# LLMs may raise the bar for creative work without replacing ideas

Markdown copy of https://realalpha.app/s/llms-may-raise-the-bar-for-creative-work-without-replacing/ on Real Alpha, the best minutes from long interviews with the people building AI. Cite and link the page, not this copy.

- Speaker: [Dan Houser](https://realalpha.app/people/dan-houser/) · Dan Houser is co-founder of Rockstar Games and founder of Absurd Ventures. · Co-created Grand Theft Auto and Red Dead Redemption, then founded Absurd Ventures to develop games and stories.
- Host: [Lex Fridman](https://realalpha.app/people/lex-fridman/) · Host, Lex Fridman Podcast · Robotics and AI researcher at MIT; started the show in 2018.
- Show: Lex Fridman Podcast
- Episode: [Dan Houser: GTA, Red Dead Redemption, Rockstar, Absurd & Future of Gaming | Lex Fridman Podcast #484](https://realalpha.app/v/o3gbXDjNWyI/), Oct 31, 2025
- Moment: 37:22 to 41:12 of the episode (4 min), playing on https://realalpha.app/s/llms-may-raise-the-bar-for-creative-work-without-replacing/; the episode from there: https://realalpha.app/v/o3gbXDjNWyI/?t=2242

## Quote

> “the **last five percent** is going to end up being about 95 of the work”
> — [Dan Houser](https://realalpha.app/people/dan-houser/) at [38:52](https://realalpha.app/v/o3gbXDjNWyI/?t=2332) · Lex Fridman Podcast

## Takeaways

1. Original ideas are harder to automate than low-level work.
2. The final polish may take most of the effort.
3. AI could make creative fields harder to enter.

## About this segment

Dan Houser expects language models to produce decent, similar-sounding work, but doubts they can originate strong concepts or capture creative magic. He thinks the tools may make it harder for newcomers to get started.

From “Dan Houser: GTA, Red Dead Redemption, Rockstar, Absurd & Future of Gaming | Lex Fridman Podcast #484”, Oct 31, 2025. [Watch the full episode](https://realalpha.app/v/o3gbXDjNWyI/)

## Transcript

Transcribed by machine (Whisper) from the episode’s audio, so a word may be off here and there.

[37:28](https://realalpha.app/v/o3gbXDjNWyI/?t=2248) how do you feel about llms being able to uh write in a way that looks awfully human i'm not that afraid of them for large stale concepts i don't think they're gonna be very good at that i think if you were i think it's harder if you know i i begun and i i was too shy to tell anyone i want to be a writer that's why i ended up in video games and i would scribble away like writing manuals and writing on like ps1 games all 12 lines of dialogue in a game sometimes i wouldn't even get that job and i just write the website copy and um and then by doing then working on little bits and pieces and then it it you know i'd luckily done enough work that when gta 3 turned up was the first thing that resembled real writing. I had all of these small bits of skills that I could assemble into it. Based on my fairly limited understanding of how language models work,

[38:28](https://realalpha.app/v/o3gbXDjNWyI/?t=2308) they're not going to replace good ideas. They can't really come up with good new ideas. What they can do is do low-level stuff. So I think it's going to be harder for people to start out in some of these spaces. If you're not a very good concept artist, you're in a lot of trouble. if you have original ideas i think you're fine but i think uh i also think that the fur they've done the sort of first 90 of the work to sound human 95 possibly in some areas the last five percent is going to end up being about 95 of the work i think that last bit in with with with tech in my experience with things like facial animation always been the last bits and pieces take far longer than the first bit. And so I'm probably a hideous Luddite, but I'm less scared than a lot of people. I think you're going to end up with a lot of work

[39:19](https://realalpha.app/v/o3gbXDjNWyI/?t=2359) that looks the same. It's going to help people be creative in some ways. It's going to get some people who probably shouldn't be in that space out of that space. But if you've got talent, I think it'll be fine. Yeah, it's... I agree with you totally, actually, and it's hard to really put a finger on it. So one way to illustrate that, I speak English and Russian, and I've been reading Dostoevsky in both languages and using LLBs to translate back and forth, because I was preparing to have a conversation with the translators of Dostoevsky.

[39:48](https://realalpha.app/v/o3gbXDjNWyI/?t=2388) Which ones? Richard Prevere and Larissa Volokonsky. Yeah, I read quite there when they first did Crime and Punishment. Mm-hmm. That was amazing. They're wonderful translators and a wonderful love story, too. But in the translation process, you get to see the LLM is missing some magic. And they're, you know, that couple of translators are world-class experts capturing the magic. And I can't quite put that into words because you said like totally novel ideas, yes.

[40:17](https://realalpha.app/v/o3gbXDjNWyI/?t=2417) But also this magic of the timing, the right word at the right time that captures the human experience. So they can do some really incredibly human-like, the 90%, like I mentioned, human-like phrasing about the bulk of the storytelling, but the magic, whether it's the endings of Red Dead Redemption 1 and 2, the timing of that, the word choice of that, everything around that. but it's hard to argue because they're incredibly impressive winning all kinds of math competitions but it's what is that magic uh and again that could be just a romantic human side of me this thing that llos won't be able to capture that maybe desperately holding on for hope i don't think they're going to come up with magic i think they're going to be fantastic at coming up with