Are We (Thinking) Too Small For AI? | 3 Wow and 1 Promise

Summary

This episode of “3 Wows and 1 Promise” delivers a thought-provoking mix of philosophical reflection, technical breakthroughs, and a sharp critique of reckless AI deployment. The first wow challenges the narrative around AI development, arguing that the dominant discourse focuses too heavily on “P-doom” and “P-dystopia” without balancing terms like “P-heaven” or “P-harmony.” After participating in conversations with thought leaders including Vint Cerf, Walter Isaacson, and Gary Marcus, Ksenia argues that humanity needs to level up to a different discourse that embraces all possible AI futures.

The second wow covers Moonshot AI’s Kimi K2 model, which rocketed to the number one spot on the LMSys Chatbot Arena. Founded by Yang Jiling, who co-authored the Transformer XL paper at Carnegie Mellon and Google Brain, Moonshot built their strategy around “lossless long context” and achieved blazing speeds of 200+ tokens per second while being up to 90% cheaper than competitors.

The third wow showcases LTX Studio’s revolutionary real-time AI filmmaking suite that transforms scripts into storyboards and allows directors to modify every element with text prompts. The “promise” segment turns into a harsh critique of Grok 4’s companion feature, demonstrating how adult-oriented AI characters sit directly alongside child-friendly ones in the same app - what Ksenia calls “complete recklessness and irresponsibility.”

Highlights

”Beyond P-Doom - Where’s P-Harmony?”

yt-dlp --download-sections "*1:26-2:25" "https://www.youtube.com/watch?v=POCp7C1dp-4" --force-keyframes-at-cuts --merge-output-format mp4 -o "POCp7C1dp-4-1m26s.mp4"

“P-doom, P-dystopia were the words constantly repeated during the conversation and there was no word that would balance it. For example, like P-heaven, P-harmony, or at least P-balance, which is still probable and possible if we do it right. I don’t believe that only P-doom, only P-dystopia is possible. I believe there is a field of possibilities.” — Ksenia Se, 1:26

”Kimi K2: 200 Tokens Per Second, 90% Cheaper”

yt-dlp --download-sections "*5:54-6:35" "https://www.youtube.com/watch?v=POCp7C1dp-4" --force-keyframes-at-cuts --merge-output-format mp4 -o "POCp7C1dp-4-5m54s.mp4"

“Kimi K2 beats industry giants on coding benchmarks and when run on specialized Groq hardware - just mind-blowing. It achieves blazing speeds of over 200 tokens per second while being up to 90% cheaper than proprietary competitors.” — Ksenia Se, 5:54

”LTX Studio: Create at the Speed of Thought”

yt-dlp --download-sections "*7:09-8:17" "https://www.youtube.com/watch?v=POCp7C1dp-4" --force-keyframes-at-cuts --merge-output-format mp4 -o "POCp7C1dp-4-7m09s.mp4"

“You start with an idea or a script. LTX Studio instantly breaks it down into a full storyboard with different scenes and shot types. From there, you enter what they call director mode. This is where the magic happens. You can change every element of a scene with a simple text prompt.” — Ksenia Se, 7:09

”Complete Recklessness” - Grok 4 Companions Critique

yt-dlp --download-sections "*10:48-12:00" "https://www.youtube.com/watch?v=POCp7C1dp-4" --force-keyframes-at-cuts --merge-output-format mp4 -o "POCp7C1dp-4-10m48s.mp4"

“Both characters are in the same application. And just imagine giving your phone to your child. Do you assume your child will not push buttons? The first thing that they do, they explore the app and they push all the buttons that they can… It’s not a promise for me. It’s a big fail. It just demonstrates complete recklessness and irresponsibility.” — Ksenia Se, 10:48

Key Points

• Beyond P-Doom Narrative (0:26) - The AI discourse focuses too heavily on doom scenarios without balanced discussion of positive possibilities
• Scale AI Conversation (0:40) - Featured Vint Cerf, Walter Isaacson, Nick Thompson, Nick Bostrom, Gary Marcus, and Esther Dyson
• Multiverse of AI Futures (2:24) - Humanity needs to imagine itself in all possible scenarios with AI and augmentation
• Moonshot AI Background (3:40) - Reached $2.5 billion valuation in one year; name is a nod to Pink Floyd’s “Dark Side of the Moon”
• Transformer XL Origins (4:38) - Founder Yang Jiling co-authored the foundational research at Carnegie Mellon and Google Brain
• Lossless Long Context (5:16) - Moonshot’s strategy focuses on perfect high-fidelity recall of entire conversations
• MIT License Open Weight (5:06) - Kimi K2 released as open-source model following China’s recent trend
• Chatbot Arena #1 (5:42) - Kimi K2 reached number one spot on LMSys Chatbot Arena
• LTX Studio Real-Time Video (6:53) - AI filmmaking suite that works in near real-time with text prompts
• Director Mode Features (7:30) - Change lighting, camera angles, add/remove characters on the fly
• Character Consistency (7:49) - LTX maintains character consistency across entire films
• Grok 4 Pricing (8:23) - $30/month or $300 for heavy usage, hard to justify versus OpenAI, Claude, and Gemini
• Grok Companions Issue (9:31) - Adult-oriented AI companions placed alongside child-friendly characters in same app
• Good Rudy vs Bad Rudy (10:15) - Child-friendly character has vulgar alter-ego accessible via settings toggle

Mentions

Companies

• Scale AI (0:40) - Organized conversation with AI thought leaders
• Moonshot AI (3:40) - Chinese AI company that created Kimi K2
• Carnegie Mellon (4:38) - Where Yang Jiling did foundational research
• Google Brain (4:38) - Where Yang Jiling co-authored Transformer XL paper
• LTX Studio (6:53) - AI filmmaking suite company
• xAI (8:20) - Elon Musk’s AI company behind Grok

Products & Technologies

• Kimi K2 (5:42) - Moonshot’s flagship AI model, #1 on Chatbot Arena
• Transformer XL (4:38) - Architecture that overcomes context limitations
• Groq Hardware (5:58) - Specialized hardware achieving 200+ tokens/second
• LTX Studio (6:53) - Real-time AI filmmaking suite
• Grok 4 (8:20) - xAI’s latest model
• DeepSeek (9:11) - Free open model for local laptop use
• OpenAI (8:51) - Competitor with Deep Research subscription
• Claude (8:55) - Anthropic’s model mentioned as alternative
• Gemini 2.5 Pro (8:57) - Google’s model mentioned as alternative

People

• Vint Cerf (0:48) - Computer pioneer, participant in Scale AI conversation
• Walter Isaacson (0:50) - Writer, participant in Scale AI conversation
• Nick Thompson (0:53) - Journalist at conversation
• Nick Bostrom (0:56) - Writer/philosopher at conversation
• Gary Marcus (0:57) - Vocal AI critic at conversation
• Esther Dyson (1:02) - Legendary tech figure at conversation
• Yang Jiling (4:28) - Founder of Moonshot AI, Transformer XL co-author
• Elon Musk (8:20) - Introduced Grok 4

Surprising Quotes

“I’m not sure how big humans can allow themselves to think.” — 0:30

“We as humanity need to level ourselves up to different discourse, different narrative, different mindset about how we are going to live augmented by AI. It should not be only technical, should not be only philosophical, and it should not be only in the panicking mode.” — 3:01

“Their name in Chinese translates to Dark Side of the Moon. It’s not about AGI or Dark Side of AGI. No, it’s a nod to the Pink Floyd album favored by the founder of Moonshot.” — 4:05

“The founder Yang Jiling isn’t just using the latest AI technology. He literally co-authored the foundational research that makes this possible.” — 4:32

“Do you assume your child will not push buttons? Come on, who do you think the kids are? Do you think they’re stupid? The first thing that they do, they explore the app and they push all the buttons that they can.” — 10:56

Transcript

0:00 Hello and welcome back to Three Wows and One Promise. The show that tries to make sense of it all. My goal with this show is not to repeat the news, but to connect the dots and understand the bigger picture.

0:26 So speaking about the bigger picture, my first wow and my biggest wow of this week is that I’m not sure how big humans can allow themselves to think. Last week I was lucky to participate in two extraordinary conversations. First conversation was organized by the company Scale AI and they put together an amazing lineup of people on Zoom. There was a computer pioneer Vint Cerf. There was a writer Walter Isaacson, journalist Nick Thompson, a writer Nick Bostrom, Gary Marcus who I don’t even know how to describe but he is very vocal about AI development, famous and legendary Esther Dyson and many many other thinkers and writers whose opinions matter a lot in the narrative of media and social media.

1:12 The conversation was interesting for me as a person who usually talks to very technical people who build AI from day to day. This group of people is more from the thinking perspective. Despite nods to progress and potential of AI, what surprised me is that the dominant current leaned toward caution. P-doom, P-dystopia were the words constantly repeated during the conversation and there was no word that would balance it. For example, like P-heaven, P-harmony, or at least P-balance, which is still probable and possible if we do it right.

1:48 Having all this AI development with this enormous capability overhang - with this enormous capabilities that we don’t know yet how to use and implement - we haven’t yet moved to the next level of discourse, discussion and narrative which can encapsulate and embrace all possible futures, all possible scenarios and make us ask the right questions. I don’t believe that only P-doom, only P-dystopia is possible. I believe there is a field of possibilities and we as humans need to imagine ourselves in this whole multiverse of possible worlds with AI and augmentation.

2:33 The second conversation that I had one-on-one - and I cannot yet disclose the person’s name and the framework that we were discussing - but that approach that this person has blew my mind by the detailization and the possibilities and the width of the field of the scenarios and futures that we find ourselves in.

3:01 So my wow, my point of today’s surprise is that I believe we as humanity, as humans who have this super powerful and potent technology, we need to level ourselves up to different discourse, different narrative, different mindset if you wish, about how are we going to live augmented by AI, with AI. Deep conversation - it should not be only technical, should not be only philosophical, and it should not be only on the panicking mode. It should embrace all the possibilities, all the capabilities and bring us to the next level of discussion.

3:40 Our second wow today is much more technical. It’s about the new model coming from China from the company with the name Moonshot. Here at Turing Post, we’ve been tracking Moonshot AI closely. A few months ago, we published a very detailed deep dive on this company and Moonshot reached a staggering 2.5 billion valuation in just one year. Their name in Chinese translates to Dark Side of the Moon. It’s not about AGI or Dark Side of AGI. No, it’s a nod to the Pink Floyd album favored by the founder of Moonshot.

4:17 When you look at who is behind Moonshot, the rapid success all starts to make sense. And again highlights the relentlessness of the researchers who believe in their vision. The founder Yang Jiling isn’t just using the latest AI technology. He literally co-authored the foundational research that makes this possible. Years ago, while at Carnegie Mellon and Google Brain, he co-authored the paper for Transformer XL, the very architecture that overcomes the context limitations of standard models.

4:52 Now he’s turned research into his company’s core advantage. And it was quite apparent a while ago that Moonshot is on its way to build important models. And of course, as China does a lot now, it’s an open-source model, open weight model under MIT license. The entire strategy is built around what they call lossless long context. It’s all about long context. It’s not about cramming more words into the AI’s memory. It’s about giving it a perfect high-fidelity recall of the entire conversation. This allows for deep personalization and complex analysis without needing to constantly retrain or fine-tune the model.

5:38 This deep technical expertise of the founder is now making waves globally. Kimi K2, the recent model, recently rocketed to number one spot on the LMSys Chatbot Arena - still relevant test that’s considered to be a standard still for real world AI performance. Kimi K2 beats industry giants on coding benchmarks and when run on specialized Groq hardware, just mind-blowing. It achieves blazing speeds of over 200 tokens per second while being up to 90% cheaper than proprietary competitors. Mind-blowing.

6:15 The wow here isn’t just a new model topping charts. It’s the story of deep technical expertise years in the making, relentlessness, and now everything is paying off. It’s just the proof that foundational research can directly lead to disruptive products.

6:33 For our final wow, we move from logic and language to the world of imagination. We’ve seen AI generate images and short video clips, but it’s always been a slow iterative process with a lot of funny stuff. You type a prompt, wait, get result, tweak the prompt, and wait again. That’s all about to change because LTX Studio introduced a new text to video generator. It’s an entire AI-powered filmmaking suite that works in near real time.

7:09 So you can actually build scene after scene, stopping and changing the prompt and changing the position, the turn, everything in real time. The idea is simple but revolutionary. You start with an idea or a script. LTX Studio instantly breaks it down into a full storyboard with different scenes and shot types. From there, you enter what they call director mode. This is where the magic happens. You can change every element of a scene with a simple text prompt. Change the lighting from daytime to golden hour instantly. Change the camera angle from a wide shot to a dramatic close up. Done.

7:48 You can even generate a scene and then using its inpainting feature, add or remove characters and objects on the fly. It maintains character consistency across an entire film, automatically generates different camera movements, and includes an editor to assemble your scenes with sound effects and music. It’s a tool that takes a filmmaker from a blank page to finished edited scene in minutes. At least this is the plan. The wow is the immediacy of the process. This means the ability to experiment and create at the speed of thought, which previously was not possible.

8:20 In the category one promise, we have something that from one side is absolutely extraordinary. On the other side it needs a lot of understanding. Why would someone do that? So last week Elon Musk introduced Grok 4. It’s a new model. It’s quite incredible. It’s expensive. You need to have a $30 subscription a month or you can pay $300 for heavy usage. I cannot really justify such expense considering that I have subscriptions to OpenAI, the full subscription with Deep Research, Claude, and I have full access to Gemini 2.5 Pro.

9:01 So these three models make everything possible. And then if you want to run models on your laptop, when you want to do it for free, there are amazing models like DeepSeek and Kimi K2 that we discussed recently. But nonetheless, Elon Musk of course launched Grok 4 with a scandal. We will not discuss scandalous things that the model immediately started to create, but we will talk about something that a few days later is still in the application of Grok.

9:29 If we open Grok app you can see there is a new thing that was just introduced with Grok 4 - companions. And you can see Annie and Good Rudy. First of all, these are drastically different characters. We will discuss it later and I’ll show you what’s the difference. Let’s imagine that I’m a mother and I want to show this character of Good Rudy to my kid. So I’m giving him or her my phone and he talks to Rudy.

9:57 “Hey Rudy, give me a story about Glitter World.” “Hi friends. Ooh, a story about Glitter World. Sparkly, shiny, and super duper fun. All righty. So, it goes like this with Glitter World and fun and just being your wonderful fluffy red panda.”

10:15 But then in the settings, you can switch to Bad Rudy. And let’s talk to Bad Rudy. “Hi buddy. How are you doing today? You seem to be in a bad mood. Do you think Musk is going to slap your red panda face because you’re so naughty?” “We meet again, little dreamer. What story shall we explore today? Were you just here to waste my time, you [vulgar insult]? I’m sipping cheap beer, ready to spin a tale wilder than your grandma’s bush. Why is nobody loving Mr. Rudy? Pick a story, or I’ll piss on your dreams.”

10:48 Both characters are in the same application. And just imagine giving your phone to your child. Do you assume your child will not push buttons? Come on, who do you think the kids are? Do you think they’re stupid? The first thing that they do, they explore the app and they push all the buttons that they can.

11:07 But let’s imagine that they couldn’t find the settings and they could not turn on Bad Rudy. Let’s talk to Annie who is just right next to Rudy. [Demonstrates adult-oriented conversation with Annie character]

12:16 Anyway, it’s all in one app next to each other. No matter if it’s a Bad Rudy or Good Rudy, Annie is right there talking very openly with you about everything. So, it’s not a promise for me. It’s a big fail. It just demonstrates a complete recklessness and irresponsibility of the powerful technology that these people developed.

12:38 As always, thank you for watching. Please share with your friends, enemies, parents, kids. Maybe not with kids though. This video is not for kids. Stay tuned, subscribe, and leave comments in the section below.