Exploring AI's Hidden Capabilities
Share
Guillermo Rauch shares his perspective on the evolution of software development and AI, contrasting traditional programming (Software 1.0) with AI-driven approaches (Software 2.0), and how entrepreneurs can leverage existing AI technologies to create new platforms without needing to build foundation models from scratch.
Software 1.0 vs Software 2.0
-
Software 1.0 is what Guillermo grew up with:
- Programming languages, data structures, algorithms
- Making things efficient through better for loops and recursion
- Very deterministic and predictable
- Engineers are in control of everything
- "There's a PM giving you tickets, I know exactly what this can do, I know exactly what it can't do"
-
Software 2.0 (term coined by Andrew Karpathy) represents a paradigm shift:
- Still uses foundations of computer science but makes the process more stochastic and probabilistic
- Instead of writing every circuit of programming, relies on training models with data
- Output resembles what a traditional program might do
- Has "magical emergent properties" not explicitly programmed by engineers
- Engineers don't have to write every if-else branch or think about every corner case
The Excitement of AI Discovery
- Even creators of AI models don't know what's possible with their creations
- "Every time a new model comes out we're all... discovering the latent space"
- "We're all trying to figure out what is even possible"
- This is "in stark contrast to software 1.0" where capabilities are clearly defined
Entrepreneurial Opportunities in AI
-
Entrepreneurs don't need to train foundation models:
- "Just need to go in and put the pieces together into opinionated workflows"
- Look for "permutations of technologies that already exist that are making new platforms possible"
-
Identify clusters of queries people are using ChatGPT for:
- These can become entire platforms
- Example: When ChatGPT came out, people asked for NDAs → legal platforms emerged
- People asked for web UI → Vercel created V0 for web development
-
Specific platform ideas:
- V0 for creating video games combining Software 1.0 and 2.0 techniques
- AI Camera app that uses the photo as input to a prompt:
- Automatically fixes common photo issues (blinking, lighting)
- Removes unwanted objects
- Provides multiple enhanced versions
- Combines existing models into useful workflows
-
Focus on creating opinionated workflows:
- "Your job will mostly be to combine models, create pipelines of models, prompting"
- "Embrace the wrapper, embrace the fact that models are fucking phenomenal"