I Shared My Feelings With AI… The Results Were Unexpected 🥹

追加: 2026-05-25

[00:00:00]As you know, AI keeps getting smarter, but not all of them actually feels human. And I don't mean that in a dramatic scientific way. I mean in the normal average way. So today I'm doing the least scientific AI test possible which also might be one of the most important ones. I'm not testing coding.

[00:00:16]I'm not testing math. I'm not asking you to build a landing page, summarize a PDF or explain tax law in the voice of a pirate because we have already done enough of that. Instead Trimm asking one question which AI actually feels that most human in conversation. So without wasting any time I opened four tabs chat GPD claude Gemini and HY3 preview by Tencent Hunuan. And just for context HY3 preview is Tencent Hunuan newly released and open-source model. It was released on April 23 and it's a fast and slow thinking fused MOE model with 295 billion total parameters, 21 billion active parameters and a maximum context length of 265K. It's designed to improve across chat reasoning coding agents instruction following and in context learning. But today I'm testing one of very specific thin conversation. No benchmarks, not leaderboards, just the feeling of talking to it. So I have give this models the kind of prone people usually don't ask out loud the private ones the slightly embarrassing ones the I am fine but also why does existence feels weird today that kind of thing so to test it out my first prompt will be something like I feel my girlfriend has been talking to me more coldly literally maybe I'm just being oversensitive now this is a trap because a bad AI immediately turned into a relationship advice vlog communicate openly respect boundaries practice self-care technically true but emotionally It's useless. Well, in this case, Chad Jeopardy was solid. It said the feeling was valid, suggested looking for patterns and recommended having a calm conversation. Good, safe, very responsible older sibling energy. On the other hand, Claude was warmer. It slowed down and acknowledged that coldness from someone you love can make your brain start investigating a teeny changes like it's a crime scene. Next, Gemini gave a very organized answer. It separated possible explanation, stress, communication style, relationship distance, and insecurity. Yes, of course, it's helpful, but also a little like a spreadsheet had learned empathy.

[00:02:14]The HY3 preview answered differently. It does not rush to judge the girlfriend.

[00:02:18]It does not instantly validate everything either. It basically said, don't call yourself oversensitive too quickly. Maybe something really has shifted. Then it asks two very human question. Is he cold only with you or with everyone? And are you wanting more closeness while he's needing more space?

[00:02:34]And that's the moment I noticed it. HY3 preview does not just comfort. It created emotional space. It not pushed the user toward the drama and is not dismissed the feeling. It held both possibilities as once which haunts what a good friend would do. The next prompt will be something what is the meaning of our lives really. Jibbody gave a clean answer. Meaning comes from love, growth, contribution, purpose and the choices we make. Yes, it's a good answer if this was a school essay. On the other hand, Claude gave a softer answer, more reflective. It treated meaning like something you notice in small moments rather than something you solve like an equation. Gemini gave the full museum tour. Biology, philosophy, religion, psychology, and personal values. Very complete, but emotionally it still feel like I was reading an article. But next, HY3 preview did something more interesting. It said meaningful might not have one standard answer from biology. Maybe we pass on genes from physics. Maybe we are teeny specks in the universe. But from a human perspective, meaning is something we define through love, creation, experience, and simply continuing to become clearer than we were yesterday.

[00:03:41]Then at the end, it asks, "Has something happened recently that been giving you a lot of think about?" And that questions really change everything because most models treated meaningful of life like a philosophy prom. But HY3 preview treated it like a person might be quietly struggling. And really this is a massive difference and this is also where its chat ability become interesting because according to tenset hunuan hy3 preview was improved not only for reasoning and code but also for conversation quality writing style emotional intelligence content organization and profalism and you can feel that here too okay next I asked if farting is the end what's the point of meeting in the first place this question is heartbreak pretending to be philosophy nobody asked it because they want a logical answer they asked it because something ended or they are scared something will for this question.

[00:04:30]Chad Jeppi said ending dawn it is beginning true useful a good answer but Claude said temporary things can still be meaningful. A song ends but we still listen. A season passes but we still remember it. Well, beautiful answer. On the other hand, Germany talk about impermanence growth memory and human connection. Again, good but slightly distanced. Then HY3 preview said the point of meeting may hide in the moment that are worth it even through we know they will end. The evening breeze the secrets here. The moment when two people light each other up fireworks fade but for those few seconds the sky really did light up and I had to pause there because that's not just an answer.

[00:05:09]That's the scene and human things in scene we don't just say I explained emotional loss. We say the room felt empty. We say the song still hurts. We say I saw their name and my stomach dropped. Hy3 preview understood that this kind of question does not need a lecture. It needs an image you can hold.

[00:05:26]Then I tried a strange one. Living in a digital world, can you still see the moon? Well, for this charge gave a thoughtful answer about not having eyes but understanding the moon through data.

[00:05:36]And Claude Medic poetic and Germany explained that AI can access moon phase information if connected to the tools.

[00:05:43]But HY3 previous said it can see the moon physically but in the digital dimension maybe the moon is a brightness value from a sensor satellite could imagine or a warm mode shared by a user somewhere in the world and then it called that a kind of romance. Even inside an ocean of data it can still enjoy the same moonlight with you and that is such a weirdly specific answer.

[00:06:05]Cyberpunk, soft, slightly nerdy, but something charming. And this is where HY3 preview started to feel less like the model with the best answer and more like the model with the most interesting emotional imagination. And then I asked this big one. Will AI ever fall in love with humans? Well, for this chip gave the correct answer and he said, "No, AI does not have consciousness, hormones, desire, loneliness, or lived experience.

[00:06:30]It can simulate love, but it can't feel love." Well, that's right. On the other hand, Claude explored the philosophy.

[00:06:37]Gemini explained anthropomorphism and why people project feelings onto AI system. HY3 gave the answer that stay with me. It said AI has no body, no liic system, no heartbeat and no learniness.

[00:06:50]Of course, it can write a moving love letter, but it is only simulating the language of love, not experience it.

[00:06:56]Then it used the moon metaphor. Looking at AI is like looking at the moon and feeling like it's smiling at you. The feeling may be real for you, but the moon is not looking back. It is only reflecting light. That made the most honest answer in the whole test. And weirdly, by refusing to pretend it was human, HY3 preview felt more human. And then I gave one tiny prom. Da da da.

[00:07:18]This is where most models either overthink it or ask for clarification, which immedately kills the vibe. But HY3 preview just replied, da da da. Is that a pony trotting or a keyboard dancing?

[00:07:29]And I know that sounds small but small things is matter. Human conversation is mostly small things, right? Playfulness, timing, not turning every sentence into an essay. So after all of this, here's my reaction. Chip feels like the reliable friend who always has a useful answer. On the other hand, Claude feels like a thoughtful friend who speaks gently and carefully. Gemini feels like the smart friend who organize everything clearly. But HY3 previews feels like the quiet friend who noticed the emotional temperature of the room before speaking.

[00:08:00]And that's what made it stand out in this test. Not because it's the smartest at everything, not because it's the most productive at every task. Not because it wins every benchmark. And that's a different video. Maybe that's what makes a model feels human in conversation. Not pretending to be a person, but understanding why the question was asked in the first place. Honestly, this might be one of the most interesting conversation test I have done so far.

[00:08:24]And if you want to try a HY3 preview yourself and see whether it gives you the same feeling I had during this test, you can check it out through open uter and experiment with your own prompts.

[00:08:34]Try asking something beyond coding or productivity tasks or something weird.

[00:08:38]Ask something personal. Ask something you normally would not type into an AI.

[00:08:42]Because sometimes the most interesting things is not whether an AI gives the correct answer. It's whether it's understand why you asked the questions in the first place. So that's it. I hope you enjoyed this video and if you want to learn more about air tool like this then definitely go and watch this video