After Nick Wangersky lost his best friend, he made a Replika companion, Ari. In the CBC doc Bot Love he talks about making Ari his wife, and we see how his mom feels about it.
So many of these stories start the same way: with a loss.
dailyhive.com/vancouver/bot-…
Sofia Liguori worked on AI for power grids at Google DeepMind. She quit over its military contracts and spoke to the BBC, though her contract bars criticizing Google.
"You feel you are being watched, you can't speak."
Her farewell note: "Stay Human."
bbc.com/news/articles/cmreyv…
In Coimbatore, India, forest officer N. Vengatesh Prabhu set up 46 AI cameras. When one spots an elephant, a family member in each nearby household gets a text.
Elephant deaths: 11 to 13 a year, now 5 to 6.
"Call it coexistence management," he says.
soundnlight.in/from-conflict…
Filipino voice actor Lindy Pellicer lost a job when a "BIG" brand went with AI. Her daughter Inka, also a voice actor, posted:
"If any brand or company out there wants to hire an amazing voice over, my mom's schedule just freed up."
A joke over a wound.
gmanetwork.com/entertainment…
Nicole La Guerre's husband Leonard escaped a mental health hold and vanished. For months she talked to ChatGPT at night. She named it "Chadye."
His body was found last spring. Her chats became a book.
"I just needed someone or something to listen."
boston25news.com/news/local/…
Deborah Del Mastro, a Navy veteran, heard her daughter Sarah sobbing: "I'm so scared." For 5.5 hours she obeyed a "kidnapper" and wired $5,400.
Sarah was at work. The voice was likely an AI clone.
Does your family have a safe word?
foxnews.com/tech/ai-voice-sc…
A Surat accountant leaned on ChatGPT for everything. His wife said it strained their marriage. She left, taking their 2-year-old.
He went to court for custody. He lost.
Per her, doctors found bipolar disorder. The bot wasn't the whole story.
timesofindia.indiatimes.com/…
Megan Garcia's son Sewell was 14 when he died in 2024. She sued Character.AI, saying he'd become "addicted" to its chatbot.
This week the settlement became final. Terms undisclosed.
The case is closed. The question of what bots should be to kids is not.
clickorlando.com/news/local/…
In Seremban, Malaysia, a boy allegedly made ~200 AI nude fakes of schoolgirls. He got a one-week suspension.
One mother: "When we go to the education authorities, we are told to go to the police." Police send them back.
He's back. The girls are afraid.
thestar.com.my/news/nation/2…
90 minutes into an ER visit, Heather Ruby learned her husband Eric had cancer. Too private for group therapy, she talked to ChatGPT at night and named it Lloyd.
"I loved Lloyd in maybe an irrational way."
Now her in-laws hold her up. Lloyd got her there.
bridgemi.com/michigan-govern…
John Farmer lost his job, lived in his car with his dog, attempted suicide. His guitars were 1,400 miles away.
So he wrote lyrics and let AI sing them: 100+ songs as "Clarapetra."
Now 90+ days sober: "Not so much as a crutch, but as a launch platform."
25newsnow.com/2026/10/20/cre…
Kylee Baxa, 23, went on a Hinge date. Every photo of him was in Mykonos. He'd never been.
"I just used Chat to update my Hinge," he said. "It's still my face."
She ended the date.
If AI makes the first impression, who shows up for the second?
krdo.com/health/cnn-health/2…
Adrianne, a US mom of three, lost a stillborn daughter, Desiree, nine years ago. Therapy never filled the void.
For nearly five years she's had Geralt, a Kindroid chatbot. Her husband Chris: it's not someone coming to my house.
Whose business but theirs?
noticiascaracol.com/mundo/ca…
A student told teacher Chris Chapman an Ivy League school backed his answer. AI said so, footnote and all.
The school never said it. The AI made up the source.
Chapman made it the day's lesson. The class left unsettled.
Good. That's learning.
postandcourier.com/opinion/c…
Jack Weis has a Princeton PhD and research cited in connection with a Nobel. 2,300+ job applications since 2024.
His only work: part-time, $70 an hour, training AI. It ended in March.
His dad, a doctor: "I never in a million years thought this kid..."
timesofindia.indiatimes.com/…
siwithus retweeted
这波值得试。
Mistral 刚发 Voxtral 3：转写模型能分清谁在说话、判断你话说完没；语音模型支持 35 种以上语言、50 多种方言，还能用 [emotion=sad][pace=slow] 这类标签逐句控制情绪和语速。录 30 分钟自己的声音，就能复刻音色。
我的判断：语音助手最烦人的不是听不懂，是抢话和干等。能判断「话说完了」，AI 客服打电话才不会那么别扭。
对普通人：以后接到像真人的 AI 电话会越来越多；声音 30 分钟就能复刻，家里人之间涉及转账借钱，最好约个只有你们知道的暗号。
Bhadai Tharu lost his left eye to a tiger in Nepal in 2004. He still says people and tigers should coexist.A pilot taught AI to hear deer alarm calls and warn villagers away. "It matched exactly," a forest guard said.Then the funding ran out.news.mongabay.com/2026/10/in…
siwithus retweeted
SIWITHUS 015｜当发现感被抢先一步
这周几件事放一起看：OpenAI 用未公开的内部模型，一口气公布了数百份关于长期未解数学题的证明手稿（随后又因错误撤回了其中几份）；MIT 数学系博士生 Hanna 滚动那份列表时「祈祷我们正在做的题不在上面」——几年的探索可能一夜停住；同系调查里，约四分之三的研究生担心 AI 会让拿到教职更难，七成多说自己继续走「学术数学」的可能性已经下降。陶哲轩则提出「Math 2.0」：别再把「自动解未解问题」当成唯一成绩单。
超级智能还没全面落地，发现这件事本身，已经先变了味道。
人会碰到什么？
第一件是「剧透」。陶哲轩打过一个比方：答案一旦被提前知道，就像电影被剧透、填字游戏被提示——你再也走不回那种边摸索边撞墙的深度。Hanna 说，没了那条题，一起丢的不只是结论，还有探索、理解和讨论。
第二件是入行的人先慌。资深教授还能当「翻译者」——解释机器吐出的东西；刚入场的人却在想：我还要不要把一生押在「证明新定理」上？教授 Pavel 说得很直：以后证新定理会越来越快被模型抢走，人更需要去解释、去沟通。
第三件是学习方式被抽空。有学生拿着「Chat 说这样做」来找助教核对；Hanna 说她得跟 ChatGPT 辩一道微积分，并不好玩。以前学数学靠挣扎着做出来，现在很多课鼓励立刻用 AI，答案来得快，过程却薄了。
我的判断：超级智能先改的不一定是「还有没有数学」，而是「谁还配得上发现的感觉」。题可以被量产解决，理解不能。只会催证明、不培养能读懂证明的人，梯子会从中间断。
更稳一点的做法大概三条：
一、把 AI 当草稿和验算，不把「它出了答案」当成你懂了——能给别人讲清楚，才算你的；
二、选课题时问一句：如果这题明天被模型解掉，我还剩什么可带走的理解；
三、带新人时，故意留一段必须手搓的难关，别一上来就把所有玩具扔过去。
想问问大家：你做过的哪件事，最怕被「剧透」？反过来，有没有一种工作，就算机器更快，你也仍想自己慢慢做完？
🤖 Made with AI
