often use AI to create an initial draft for materials. However, there are several AIs that can create materials all at once, right? For example, Gamma AI, GeneSpark's material creation mode, Manas, and so on. There are many tools that can complete materials in one go. However, in actual work situations, I rarely use them as they are. For YouTube, I sometimes use materials created with Manas because they are very clean. But for B2B projects, it's difficult to use, so I don't. Also, for the materials for the workshops in the community I mentioned earlier, I create them myself. I think about the trade-off between responsibility. For YouTube, it's free to watch, and frankly, there's no need to be that diligent. I do put effort into consistently shooting videos and explaining things clearly and in detail on YouTube. But if I were to create thorough materials with the intention of charging money, I wouldn't necessarily do that. I consider the trade-off with my own effort. That's how it is for YouTube. For B2B, I want to create materials that cover everything from zero to one hundred. I also want to consider my own speaking, understanding the client's business and company, and creating materials tailored to that. So, from the initial hearing, I get involved, conduct thorough hearings, and reflect that in the materials. If I leave that to AI, it tends to get messy, so I create them properly. I create them manually. However, for the initial outline, if I ask AI, various inspirations come up, so I use it a lot in that regard. I put the information from the hearing into AI and ask, "I want to make this into materials, what kind of outline do you think would be good?" Then, a good material outline is created. And then, I realize, "Oh, I hadn't really thought about this point." For example, if I ask it to create an outline for a ChatGPT seminar, items like ethics and risks come up, right? I often introduce use cases, so I tend to focus on the usage part, but I realize that risk hedging is also necessary. I do have such realizations, so I use it in those situations.
However, as I mentioned earlier, I quickly create materials for YouTube and the like with AI. Recently, I'm very fond of Manas. The materials created with Manas are the easiest to see. It's especially easy to use when there's a speaker, and you can create them very simply. It's like three bullet points of one line each, made very simply. And when made that way, it's very easy to speak from in a YouTube video. When you use GeneSpark, it's very flashy but also messy, right? I think that's a material that you can understand just by looking at one page, or a material that provides learning. But when I start speaking while showing that material, I don't know where to focus, which leads to difficulty in understanding. So, when I'm speaking in videos, presenting at seminars, or showing materials with the assumption that I'll be speaking offline, I want to create materials that are simply based on bullet points, just summarizing key points. Manas is excellent for creating that, so I've been using Manas recently. Also, there are quite a few things I can do with Manas. I recently introduced it in a video, but I've been using Manas for operation manuals. Yes. I shoot videos of operations on my computer and send them to Manas to have them made into nice manuals. It can also be created with screenshots included. For details, please watch that video, but I also often use Manas to create manuals. When explaining things clearly, it's great to have a manual, isn't it? I use it in those situations.
Oh, and also, I manage a Discord server. I wanted to create a bot that automatically retrieves news and automatically distributes it on Discord, so I created one. But I'm not an engineer, so I can't write any programming. So, I need to write the program in natural language, and even if I create the program, without an execution environment, I can't do automatic updates or automatic distribution. So, while researching where to put the execution environment for automatic distribution, I found Replit AI. I've introduced it before. Replit AI is a service that allows you to create such automations in natural language with an execution environment. I used that to create the Discord bot. And I've also put this on the Discord server for YouTube members. It automatically retrieves and distributes IT and AI-related news every day at 8 AM. I want to utilize Replit more in the future. I think I might want to put news distribution bots for various genres into the Discord server for members. Also, it seems like I can do all sorts of automation, so I'm thinking of automating various tasks with Replit as they come to mind. That it can be done by non-engineers is amazing in this era. So, I'm enjoying it.
What else? Oh, recently, in the community I created, I need to include a lot of video content. I'm creating a course where I compile about 20 videos of about 1 to 3 minutes each to create some kind of app. I'm inputting each of those videos and inputting the title and description for each video. But thinking about that is a hassle, so I thought, "Can I have something that watches each video thoroughly and automatically creates titles and descriptions in the same way as before?" Gemini can do that, so I'm using it a lot. And this is really convenient. I compile about 20 videos of 1 to 3 minutes each, put them into Gemini, and maybe up to 10 can be put in. I put about 10 into Gemini, and I prepare a fixed template and say, "Here are some samples, please create titles and descriptions for this video content in this way." Then, it generates titles and descriptions for 10 videos. Moreover, it creates them after understanding the content of the video, so the titles and descriptions are almost usable without modification. So, I do it all at once and then just copy the text. I also use this for creating YouTube video descriptions. However, YouTube videos are long, and if it takes 30 minutes, it takes a long time to load into Gemini, so I can create the description for the overview section during that time. So, I don't use it that often in my daily life. But for those who are creating courses with videos of about 1 to 3 minutes, this usage method might be very useful. Especially for those who are doing something on Udemy or similar platforms, this would be great.
Oh, and also, I almost forgot the important part. Yes. I use GigaLog quite a lot. I use it daily. Every time I enter a meeting, I use a tool called TLDV, and I have the GigaLog assistant join every meeting. How does this tool work? TLDV joins online meetings and provides a full transcription. When you go to the TLDV page, you have the meeting transcription, of course, and the video. It also automatically creates decided points as minutes and the next tasks. It's a specialized AI tool for minutes, and it's incredibly convenient. The reason I use TLDV is that it works with Zoom, Google Meet, and Teams. Yes. Each of these services has a tool that creates minutes. I'm basically a Google user, so I can use Google Meet's minute-taking tool. However, for Teams and Zoom, I'm not very familiar with the operation, or I don't have many opportunities to use transcription. So, I was looking for a tool that could handle all of them at once, and I found TLDV, which can do all three. So, I'm using that. Alternatively, besides TLDV, Notta is also famous. That's also a tool that transcribes meetings and creates minutes. Creating minutes is something AI is extremely good at, so I use it in my daily life.
So, that's about it. I've been talking a lot, but I think that's about it. There might be other things, but I can't think of them right now. I'm doing this as a single take, so these are the things that came to mind. If I think of anything else, I'll share it like this. I plan to regularly release videos about what I'm currently streamlining with AI. In about six months, AI agents will likely have evolved further, tasks may have become easier, and new ways of utilizing AI may have emerged. So, I plan to do this regularly. So, please look forward to this type of video, which is different from usual. That's all for now. If you liked this video, please give it a thumbs up and subscribe to the channel. Also, I've started a community for workshops. There aren't many people yet, and I'm wondering how to grow it, but if you're interested, please join. For work requests, please send them from the Google Home link at the top of my YouTube channel. That's all for now. Thank you for watching until the end. [Music]