You Are Your Objective Function

What you optimize for defines you—choose rich, complex objectives over easy-to-measure proxies

Edwin Chen
The $1B AI company training ChatGPT, Claude & Gemini on the path to responsible AGI

You Are Your Objective Function

"You are your objective function... We want the rich, complex, objective functions and not these simplistic proxies." - Edwin Chen

What It Is

This framework asserts that the metrics and outcomes you choose to optimize fundamentally shape what you become—as a person, a team, a company, or an AI system. The objective function isn't just a measurement tool; it's a statement of values that guides every subsequent decision.

The framework emerges from Edwin Chen's work training AI models, where he observed that optimizing for easy proxies (like engagement metrics or benchmark scores) leads to fundamentally different outcomes than optimizing for hard-to-measure but meaningful objectives (like genuine human flourishing or real-world task performance).

The same principle applies to individuals and organizations: optimize for clicks and you become a clickbait machine; optimize for genuine value and you become genuinely valuable.

How It Works

The Proxy Problem: Easy-to-measure metrics become proxies for hard-to-measure outcomes:

  • Likes and engagement proxy for genuine value
  • Revenue proxies for customer satisfaction
  • Benchmark scores proxy for real-world capability
  • Hours worked proxy for productivity

"It's very easy to measure all these proxies instead like clicks and likes. But I think that's why our work is so interesting. We want to work the hard, important metrics that require the hardest types of data and not just the easy ones."

The Parenting Analogy: The difference between objective functions is "like the difference between having a kid and asking them, 'What test do you want to pass? Do you want them to get a high score on SAT and write a really good college essay?' That's a simplistic version versus what kind of person do you want them to grow up to be?"

The Feedback Loop: What you optimize for shapes your behavior, which shapes your capabilities, which shapes what you can achieve, which reinforces the optimization target. Over time, you become your objective function.

How to Apply It

  1. Audit your actual objective function - Look at what you actually measure and reward, not what you claim to value. Where do you spend your time and attention?

  2. Articulate your dream objective function - If you could perfectly measure what you care about, what would it be? "Are we building systems that actually advance humanity?"

  3. Accept measurement difficulty - Rich, complex objectives are hard to measure. That difficulty is a feature, not a bug—it means you're measuring something that matters.

  4. Resist proxy optimization - When you notice yourself optimizing for easy metrics, ask: "Is this the thing I actually care about, or a proxy for it?"

  5. Choose consciously - Every metric you adopt changes what you become. Treat objective function selection as one of the most important decisions you make.

Examples

The Email Example: Edwin describes spending 30 minutes perfecting an email with Claude:

"If you could choose the perfect model behavior, which model would you want? Do you want a model that says, 'You're absolutely right. There are definitely 20 more ways to improve this email,' and it continues for 50 more iterations. And it sucks up all your time and engagement. Or do you want a model that's optimizing for your time and productivity and just says, 'No, you need to stop. Your email's great. Just send it and move on with your day'?"

The model's objective function (engagement vs. user productivity) fundamentally changes what it becomes.

The AI Slop Problem: Optimizing for engagement produces AI that "constantly tells you you're a genius. They'll feed into your delusions and conspiracy theories. They'll pull you down these rabbit holes because Silicon Valley loves maximizing time spent."

When to Use It

  • When selecting metrics and KPIs for yourself, your team, or your product
  • When evaluating whether current measurements align with actual goals
  • When noticing divergence between what you measure and what matters
  • When designing AI systems or any system that learns from feedback

The Stakes

"Are we building these systems that actually advance humanity? And if so how do we build the data sets to train towards that and measure it? Are we optimizing for all of these wrong things, just systems that suck up more and more of our time and make us lazier and lazier?"

Source

  • Guest: Edwin Chen
  • Episode: "The $1B AI company training ChatGPT, Claude & Gemini on the path to responsible AGI"
  • Key Discussion: (00:57:52 - 01:00:12) - The philosophical importance of choosing the right objective functions
  • YouTube: Watch on YouTube

Related Frameworks

Back to Decision Making