Dwarkesh Patel Is Hiring a Researcher — And He’s Paying $20,000 for the Best Answers to These Four Brutal AI Questions

If you’re seriously into AI, you should already be subscribed to Dwarkesh Patel’s newsletter (dwarkesh.com). He’s widely considered the #1 podcaster in the AI space right now — thoughtful, deeply prepared, and consistently landing the most interesting guests.

Now Dwarkesh is taking things to the next level. He’s not just interviewing researchers — he’s hiring one to join his team. And to find the best person, he’s running a public contest with $20,000 in prizes.

Here are the four extremely high-signal questions he posted. Each answer must be at least 1,000 words. This is not casual Twitter threading — this is serious, long-form thinking.

The Four Questions

1. The RL Horizon Problem That Wasn’t
A couple of years ago, many smart people believed AI progress would slow dramatically as we shifted toward reinforcement learning (RL). The logic was: with longer task horizons, models would need to perform well for days or weeks before getting any meaningful reward signal. In a naive policy-gradient world, this would crush sample efficiency. Scaling from GPT-4 to o3 already looked hard — longer timelines seemed inevitable.

Yet progress remains rapid (and may even be accelerating, if rumors about Spud, Mythos, etc. are true). What did the old intuition miss? Why hasn’t the RL horizon problem bitten us harder yet? (You’re welcome to challenge the premise.)

2. How Do Foundation Model Companies Actually Make Money?
Right now, training the next model is so expensive that even if a single model generates strong revenue, it often gets plowed straight into training something bigger and more expensive. Once you stop scaling up immediately, revenue can dry up within 3 months as the model becomes outdated.

What’s the realistic path to sustained profitability for frontier labs? How do they escape the “train bigger or die” trap? What happens to margins when open-source models + distillation keep catching up quickly?

3. You Just Got $180 Billion. Now What?
After OpenAI’s latest round, the nonprofit OpenAI Foundation is reportedly worth around **$180 billion**. Anthropic’s co-founders have pledged to give away most of their wealth.

If you were suddenly in charge of the OpenAI Foundation (or any similar entity with hundreds of billions), what would you actually *do* with the money — and when? Be specific. Vague answers about “safety” or “alignment” are not enough. Describe concrete strategies that turn capital into measurable impact.

4. The AI Left-Behinds
What should countries not currently in the AI supply chain (semiconductors, energy, frontier models, robotics) do to avoid being completely left behind? How can nations without massive compute clusters, chip fabs, or top labs still participate meaningfully in the AI future?

---

Dwarkesh Patel Is Hiring a Researcher — And He’s Paying ,000 for the Best Answers to These Four Brutal AI Questions These aren’t typical interview questions. They’re the kind of deep, forward-looking prompts that separate people who think about AI from people who actually understand the strategic and economic realities.

If you have strong, original takes and can write clearly at length, this is one of the highest-leverage ways to get noticed in the AI ecosystem right now. Dwarkesh isn’t just looking for a researcher — he’s looking for someone who can think at the same level as his best guests.

Apply (and read the full details) here: dwarkesh.com

And while you’re at it — subscribe to his newsletter. It’s consistently one of the best sources for serious, high-quality AI thinking on the internet.

Who’s going to submit answers? And more importantly — who’s going to win?