Discover more from nathan
what i've been reading (38)
OpenAI Keynote, Bill Gates, Executive Order, Colleges Hacking Rankings
Trying out a new format where I have opinions and use capital letters.
openai keynote (video)
Overall: Meets expectations. No crisp vision, doing some of everything, incredible momentum. Experimenting with UX innovation.
Security: Have we solved prompt injection? The more plugins and data sources and access to personal services OpenAI provides the more opportunities for hackers. I’m not convinced there’s no fundemental vulnerability in OpenAI’s approach to plugins.
Network Effects: I wasn't sure OpenAI would have sufficient a network effect to win with their first mover advantage. Now, if someone comes out with a better model in a year, that might not be enough to get people to slowly switch, as there will be "GPTs" and Plugins only available on OpenAI.
Satya Nadella: He goes on stage for only 3 minutes, and he drops this line shortly before leaving: "Ultimately, AI is only going to be useful if it truly does empower." He’s clearly focused on building tools to empower. Sam Altman just caught the wave of LLMs, which is enough to be hugely successful, but he doesn’t have the vision that Nadella does.
To this day, Bill Gates thinks Microsoft was months away from winning in mobile. Windows Mobile arrived way before iOS and Android. By 7 years.
Steve Jobs did not expect maps to come to mobile. But Apple built a device that enabled services like maps. Visionaries enable innovation during uncertainty, they don’t always innovate themselves.
The AI development/regulation implication: Embrace uncertainty.
“Require that developers of the most powerful AI systems share their safety test results and other critical information with the U.S. government”
I am not convinced the Goodhart’s Law (“Any measure that becomes a target ceases to be a good measure”) is a trap. In theory, over-optimization should be hackable if you set good goals.
That is the trap of “Optimize Guys” - SBF optimizing for EA, Richard Freeland optimizing for US News college ranking. It worked for Freeland.
US News college ranking wall-of-shame - list of schools that reported incorrect data to be used in ranking calculations.
Nothing has changed.
Effective executives do the right things.