Oscar Le's Blog

A practical guide to selecting LLM models and providers based on your product's specific needs and constraints.

Why Zillow lost $550M on predictions, finding founding employees, and how LLMs democratize enterprise AI.

Testing Lovable, Replit, and Bolt.new shows their struggle with custom logic. A comparison of no-code platforms' real capabilities.

While no-code tools dominate, developers who do everything need scriptable, API-first tools. This underserved market is growing fast.

Success in LLM applications comes from niche focus, speed to market, and cost optimization, not big funding or features.

Building AI that excels at research with AlphaGo-style self-improvement might reach AGI faster than solving general intelligence directly.

The best path to big results is working at the frontier of your domain, where technical breakthroughs create lasting business moats.

Like ML's loss function, your product needs one metric to optimize while others serve as guardrails. Choose wisely.

As frontier model improvements slow down and trust increases, enterprises have a unique window to adopt LLMs without the risk of rapid obsolescence.

While others scramble for chips and users, Google quietly holds every card needed to win the AI arms race.

LLMs excel at making sense of messy, unstructured input. That shifts the burden of precision from people to systems. This capability unlocks massive opportunities in business.

Get hands-on with the future of web browsing through BrowserOS, a free open-source alternative to invitation-only Perplexity Comet.