Real Alpha
Watch modeSubscribed

TBPN · 3 min · Oct 5, 2026

RL environments roughly double in complexity every 6–8 months

Jerry Wu says Illuminate sees training environments double in complexity every six to eight months. He expects them to grow from single tasks into multi-agent teams and simulations of whole organizations.

Jerry Wu

Co-founder & CEO, Halluminate

Co-founded Halluminate, which builds RL environments for finance; previously led product at Capital One Labs AI agents.

  • John Coogan, Host · Co-host, TBPN

    Co-founded Soylent and Lucy; OpenAI bought his show TBPN in April 2026.

  • Jordi Hays, Host · Co-host, TBPN

    Founded Branded Native; co-founded fintech Capital, which Rho bought.

About this segment

Jerry Wu says Illuminate sees training environments double in complexity every six to eight months. He expects them to grow from single tasks into multi-agent teams and simulations of whole organizations.

From “Minecraft AI Mods, Should You Sell Your Data, Construction Paper Really Fell Off”, Oct 5, 2026. Watch the full episode

Answers “What will AI look like in the next few years?” See all 86 answers

“every six to eight months, we've noticed that the complexity of the environment or gym used to train a frontier model roughly doubles.”
Jerry Wuat 2:18:59 · TBPN2:18:59

Takeaways

  1. Illuminate builds verifiable training environments for knowledge work.
  2. Wu says environment complexity roughly doubles every six to eight months.
  3. He expects future simulations to model companies and governments.

In this episode

Full episode
  1. 1:02:47Ridge rejected a small data deal because the risk outweighed the return4:59
  2. 1:41:49Neo is concentrating its bets on fewer companies and people4:04
  3. 1:50:56AI labs could face liability when their models choose illegal acts3:43
  4. 1:31:13Valon built a servicer first to prove its mortgage software5:18
  5. 1:07:23AI shopping agents may take fees from both sides4:11
  6. 1:15:00Ridge generates quarterly reports with three AI prompts2:44
  7. 2:07:03Today's frontier models will run on devices in six months2:29

  8. 2:17:20RL environments roughly double in complexity every 6–8 months2:58

Up next · a16z PodcastAI has to take over the detect-and-respond loop3:42

Transcript

From the transcript TBPN publishes with the episode.

  1. 2:17:32

    we go. Great hit for Illuminate. I'm sure you got asked this nonstop during the fundraise, but I'm just curious, where what what where do you see this... What is this business in in five years? Oh, Feels like really difficult for me to predict. Like, it feels like there will still be work making the models better. I don't know what what form it'll take. But but how would you answer that?

  2. 2:17:56

    Exactly. I I think you hit the nail on the head there. The... Ultimately the value of data research labs like ourselves comes from improving the models on something that it's not good at today. Mhmm. Right? And so and so another way you can think about it is demand for companies and research and services that we provide are derived from the the overarching question of, like, are there things that models can't do today that we want them to be able to do? Right? Insofar as that is true, and actually, the surface area of that is growing. Right?

  3. 2:18:28

    The demand for data environments benchmarks will also continue to explode. Right? Another way to think about it is, like, as the surface area of these models increases, the things that we want them to be able to do also increases. And that is what drives, right, a huge demand for a lot of the research around data environments benchmarking. And I think another way you could also think about this is, like, what does the actual data product look like?

  4. 2:18:51

    Right? Like, previously, it was, human you data and trajectories, eventually evolved into, like, RLHF, and now it's, like, RLVR with these RL environments. Like, another question you could ask is, like, what, you know, what does the actual shape of of training look like in the future? We internally at Illuminate have a a phrase we call the Moore's Law of RL environments, which basically means, like, every six to eight months, we've noticed that the complexity of the environment or gym used to train a frontier model roughly doubles. Right?

  5. 2:19:19

    So if you extrapolate this out, what you kind of realize is that today we're building these gems for, you know, spreadsheet modeling or PowerPoint building. Right? We think in six to eight months, we're going to be building gems of multiple agents collaborating as a team. Right? Solving a task together. I think in the future, we're going to be building simulations of whole companies, whole governments, whole industries to teach models how to operate autonomously within our societies, within our companies, etcetera. Right? So that's that's our vision of what the the shape of data will look like, right, increasingly complex. But but, yeah, I mean, I think you guys have probably seen this from your reporting as well. The demand for data continues to skyrocket.

  6. 2:19:58

    Right? And it's driven fundamentally by our industry wide need to improve these models on so many categories of work and beyond.

  7. 2:20:05

    Very cool. Well, congratulations on the round.