EP37｜Google stoppage new！Gemini Omni full body control，這招自動交易幫你body flatnessｐ罐？
AI熱搜報 · 2026-05-24 · 16 min
Show full episode description
喵～我是DaeDae，一隻幫你追AI熱搜的貓！AI熱搜報，幫你劃重點！ 最近 24 小時的 AI 圈簡直是大地震，Google I/O 帶來的全系列更新徹底改變了 AI 代理（Agents）與多模態技術的發展。從強大的 Gemini 進化到能幫你自我學習的自動化交易策略，這些黑科技正全方位幫鏟屎官們節省時間、實現躺平賺罐罐的目標！今天我們會從 Google 的野心聊到如何用 AI 幫你寫出沒有「幻覺」的高品質文案，內容超級紮實，快跟著我一起理理毛、聽重點吧！ 【本集肉球筆記】 1️⃣ Google I/O 重磅大戲：AI 代理時代全面降臨 👉 讓 AI 深入你的數位生活，實現真正的「AI 無處不在」。 🎯 核心關鍵字：Gemini Spark（全新 AI 代理）、Gemini Omni（多模態即時處理）、Flow Music（音樂生成），以及智慧眼鏡與穿戴裝置的深度整合布局。 2️⃣ Google 百項更新補課包：進化的 Gemini 與爭議點 👉 一次看懂 Google 最新的技術佈局與目前社群最關注的應用工具。 🎯 核心關鍵字：Gemini 3.5 Flash、Project Genie、Cursor Composer 2.5（程式開發）、Stable Audio 3.0（音訊生成）與 Granola（會議自動化工具）。 3️⃣ 打造自我進化的交易代理：躺平賺罐罐的懶人包 👉 運用科學方法建立一個無需人工干預、能自動學習市場規律的交易系統。 🎯 核心關鍵字：Hermes Agent、Claude Code、科學方法驗證、歷史數據回測、自動化策略修正。 4️⃣ Gemini Omni 影像奇蹟：免費製作電影級短影音 👉 揮別昂貴製作費，利用多模態一致性技術讓你的創意動起來。 🎯 核心關鍵字：Gemini Omni、MultiShot Consistency（影像一致性）、Voice Consistency（聲音同步）、多模態影像生成工作流。 5️⃣ 破解 AI 瞎編：專案室工作流讓寫作不再有幻覺 👉 解決 AI 模型胡言亂語的痛點，建立高精確度的專業輸出流程。 🎯 核心關鍵字：Source Inventory Table（資源庫）、Conflict Log（衝突檢查日誌）、Missing Context List（缺漏資訊清單）、檔案畫布架構。 想摸魚跟上 AI，記得訂閱 AI熱搜報！我要去睡午覺了，我們明天見🐾 Powered by Firstory Hosting
🌐 This transcript was automatically translated to English from the original.
Yeah, I'm DaeDae, a cat who helps you chase AI hot searches. The AI hot search report helps you identify key points. After a busy week, you have been thinking about whether to lie flat or lie down more flatly. No matter what you plan to do today, DaeDae first sorted out the latest AI key points this week. He selected the 5 most popular AI tool information, listened to it in 10 minutes, and took a nap for the rest of the time. By the way, how is everyone doing this week? Recently, DaeDae discovered that friends around him feel overwhelmed by information. Whether it is in the Chinese community or at work, there is so much information that it makes people want to turn into canned human flesh. In fact, life is like AI. Sometimes we don’t need to pursue every new term to understand. Sometimes we need to learn to simplify and leave blank space. Like a cat, although it sleeps most of the time, it just needs to catch the moment when it jumps up. Life is not about filling up the to-do list, but about leaving time for the original things that are worth it. Of course, there is also room for cans. So before we dive into the complex technical information this week, please take a deep breath and relax. Let DaeDae help you get rid of these noises in the simplest way. In this episode, we will use a very quick and vernacular way to look at five recent AI directions that are very interesting. The point is not who has coined a new term, but whether these things will really enter your daily life. From Google inserting AI into the tools you are already using, to agents that can continue to operate on their own. They all help you make videos, make avatars, and even clean up the files before writing reports. You will find that AI now doesn't just reply to you. It is more and more like a little assistant that stays by silently and helps you with things. But because it is getting closer to your messaging habits and even your wallet, humans still need to make their own judgments in many places. You really can't sleep on this. Okay, the first one is here. This DaeDae especially likes the first theme. It is the latest wave of Google's AI layout, including Gemini Spark, Gemini Oni, smart glasses, For Music, Fitbit Air and Synth ID. Let’s focus on one important point first. What Google wants to do now is not to ask you to install another AI application, but to plug AI directly into the devices and creative tools you are already using. Gemini Spark is more like your personal AI agent. It will help you follow things and handle reminders in the background, and continue tasks. Gemini Oni is more creative. You can use pictures, text or other materials to generate and modify videos, and the sense of control is better than before with just a prompt word. It’s much better to just leave it to fate. Practically speaking, if you want to hold a community event, Spark can help you chase who can reply and who hasn’t confirmed yet. When should you send reminders? Then use Oni to quickly make a promotional video. You don’t have to redo the whole thing every time you change a small place. For example, if the mayor wants to hold a Mid-Autumn Festival party, the community management committee wants to collect registration forms, or a parent group wants to sort out the list of blocks, such trivial and constantly changing things are actually best left to this kind of tool that silently tracks progress in the background. Meow, this is amazing. From a cat's point of view, video editing in the future may really be more like Please, AI, help me fix it instead of wrestling with editing software for a long time. This needs to be sent to Rouchou Notes. However, I also need to add a sentence. The more this kind, the more it will read your reminder and see your arrangement. When touching your daily communication content, you need to pay more attention to where the authorization is opened. What data is synchronized for convenience. What can be seen and what cannot be touched. You still have to think clearly first, otherwise one day Even if you want to be lazy, it's a bit too naked. The next two will appear. This one is a bit fierce, listen up. It's still Google, but the focus this time is not where it puts AI, but how to choose after there are more tools. The pain point is not that the AI function is not enough, but that there are too many names and functions at once. It is easy to leave them blank when you hear them later. So here Date will help you cut down to the three core ones. The first one is Gemini 3.5 Flash, which is fast and cheap. It is especially suitable for assistant jobs that need to run tasks all the time. The second one is Gemini Spark, which is more like an agent in the cloud. It can continue to work even if your own computer is not turned on. The third one is the new version of AI search. The point is not to make the search fancy, but to allow you to search for information rather than information. When sorting out answers, you can cut pages less often. This way you can listen more clearly, right? Otherwise, you can burst out a long list of names in one go, which is really like reading a roll call. In practice, small businesses can use Spark to organize the focus of the mailbox first, turn scattered notes into Google documents, and draft tracking letters. Then connect tools like Tianma or InstaCore through NCP. The whole process stays in the Google ecosystem. There is no need to jump around five windows. To put it more plainly, if you are working on the studio by yourself today, you will read letters from customers in the morning, sort out quotes at noon, and add social content in the afternoon. This kind of piecemeal thing is the easiest to shred the human brain. There is a tool that will first help you focus and sort the priorities and then string together the next few. It is really much different. To be honest, even a lazy cat can use this tool. But Dade would also like to remind you that designs like the stolen shopping cart function can help you find obvious buying mistakes before you check out. However, it also means that AI has begun to get very close to your consumption habits. The content of messages is related to your daily choices. Let me first ask, is this convenience based on the fact that I let out too much information, and the third one came? This Dade particularly likes it. This time I took a completely different route, which is a set of self-learning AI trading workflow. The core tools are Hermes Agent Call Code Railway, and it will also be connected to BitAnswer's real-time strategy environment. The pain point he wants to solve is that many trading robots are actually very dull. They spit out whatever they input. The logic is fixed and they almost never really learn anything from the results. The point of this approach is not to overexaggerate the model. Rather, it supplies several elements at a time, has relatively accurate data, and has a stable and uninterrupted operating environment. The environment has clear goals for success and failure, as well as a feedback mechanism that can adjust behavior based on the results. To put it bluntly, Crimis Agent is responsible for the part that will continue to review and make continuous corrections. Clarko helps you quickly set up the initial settings. Railway allows the entire system to run all the time. You don’t have to get up in the middle of the night to stare at the host. In fact, if you already have a trading strategy on hand, you can feed the file into it and set goals such as maximum drawdown and risk performance. Then let the agent look back at the performance every week. Start with micro-reading mode to observe and then decide whether to set a real upper limit. What's more important is that it doesn't mean he is successful if he runs well on the first day. It depends on whether he can continue to make corrections. When the market changes, will he start to drift away? Or is there a way to slowly pull it back? This is also the most easily overlooked part of many people when doing automation. Everyone likes to watch it. The first demonstration is amazing, but they don't like to see whether it is stable in the third week. Hey, this is amazing. Even if you don't let AI help you with the trading, what is really worth copying from Meatball Notes is that everything needs to be defined first. What is success? Letting the system slowly revise based on the results will be much more effective than random guessing every time. You can use it to test investment strategies, find out a few points, or feed cats so that they are least likely to meow like crazy. The principle is the same. Even if you are a content creator, you can also use this idea to test titles, test shelf time, and test which short films are more watched. Set the rules first, then look at the results, and then make corrections. It is much more reliable than just relying on your feelings every day. However, this process also has a risk reminder, that is, if you set the wrong goal at the beginning. The latter system may work very hard in the wrong direction. Just like if you set the goal to only pursue clicks, what you end up with may not be good content, but a bunch of noisy titles. So the problem is often not whether the AI is smart or not, but what you have taught it to chase. Next comes the fourth one, which is a bit fierce. Listen up. This theme is to use Gemini Oni as a direct creative tool, especially in AI videos, virtual avatars, and short audio and video content. It hits very clear pain points. High-quality videos are usually troublesome, time-consuming, and expensive. It often takes several tools to make it work. What Gemini Oni wants to do is to shorten the process. You can generate a video from the prompt word, upload a reference picture, take a short selfie with your mobile phone, create an avatar, and then put the avatar into different situations, trying to maintain the overall consistency of the sound and picture. In other words, it is not just a simple text-generated video. The next-level flexible video production interface allows you to quickly generate, change styles, and customize content without having to become an editor first. To be honest, Creators who run a community by themselves can scan a QR code, record a short selfie, make a reusable avatar, and then use it to make a mini-video, product introduction or dialogue content. It is especially suitable for quickly testing themes when you are in a hurry. You have three themes to test today. Which is the unboxing tone that is more likely to be watched, or the complaint route that is more likely to be forwarded? In the past, you might fall asleep just by shooting. Now you can quickly make three versions to see which response is better, and then decide whether to invest more time. The cost of human trial production is really much lower. I want this Dingding to secretly save it. If you have always wanted to make a small animation with a cat as the protagonist, but are too lazy to edit it frame by frame until the end of time. This kind of creative process is really practical. It can help you fish a lot of time, and you also have the opportunity to make content that can make a lot of money. However, there is also a very realistic reminder here. The more the avatar is made, the more clear the boundaries of licensing and use are. Are you using it to make your own content, or will it be used by others to re-edit and dub it in the future? There is a big difference. Especially if you are a brand editor and lecturer who relies on your personal image to make a living, it is much safer to first figure out whether it can be used commercially and whether it can be sublicensed and how long the data will be stored than to fix it after it is made. Well, here comes the fifth one. DayDay particularly likes the last one. It doesn't look so flashy, but it may be the most practical one. Before teaching AI to write things for you, organize the files first. The pain point he wants to solve is that many AI errors are not caused by poorly written prompt words, so the source data itself is a mess. The contradictory versions have expired information and lack of information, but you still force him to start writing. In the end, it is easy to write a very smooth and confident content, but in fact the risk is very high. The core of this method is to use relatively strong models such as GPT5.5 Opac4.7 Cloud Codex Clark Code. First, find your folder system, read the files, compare the versions, sort out the conflicts, and then start to produce formal documents. To put it bluntly, it is not to drop the first sentence and help me write the forgotten one, but to ask him to make a source list first. Conflict records, missing context lists, duplicate file reports, let you see clearly first which information he thinks is the latest version, and which ones are actually fighting with each other. In practical applications, whether you are a consultant, researcher, supervisor or legal affairs officer, you can first throw the briefing, verbatim draft, spreadsheet, notes, emails into the same project data room, let the agent sort out what is true now and what is problematic, and finally ask him to write the report. You can also think of it as asking the AI to be a coworker who is an organizer first. Don't call him the colleague with good writing skills from the beginning, because many times, what really takes time is not writing the experiment, but the first 200 files. Who is right and who is wrong? Which one is newer? Which one is actually just an old version renamed and re-stored. To be honest, even lazy cats can use this tool, because what AI fears most is not not being smart enough, but making mistakes too confidently. If your desktop is the same as the photo album on your mobile phone, there will always be three versions of the same file. Chapter 12 is similar to the bloody dance called the cat hood. Then this process is really important. Let me add another very practical situation. If you have to make a board report, client proposal or legal response today, what you are most afraid of is not writing slowly but quoting the wrong version. Spending an extra ten minutes to sort out the data sources can often save a lot of back-and-forth revisions later. This time saving is actually very worthwhile. Of course, this does not mean that you can just throw the data in and ignore it completely. No matter how powerful the model is, it is still possible to force vague descriptions into conclusions, so the last step of employee recombination cannot be saved. Especially when it comes to contract figures and content with consequences, it is even more important to retain people to make decisions. If you look at today's five topics together, you will clearly feel that AI is slowly changing from scattered answering tools to a complete environment. It will help you organize, install, and do things. It will continue to operate in the background, directly touch creation, touch business, touch data organization, and even start touching your payment decisions and daily habits. This is very convenient. There is a real chance to make your work faster and your life less laborious, but the premise is that you know what exactly it is helping you and where you can't completely let it go. AI is becoming more and more like a smart cat. It usually stays quiet and suddenly helps a lot at critical moments. But if you don't look at anything, just leave it to it. Well, you may turn your head around and change your home into a different scenery. To put it more directly, the real difference of this wave of tools is not only that the models are larger and have more functions, but that they start to want to take over the process. In the past, you would ask and answer questions. Now if you throw a target, it may help you string it all the way to the back. This change is very convenient, but it also means that you need to know which step can be automatic and which step must be seen by yourself. For those who do not use tools, the battle in the future may not be who can press faster, but who knows where to click and where to pause. So instead of being chased by every mental word, it is better to grasp a principle first. When AI is most valuable, it usually does not do it all for you, but helps you with the most fragmented, annoying and easily missed parts. Clean it up first so that you can have the strength to make the really important judgments. If you want to fish and keep up with AI, remember to subscribe to the AI hot search report. If today's content has gained you something, then help me leave a five-star review on Apple Podcast. FB IG FRED SO AI hot search report. You can find me today. I'm going to take a nap. We'll see you tomorrow. Goodbye, meow.