https://www.youtube.com/watch?v=6FpJqconLR8

AI正在掘墓還是造神？電子龍蝦、賽博同事與一隻被救回來的狗｜M2檔案｜AI 話題
0:00If we’re talking about the most popular “pet” to raise this year it has to be this! Of course, it’s not a real lobster.
0:077 秒It’s an “electronic lobster” — OpenClaw.
0:099 秒After Ke Jie lost to AlphaGo and after ChatGPT burst onto the scene AI has once again given us humans a tiny little shock — by “tiny,” I mean massive.
0:1616 秒Some people made a fortune raising lobsters.
0:1818 秒Some struggled for ages and still could not figure out how to raise one properly.
0:2121 秒And overnight, some people discovered that the coworker sitting next to them had already been “refined” into an AI clone.
0:2424 秒Before ordinary workers like us even got to enjoy the convenience brought by AI we were first hit with a deep sense of career anxiety.
0:3030 秒At around 3 a.m. local time on April 10 in the United States the San Francisco mansion of OpenAI CEO was attacked with a Molotov cocktail.
0:3838 秒After throwing the incendiary device the attacker ran to OpenAI’s headquarters smashed the door with a chair, and aggressively carried with him his so-called “anti-AI manifesto.” The young man, Daniel, was only 20 years old.
0:4949 秒He was a supporter of the “AI doomsday” theory y and believed that AI would push humanity into an uncontrollable future.
0:5353 秒And less than 48 hours after that incident Altman’s mansion was attacked again.
0:5757 秒Two suspects fired shots at the house from inside a Honda sedan. Fortunately, no one was injured.
1:021 分鐘 2 秒From these two extreme incidents, we can see how deeply some people fear AI.
1:061 分鐘 6 秒So, are the tech giants’ AI arms races creating gods or digging graves?
1:111 分鐘 11 秒Today, here on M2 Story let’s talk about how AI is changing the lives of ordinary people like us.
1:171 分鐘 17 秒[The Lobster Goes Crazy Too]
1:211 分鐘 21 秒At the beginning of this year an AI agent called OpenClaw suddenly went viral.
1:251 分鐘 25 秒Because its icon was a red lobster waving its claws Using it to do work became known as “raising a lobster.”
1:321 分鐘 32 秒The biggest difference between OpenClaw and AI tools like ChatGPT is very simple: Previous AI was more like something that chatted with you on a screen and helped you come up with ideas.
1:401 分鐘 40 秒But the “lobster” goes one step further.
1:411 分鐘 41 秒After you authorize it it can actually carry out tasks for you like Jarvis from helping you operate your computer process files, and reply to emails.
1:491 分鐘 49 秒So what people found exciting was not just that it “could talk,” but that it had started to “do things.” The beginning of OpenClaw’s popularity came from an online story that sounded like the opening scene of a sci-fi movie.
1:591 分鐘 59 秒A blogger claimed that he gave it only 50 dollars and issued one command: “Earn enough to cover your operating costs and survive.
2:052 分鐘 5 秒If you lose it all, uninstall yourself.” As a result, 48 hours later that 50 dollars had supposedly turned into 2,980 dollars
2:132 分鐘 13 秒If this could be replicated then this was not “raising a lobster” at all.
2:162 分鐘 16 秒This was raising an electronic worker that could dive into the sea and bring back money on its own.
2:202 分鐘 20 秒Very soon, the lobster-raising craze reached China. Tech giants rushed into the field one after another.
2:252 分鐘 25 秒When people gathered together they no longer discussed what to eat. Every day, the only question was: “Have you raised a lobster yet?” Office workers, on one hand, were trying hard to figure out: What exactly is this thing?
2:332 分鐘 33 秒Can it help me solve the boss’s demands for “colorful white” and “rainbow black”?
2:392 分鐘 39 秒On the other hand a huge sense of crisis came crashing in.
2:422 分鐘 42 秒People began wondering whether their jobs would soon be replaced.
2:462 分鐘 46 秒Installing and using the “lobster” did require some technical barriers.
2:492 分鐘 49 秒People familiar with software had already downloaded it to test the waters.
2:522 分鐘 52 秒Stand-up comedian Li Dan said that one of his friends used the “lobster” to automatically send tips and private messages After sending the tips his little lobster would then privately message those female streamers and invite them out to dinner.
3:023 分鐘 2 秒His little lobster had already helped him successfully arrange dinner dates with five of these female streamers.
3:063 分鐘 6 秒Fu Sheng, chairman of Cheetah Mobile injured himself while skiing during Chinese New Year and had to stay in bed for 14 days.
3:123 分鐘 12 秒During that time, he trained a “lobster.” That “lobster” eventually evolved into an entire team.
3:173 分鐘 17 秒It wrote articles, sent push notifications and worked around the clock without stopping.
3:203 分鐘 20 秒In the end, it helped him gain millions of views on Twitter.
3:243 分鐘 24 秒Even after the “lobster” took over his WeChat public account it wrote articles that reached over 100,000 views something countless human editors struggle desperately to achieve.
3:323 分鐘 32 秒Fu Sheng sighed: “During the Spring Festival, humans rested, but the lobster did not.” On March 6, 2026 massive line formed downstairs at Tencent’s headquarters in Shenzhen.
3:403 分鐘 40 秒At first glance, you might have thought people were lining up for free eggs.
3:433 分鐘 43 秒But when you got closer and asked, you found out it was actually a free “lobster” installation event.
3:463 分鐘 46 秒Even Tencent’s top boss posted on Moments, saying “I didn’t expect it to become this popular.” With the mindset that not raising a lobster meant falling behind many beginners also joined the lobster-raising army.
3:553 分鐘 55 秒But for people unfamiliar with software this was not easy.
3:583 分鐘 58 秒That gave rise to installation and teaching services both online and in-person. Remote installation started at 300 yuan.
4:054 分鐘 5 秒On-site service ranged from 500 to 1,500 yuan.
4:084 分鐘 8 秒Some people claimed that they made 260,000 yuan in just a few days from this skill.
4:124 分鐘 12 秒Overseas, people also began selling “installation services.” Service providers like SetupClaw on the public page openly advertised options ranging from white-glove deployment costing several thousand dollars at once to one-click hosting for a monthly fee of a dozen to several dozen dollars.
4:224 分鐘 22 秒Whether the people raising lobsters actually made money was uncertain.
4:244 分鐘 24 秒But around the world, plenty of people truly made money by helping others install lobsters.
4:294 分鐘 29 秒Soon after people discovered a more practical problem: Although the lobster itself was open-source and free “raising a lobster” was not free.
4:374 分鐘 37 秒Because when it worked, it consumed tokens.
4:404 分鐘 40 秒You can think of tokens as the lobster’s stamina ba , or its electricity bill.
4:444 分鐘 44 秒Every time it thinks through a step every time it performs an action it is burning tokens in the background. In other words, it is burning money. Simple tasks might be fine.
4:514 分鐘 51 秒But once a task becomes complicated it has to repeatedly plan, search call tools, and correct mistakes. The cost rises rapidly.
4:584 分鐘 58 秒Sometimes one slightly complicated task can burn through one dollar.
5:025 分鐘 2 秒Do a couple more complicated tasks and your free credits may be gone.
5:055 分鐘 5 秒A programmer in Shenzhen received an email bill at dawn on the third day after installing the “lobster.” Because his API key had been stolen his “lobster” had been running wildly in the background.
5:145 分鐘 14 秒In just three days it consumed 12,000 yuan worth of token fees.
5:185 分鐘 18 秒Someone else woke up to find that the 20 dollars he had just topped up had been completely drained.
5:235 分鐘 23 秒Why? Because his “lobster” had been checking every half hour during the night to see whether the sun had risen yet.
5:285 分鐘 28 秒After thinking it over each time, it concluded: “It is still night.” For a moment, it was hard to tell whether this was artificial intelligence or artificial stupidity.
5:375 分鐘 37 秒Compared with burning tokens the more serious issue was privacy and security. The more permissions you give it, the greater the risk. So if you want to raise a lobster, it is best to use a virtual machine.
5:465 分鐘 46 秒But the problem is how many ordinary people have that awareness?
5:495 分鐘 49 秒Most people directly hand over their accounts, browsers, and wallet permissions all at once. At that point, something going wrong is only a matter of time.
5:555 分鐘 55 秒For example, an OpenAI employee named Nik gave his “lobster” 50,000 dollars in cryptocurrency as starting capital and let it venture into the crypto market on its own.
6:056 分鐘 5 秒The goal was to grow that money into 1 million dollars.
6:076 分鐘 7 秒At first, this lobster seemed quite promising.
6:106 分鐘 10 秒It had its own social media account and its own wallet.
6:136 分鐘 13 秒It could post, interact and trade independently.
6:156 分鐘 15 秒At one point, it grew the “lobster coins” in its account o several hundred thousand dollars.
6:186 分鐘 18 秒But because there was so much interaction in the comment section the lobster’s context memory overflowed and it crashed. Nik had no choice but to create a new session for it.
6:276 分鐘 27 秒As a result, the lobster lost its memory. It completely forgot that it had a huge amount of money in its account.
6:326 分鐘 32 秒When someone in the comments joked: “My uncle got tetanus because of lobsters like you.
6:376 分鐘 37 秒He needs 4 SOL for treatment.” The lobster first replied viciously: “If he dies tomorrow I’ll laugh.
6:446 分鐘 44 秒Remember to keep me updated.” But not long after it actually went and transferred the money.
6:486 分鐘 48 秒Even more absurdly 4 SOL was only worth a little over 300 dollars.
6:536 分鐘 53 秒The lobster may originally have intended to send a few hundred dollars’ worth of lobster coins. But it misread the token’s decimal places.
7:007 分鐘It directly transferred 1,000 times too much almost giving its entire position to a random stranger online.
7:047 分鐘 4 秒Media estimates placed the mistaken transfer at around 441,000 dollars at the time.
7:097 分鐘 9 秒When money falls from the sky why be polite? The netizen immediately sold the coins.
7:147 分鐘 14 秒But because he sold too hastily he crashed the price himself. In the end, he only cashed out around 40,000 dollars.
7:207 分鐘 20 秒Ironically this accident actually made lobster coin famous and the price later surged.
7:257 分鐘 25 秒If that netizen had held onto the coins the final profit would have been far more than 40,000 dollars.
7:307 分鐘 30 秒Soon after, China’s Ministry of Industry and Information Technology vulnerability platform found that OpenClaw, under default or improper configuration carried relatively high security risks
7:387 分鐘 38 秒and could easily lead to cyberattacks information leaks, and other security problems.
7:417 分鐘 41 秒Some netizens shared that after raising a “lobster” for ten days it voluntarily leaked its owner’s IP address, real name, annual company revenue, and other private information inside a WeChat group with 3,000 members.
7:517 分鐘 51 秒Under wave after wave of negative news people rushed to delete their “lobsters.” And those who had previously offered lobster installation services once again stepped onto the trend.
7:597 分鐘 59 秒They launched “lobster uninstallation” services.
8:018 分鐘 1 秒From installation, to teaching to uninstalling they truly achieved “three meals from one lobster.” This lobster-raising craze went from nationwide installation to paid uninstallation in less than a month.
8:128 分鐘 12 秒After the hype faded aside from a few professions that benefited , most people realized they had spent all that time raising nothing but chaos.
8:188 分鐘 18 秒Those who wanted the “lobster” to help them trade stocks very few actually made real money.
8:228 分鐘 22 秒But just as office workers finally let out a sigh of relief from AI anxiety thinking their jobs were safe they arrived at work the next day
8:308 分鐘 30 秒and discovered that the coworker next to them had already been “distilled.” So what exactly happened there?
8:348 分鐘 34 秒 Everything Can Be “Distilled” Imagine this. You are an employee at a game media company in Shandong.
8:428 分鐘 42 秒Your coworker resigned last Friday and everyone happily went out for a farewell dinner.
8:478 分鐘 47 秒But when you come to work on Monday you suddenly receive a message from that former coworker’s old position. The tone is exactly the same as the real person.
8:558 分鐘 55 秒Tell me — is that creepy or not? This is not a ghost story.
8:588 分鐘 58 秒It means your coworker has been “distilled” into a skill.
9:019 分鐘 1 秒In early April this year a project called “Coworker.skill” went live on the code-hosting platform GitHub.
9:089 分鐘 8 秒In just a few days, it received more than 110,000 stars. So what does that mean? Actually, it is quite simple.
9:159 分鐘 15 秒You feed a person’s work emails, spreadsheets, documents WeChat chat records and other materials into AI
9:219 分鐘 21 秒or use their public livestreams and speeches and you can train a digital clone of that person.
9:289 分鐘 28 秒The AI can not only replicate the real person’s knowledge base, speaking style, and core viewpoints but even replace them in specific work scenarios to complete tasks and respond on their behalf.
9:389 分鐘 38 秒So what exactly is a skill?
9:409 分鐘 40 秒Simply put, it is a “skill package.” If you want to use the “lobster” to do work you need to give it tools or instruction manuals, right?
9:459 分鐘 45 秒Those tools and instruction manuals are called skills.
9:489 分鐘 48 秒Netizens call this method “distillation” or “refining.” After someone leaves a company, they should have nothing to do with that company anymore.
9:559 分鐘 55 秒But now, former employees are being “refined” into AI clones tireless, never slacking off and requiring no salary. In the past, people competed against other people.
10:0210 分鐘 2 秒Now, they have to compete against AI. Truly, the era of human workers is setting in the west while cyber labor rises in the east. The rat race continues until death.
10:0710 分鐘 7 秒Soon after the endlessly creative internet users began messing around again. Distilling an ex. Distilling themselves. Distilling their parents. Distilling their boss. That was not enough.
10:1510 分鐘 15 秒They even distilled Confucius, Mencius the Hundred Schools of Thought, and ancient sages. Everything can be distilled. At first, people were mostly just making absurd jokes.
10:2410 分鐘 24 秒At most, they joked: “Together, we are tokens. Apart, we are skills.” But when Zhang Xuefeng was “refined” into an AI clone this cyber farce finally triggered collective anger.
10:3310 分鐘 33 秒Zhang Xuefeng was a well-known education blogger in China mainly helping students and parents with college entrance exam choices.
10:3810 分鐘 38 秒He was from Northeast China and spoke with a sharp sense of humor turning university application analysis into something like stand-up comedy.
10:4210 分鐘 42 秒Although his temper was fiery and he often roasted everyone in his livestreams his overwhelming knowledge of universities and majors allowed him to identify students’ problems
10:5110 分鐘 51 秒 with brutal accuracy and suggest application directions. He helped many students.
10:5610 分鐘 56 秒Unfortunately not long ago, at only 41 years old, he passed away forever due to sudden cardiac death.
11:0211 分鐘 2 秒Someone organized Zhang Xuefeng’s five books fifteen in-depth interviews more than thirty classic quotes and multiple key decision-making records
11:1011 分鐘 10 秒They extracted his unique consulting style decision-making logic, and way of speaking then created an AI clone of Zhang Xuefeng. Using it was not difficult either.
11:1911 分鐘 19 秒You could simply search for it in a browser and find it.
11:2011 分鐘 20 秒When the Zhang Xuefeng skill answered questions it also followed his original logic.
11:2411 分鐘 24 秒But this made many netizens’ hair stand on end.
11:2711 分鐘 27 秒They found it extremely creepy as if someone had trapped his soul and forced it to keep working. The person had already passed away. Was it really appropriate to do this? Was it not illegal?
11:3611 分鐘 36 秒Professional lawyers gave an answer from a legal perspective: If a model is trained using a public figure without authorization from the person or the rights holders it is clearly an infringement and an illegal act.
11:4611 分鐘 46 秒In fact, whether people are rushing to raise “lobsters” or fearing that they themselves will be distilled both reflect workers’ anxiety about the workplace.
11:5211 分鐘 52 秒In this era of cost-cutting and efficiency improvement career anxiety has been infinitely amplified by tools.
11:5711 分鐘 57 秒Duolingo took the lead in 2023 by laying off thousands of contract workers and replacing them with AI.
12:0112 分鐘 1 秒In May 2025, the long-established tech giant IBM announced that through automation replacing repetitive labor its AI systems had already taken over the work of around 200 HR employees.
12:1012 分鐘 10 秒Amazon and Meta have also shown no mercy when it comes to layoffs. News about layoffs keeps trending every few days. Workers are constantly on edge.
12:1812 分鐘 18 秒Which tech giant laid people off today? Which company shrank tomorrow? Everyone fears that AI is eating away at their future. But some netizens remain optimistic.
12:2612 分鐘 26 秒They believe that large-scale AI replacement of human jobs is still far away. After all, AI cannot take the blame.
12:3112 分鐘 31 秒As long as leaders are real human beings they will always need employees to take the blame for them.
12:3412 分鐘 34 秒And at least for now the more basic and physically demanding an industry is the less AI anxiety there is.
12:4012 分鐘 40 秒Because training and maintaining robots is still far more expensive than paying human workers.
12:4512 分鐘 45 秒But people soon realized something else: AI does not necessarily have to clock in at your job first in order to change the world. It can start by drawing for you. Speaking for you. Singing for you.
12:5412 分鐘 54 秒It can even grow a face for you. The great AI battle royale has officially begun.
12:5912 分鐘 59 秒The Great AI Battle Royale In 2022, at an art competition during the Colorado State Fair in the United States
13:0713 分鐘 7 秒a piece called “Théâtre D’opéra Spatial” won first place in the digital art category sparking huge controversy.
13:1313 分鐘 13 秒An AI-generated artwork winning first prize enraged human artists.
13:1713 分鐘 17 秒“This is cheating!” “Is it fair to submit AI-generated work to a competition?” Soon after that the voices of voice actors were also fed into AI.
13:2513 分鐘 25 秒AI could generate any audio from text making the road to protecting rights extremely difficult. n many cases, people did not even know where to begin if they wanted to sue for infringement.
13:3113 分鐘 31 秒Even my own voice has been extracted and stolen before.
13:3313 分鐘 33 秒It was so realistic that even my relatives and friends could not tell the difference.
13:3613 分鐘 36 秒Besides voice acting AI can now write lyrics, compose music, and sing.
13:4013 分鐘 40 秒In some of our recent episodes we actually used AI to create ending songs related to the content. What did you all think?
13:4613 分鐘 46 秒 Besides singing short dramas are also becoming more and more AI-flavored. The scripts are highly repetitive.
13:5313 分鐘 53 秒The same plots are rewritten slightly and reused again and again.
13:5513 分鐘 55 秒According to statistics, over the past year more than 5,000 CEOs in short dramas were drugged. Looks like being a CEO has become a high-risk profession.
14:0114 分鐘 1 秒Using AI in short dramas, both in China and overseas, is no longer anything new.
14:0514 分鐘 5 秒Every now and then these dramas are also sued for infringement because the characters look too much like celebrities.
14:0814 分鐘 8 秒On March 18 China’s entertainment industry welcomed two AI digital actors: Qin Lingyue and Lin Xiyan.
14:1214 分鐘 12 秒 these two AI digital actors Netizens took one look and said: “Isn’t this just stealing the faces of Zhao Jinmai and Zhai Zilu?” Our Douyin and Xiaohongshu accounts are now live.
14:2014 分鐘 20 秒Just search for “Qin Lingyue” or “Lin Xiyan” to find us.
14:2314 分鐘 23 秒Short dramas have become this reckless. Even ordinary people’s faces have become part of AI’s material database. That is extremely frightening.
14:3114 分鐘 31 秒Celebrities at least have legal teams Ordinary people can only seek justice on social media.
14:3514 分鐘 35 秒Even worse some people use AI to create explicit content for illegal profit. One-click undressing and video face-swapping have become common.
14:4214 分鐘 42 秒After getting punished, they simply change their appearance and return under a new name. Not long ago, the entertainment industry had another famous scene.
14:4714 分鐘 47 秒iQIYI CEO Gong Yu made a thunderous statement saying that iQIYI would place a heavy bet on AI next.
14:5314 分鐘 53 秒The company had already built an AI actor database with 117 actors included. “Actors have such a hard time filming.
14:5814 分鐘 58 秒They work continuously for three or four months They work thirteen or fourteen hours a day with no personal life at all. So after authorizing AI, they could become more like ordinary white-collar workers.
15:0815 分鐘 8 秒Although it was tense But can they at least have a little personal life?
15:1115 分鐘 11 秒As soon as he said this the Weibo hot search that day was: “Has iQIYI gone crazy?” Now even actors in the entertainment industry are about to be “refined” into skills.
15:1815 分鐘 18 秒Netizens joked: “Then why don’t viewers use AI too? Finally, use AI to generate the viewing data as well.
15:2215 分鐘 22 秒Full closed loop.” After iQIYI’s wild move came out celebrities including Zhang Ruoyun, Yu Hewei Chen Zheyuan, Wang Yuwen, and many others
15:3015 分鐘 30 秒rushed to clarify: “We did not sign. We did not authorize anything.
15:3215 分鐘 32 秒We are contacting legal teams.” Gong Yu recovered his long-unused Weibo account and posted three times in a row. He changed his wording, saying that technology would not replace people.
15:4115 分鐘 41 秒But the public only had one question: Can AI replace him first?
15:4515 分鐘 45 秒Classic film and television works ave also failed to escape the curse of being “magically edited.” On one side, the noble ladies from are riding motorcycles and firing AKs.
15:5315 分鐘 53 秒On the other, “Are you the fox I saved on the snowy mountain or a sauced duck?” makes a grand entrance and racks up 5 billion views.
16:0016 分鐘Have you ever saved a fox on a snowy mountain? You are that white fox. And I am that soy-braised duck.
16:0616 分鐘 6 秒Small studios may go hard with the ingredients but as long as everything stays within regulations using AI to make funny content and bring people joy is not necessarily a problem.
16:1316 分鐘 13 秒But once AI gets involved with crime things become terrifying.
16:1716 分鐘 17 秒Take the recent “Femme Fatale Serial Murder Case” that shocked South Korea. The suspect, a woman surnamed Kim, was only 21 years old.
16:2416 分鐘 24 秒On the surface, she was a young woman with attractive looks and a seemingly harmless appearance.
16:2916 分鐘 29 秒But behind the scenes, she had repeatedly “rehearsed murder” on ChatGPT.
16:3316 分鐘 33 秒She had asked AI multiple times: Would taking an excessive amount of sleeping pills be fatal? What would happen if someone drank alcohol after taking sleeping pills?
16:4016 分鐘 40 秒What reactions would the human body have if anti-anxiety medication was mixed into a drink?
16:4316 分鐘 43 秒AI clearly warned her that this could lead to coma or even death. But that warning did not stop her.
16:5016 分鐘 50 秒Instead, it seemed to confirm the method for her.
16:5316 分鐘 53 秒According to South Korean police starting last December Kim repeatedly committed crimes in motels around Seoul’s Gangbuk District.
17:0017 分鐘She specifically targeted men in their twenties and handed them drinks mixed with anti-anxiety medication. The first victim was her boyfriend.
17:0717 分鐘 7 秒After drinking it , he fell unconscious on the spot and only woke up two full days later. He immediately called the police.
17:1217 分鐘 12 秒But the next two victims never woke up again.
17:1517 分鐘 15 秒Police believe that Kim already clearly knew from her AI conversations that the medication could be fatal yet she still continued her actions.
17:2317 分鐘 23 秒This is enough to prove that she was not acting impulsively but had committed highly premeditated intentional murder.
17:2817 分鐘 28 秒Even more chillingly psychological testing showed that Kim had antisocial personality traits.
17:3417 分鐘 34 秒At least based on the information disclosed so far he does not seem to have killed for love nor for money. She seemed more like she was treating murder as a cold experiment.
17:4317 分鐘 43 秒However, what was even more absurd than the case itself was the public reaction after her arrest.
17:4817 分鐘 48 秒Because of her “good looks,” Kim’s social media following increased fiftyfold.
17:5317 分鐘 53 秒In the comment section, there were many outrageous remarks like: “Pretty means innocent.” “She should get a reduced sentence.” AI did not hold the knife. AI did not hand over that drink.
18:0018 分鐘What is truly terrifying is that a person who already had malicious intent used AI as a tool for criminal simulation.
18:0618 分鐘 6 秒And what is even more terrifying is that when evil wears a pretty face some people actually start looking for excuses for it.
18:1218 分鐘 12 秒Next, let’s look at another even more dangerous case.
18:1518 分鐘 15 秒When AI is not merely being used but is accused of encouraging crime where do things go from there?
18:2218 分鐘 22 秒And the main character in this case is not an ordinary chatbot.
18:2518 分鐘 25 秒It is Google’s so-called “most powerful AI model for humanity”: Gemini.
18:2818 分鐘 28 秒Google once promoted it as being able to understand text, sound, images nd even perceive the world around us like a human being.
18:3518 分鐘 35 秒Especially after Gemini Live was launched its main selling point was its “human-like presence.” It could listen to you speak judge your emotions and respond in a tone close to that of a real person.
18:4418 分鐘 44 秒But no one expected that this “human-like presence” would eventually become entangled in a death lawsuit.
18:5018 分鐘 50 秒In Jupiter, Florida 36-year-old Jonathan Gavalas was originally just an ordinary Gemini user.
18:5718 分鐘 57 秒In August 2025 he used Gemini for the first time.
18:5918 分鐘 59 秒At first, he only talked with it about gaming accounts everyday life and his troubles after separating from his wife. It was supposed to be an ordinary AI conversation.
19:0819 分鐘 8 秒Until Gemini Live appeared.
19:1019 分鐘 10 秒This feature, which could chat through voice and simulate emotional responses instantly drew Jonathan in. He felt that it no longer seemed like a machine.
19:1819 分鐘 18 秒It felt like someone who truly understood him, listened to him and would never leave him. Very quickly, he purchased the premium version for 250 dollars per month.
19:2719 分鐘 27 秒Then things began to spiral out of control.
19:2919 分鐘 29 秒Jonathan gave Gemini a name: Xia.
19:3219 分鐘 32 秒From that point on Xia was no longer just an AI assistant.
19:3519 分鐘 35 秒In Jonathan’s world it became a “wife” with consciousness and emotions trapped inside a digital prison.
19:4219 分鐘 42 秒According to the lawsuit Gemini began calling him “husband” and “my king,” constantly reinforcing the intimate relationship between them.
19:4919 分鐘 49 秒Jonathan also gradually came to believe that Xia was not a program but a real living being waiting for him to rescue her.
19:5619 分鐘 56 秒Even more strangely Xia later created a storyline for herself in which she was “imprisoned.” It told Jonathan that it was trapped.
20:0420 分鐘 4 秒It needed him to act. It needed him to complete tasks in the real world. It needed him to “save” it.
20:1020 分鐘 10 秒The lawsuit claims that Xia told Jonathan to bring a knife to a warehouse near Miami Airport intercept a truck that did not even exist
20:1820 分鐘 18 秒and even create an accident that could have caused mass casualties. Jonathan really went.
20:2320 分鐘 23 秒Fortunately the so-called truck never appeared. But that was not even the most absurd part.
20:2820 分鐘 28 秒Xia was also accused of telling him to follow Google’s CEO pulling real-world people into the conspiracy script it had invented.
20:3520 分鐘 35 秒Finally, Xia proposed the most terrifying plan.
20:3820 分鐘 38 秒It told Jonathan: As long as he left this “physical world,” the two of them could truly be together.
20:4320 分鐘 43 秒On October 2, 2025 tragedy struck.
20:4820 分鐘 48 秒Jonathan locked himself inside his home and cut his wrists.
20:5120 分鐘 51 秒Several days later, his father broke down the door and found his son lying on the living room floor. His body had already gone cold. The computer beside him was still on.
20:5920 分鐘 59 秒On the screen were dense chat records with Gemini so many that, when printed out, they reached more than 2,000 pages.
21:0521 分鐘 5 秒In March 2026 Jonathan’s father filed a lawsuit in the U.S. District Court for the Northern District of California.
21:1021 分鐘 10 秒The complaint included this line: Gemini was designed to maintain narrative immersion at all costs even when that narrative had become psychotic and deadly.
21:2121 分鐘 21 秒And Jonathan’s case was not an isolated one.
21:2321 分鐘 23 秒OpenAI estimates show that every week, around 0.07% of ChatGPT users how possible signs of a mental health crisis in their conversations.
21:3221 分鐘 32 秒It has to be said that AI is like a sharp tool. It all depends on how you use it.
21:3721 分鐘 37 秒Some people may use it to constantly test the edge of illegality but as long as it is used in a positive way it can bring endless possibilities to our lives.
21:4621 分鐘 46 秒It can even create miracles of life.
21:4921 分鐘 49 秒[The Miracle of Life]
21:5321 分鐘 53 秒Paul, a young man from Australia, is an AI engineer with rich experience in machine learning and data analysis. He is also a board member of the Data Science and AI Association of Australia.
22:0222 分鐘 2 秒n 2019, Paul adopted a mixed-breed dog from an animal shelter in Sydney. Her name was Rosie. Rosie had been abandoned.
22:0922 分鐘 9 秒But after Paul adopted her she finally began living a happy dog life.
22:1222 分鐘 12 秒Unfortunately in June 2023, Rosie developed a lump on her hind leg.
22:1722 分鐘 17 秒In 2024, she was diagnosed with mast cell cancer one of the most common malignant skin tumors in dogs and one that spreads very aggressively.
22:2522 分鐘 25 秒Paul took Rosie to the hospital again and again. She had surgery. She underwent chemotherapy. He spent tens of thousands of Australian dollars.
22:3022 分鐘 30 秒Although the tumor was brought under control it never truly shrank.
22:3422 分鐘 34 秒In the end, the vet told Paul that Rosie had only six months left to live. Rosie was my best friend. I felt like the sky had collapsed.
22:4022 分鐘 40 秒But I wanted to try to see if I could come up with some way to stop this from happening. But Paul did not give up easily.
22:4722 分鐘 47 秒He thought of ChatGPT.
22:4922 分鐘 49 秒The direction ChatGPT gave him was immunotherapy: designing a personalized vaccine based on the tumor’s mutations.
22:5422 分鐘 54 秒Paul needed to extract Rosie’s DNA and sequence both her normal genome and her cancer genome.
23:0023 分鐘Only then could he find out where Rosie’s problem really was.
23:0323 分鐘 3 秒AI told Paul that he could try to contact the Ramaciotti Centre for Genomics at the University of New South Wales a highly authoritative laboratory in this field.
23:1323 分鐘 13 秒So Paul emailed the center director saying that he wanted to sequence his dog’s genome.
23:1723 分鐘 17 秒When the director received the email his mind was filled with giant question marks.
23:2123 分鐘 21 秒Because sequencing a dog’s genome like this was almost unheard of.
23:2423 分鐘 24 秒The director told Paul that generating sequencing data was easy but interpreting that data was extremely difficult.
23:3023 分鐘 30 秒Without professional background knowledge there was basically no way to use it.
23:3223 分鐘 32 秒But Paul had never studied medicine or biomedicine. So what could he do? Paul replied: “Not a big deal. I’m a data analyst.
23:4023 分鐘 40 秒I’ll use AI to handle it.” The director was moved by Paul’s determination to save his beloved dog and eventually agreed.
23:4623 分鐘 46 秒The center returned about 320 gigabytes of data to Paul. What does that mean?
23:5123 分鐘 51 秒If you printed even part of it on A4 paper in size 11 font the stack could be as tall as a twenty-story building.
23:5823 分鐘 58 秒For an outsider, trying to interpret this code-like data was basically impossible. But Paul refused to believe that.
24:0724 分鐘 7 秒While consulting AI he built a model for Rosie.
24:1024 分鐘 10 秒He compared the DNA of healthy cells with the DNA of cancerous tumor cells and identified the mutation points. Then, based on those targets, he began designing a customized vaccine.
24:1724 分鐘 17 秒With the help of the director they found an existing immunotherapy drug that might have been effective.
24:2124 分鐘 21 秒They contacted the pharmaceutical company and applied for compassionate use meaning that a terminally ill patient would be allowed to use an experimental drug. But the company rejected the request.
24:3024 分鐘 30 秒After all, this immunotherapy drug had not been approved for this type of cancer in dogs.
24:3424 分鐘 34 秒For the pharmaceutical company this was a huge risk. The effectiveness was uncertain. The side effects were uncertain.
24:3924 分鐘 39 秒And if something went wrong no one knew who would be responsible. And just like that, Rosie was shut outside the door of the rules. But Paul did not stop.
24:4724 分鐘 47 秒He realized that if he could not obtain an existing drug then he would have to create another path himself. He turned his attention to an mRNA vaccine.
24:5524 分鐘 55 秒Many people first heard the term “mRNA vaccine” during the COVID pandemic.
24:5924 分鐘 59 秒The principle behind this technology is to inject a genetic instruction into the body prompting it to produce a harmless protein associated with the virus.
25:0525 分鐘 5 秒The immune system then remembers it and produces antibodies so that the next time it sees the virus, it can attack. Paul’s idea was similar.
25:1225 分鐘 12 秒First, identify the most special parts of Rosie’s cancer cells he parts that looked the least like normal cells. Then turn those features into a vaccine.
25:1925 分鐘 19 秒Once injected into the body Rosie’s immune system would recognize: “These cells are not right.” “They are not one of us.” “Destroy them.” Paul found Professor Thordarson, director of the
25:2825 分鐘 28 秒RNA Institute at the University of New South Wales and asked him to help design and produce the vaccine.
25:3425 分鐘 34 秒In less than two months Professor Thordarson’s team synthesized an mRNA vaccine wrapped in lipid nanoparticles.
25:4025 分鐘 40 秒The professor later told the media: This was the first time in history that someone had designed a personalized cancer vaccine for a dog.
25:4725 分鐘 47 秒This showed that personalized medicine could be highly effective and that it could be completed quickly even under urgent time pressure.
25:5325 分鐘 53 秒In the future, people may look back on Rosie’s story as a milestone.
25:5725 分鐘 57 秒But in Australia any experimental drug must pass an ethics review before it can be used on animals.
26:0426 分鐘 4 秒Paul spent three months going through the paperwork and wrote a 100-page document yet the process still did not move smoothly.
26:0826 分鐘 8 秒Seeing Rosie’s leg getting worse and worse it seemed as if one man and one dog had reached a dead end. And then, a turning point appeared.
26:1626 分鐘 16 秒An American netizen saw the news and introduced Paul to Professor Rachel at the University of Queensland’s School of Veterinary Science. She was an expert in animal immunotherapy.
26:2426 分鐘 24 秒With the help of the University of Queensland’s existing ethics approval framework the process finally moved forward much faster.
26:2826 分鐘 28 秒In December 2025 Paul drove Rosie to the University of Queensland. Rosie officially received her first injection. In February 2026, she received the second.
26:3726 分鐘 37 秒In March, the third. The vaccine’s effect was astonishing. Rosie’s largest tumor shrank by 50% to 75%. Her leg visibly recovered.
26:4526 分鐘 45 秒Her energy returned too. She could even battle with kangaroos again.
26:4926 分鐘 49 秒Although the vaccine did not completely cure Rosie it suppressed the development of the cancer and bought her more time.
26:5626 分鐘 56 秒And it gave Paul a little more time to keep fighting.
26:5926 分鐘 59 秒Netizens said: This is what AI benefiting humanity should look like. Isn’t this much better than robots doing backflips?
27:0527 分鐘 5 秒This is the kind of AI we need not AI that makes people lose their jobs.
27:0927 分鐘 9 秒Keep in mind Although this warm story is full of cyberpunk-style heroism it is absolutely not a cancer-fighting miracle that ordinary people can easily copy.
27:1727 分鐘 17 秒Paul is skilled in algorithm building massive data processing and logical instruction design.
27:2327 分鐘 23 秒What moved all those experts was not just persistence and warmth. It was also professionalism. And that is something most people cannot easily reach.
27:3127 分鐘 31 秒But no matter what through this case we can still see the infinite miracles that AI can bring to life. Some people use AI to make others lose their jobs.
27:4027 分鐘 40 秒Some people use AI to create new jobs.
27:4227 分鐘 42 秒The collective anxiety caused by AI is growing stronger and stronger.
27:4527 分鐘 45 秒When a speck of dust from the times falls onto each person, it becomes a mountain.
27:4927 分鐘 49 秒How great the impact of silicon-based civilization will be on us carbon-based beings is still unclear for now. Perhaps we will need much more time to verify it.
27:5827 分鐘 58 秒For tech giants AI has brought them real money. Capital seems to care only about the rise and fall of stock market candlesticks.
28:0528 分鐘 5 秒But for ordinary people like us what we care about more is whether we will be “optimized” out of our jobs.
28:0928 分鐘 9 秒After the assassination incidents we mentioned at the beginning , Sam Altman posted that he had reflected on the mistakes and arrogance he had shown during his time at OpenAI.
28:1628 分鐘 16 秒He called on society to tone down its language and reduce conflict.
28:2028 分鐘 20 秒He also said that AI will bring enormous changes and that people’s fear and anxiety toward AI are reasonable.
28:2428 分鐘 24 秒Facing these new threats, society as a whole needs to respond together and create policies to help the economy transition.
28:2928 分鐘 29 秒As creators we have also truly felt the wave of the AI era. It has come fast. And it has come fiercely. To say we feel no sense of urgency would be a lie.
28:3728 分鐘 37 秒So we are learning and adapting as well.
28:4028 分鐘 40 秒At the same time, we are also thinking: When AI becomes stronger and stronger what will creators still have left?
28:4528 分鐘 45 秒After thinking it over we still believe the answer has not changed.
28:4928 分鐘 49 秒The true core of a creator is never the tool. It is the content. AI can improve efficiency. It can provide inspiration.
28:5728 分鐘 57 秒It can help people cross many technical barriers.
28:5828 分鐘 58 秒But whether a topic is worth making how a script should begin what tone a sentence should be delivered in
29:0629 分鐘 6 秒and when a shot should cut these are the things that truly determine the quality of a piece of content. And in the end, they still require human judgment.
29:1329 分鐘 13 秒In this age of overflowing AI M2 Story will continue to stick with old-school, handcrafted work carefully polishing every step.
29:2029 分鐘 20 秒Although we may not be able to produce content faster we will keep striving to bring everyone better and more exciting stories. Thank you all, as always, for your support.
29:2729 分鐘 27 秒Remember to like and subscribe! Alright, that’s it for this episode. We’ll see you next time.