Announcements
Happy fourth of July everyone!
Weekly Shoutout
@Val and I got to hang with @Alice this week and treat her out to dinner as a belated wedding gift!
Food for Thought
This week I’m focusing on the bear case against LLMs — it’s good to steelman every once in a while ahah
First, it’s clear that we are a long way from “agents”, mostly because the longer the task the more the LLM starts to break down, even in domains like coding
Second, the limits of LLMs are more clear in domains we are already experts in, so they can make us overconfident in domains we are not — just like newspapers
Third, most of these LLM companies are just burning through money, which means that eventually they’ll be forced to quantize their models right after we’ve become addicted to higher levels of intelligence
Fourth, we have hit diminishing returns on the data we can get from the internet, so now we are forced to use AI to make new data which is…
Crowd Pleasers
“If women are going to have it all, then it will only be possible with a really supportive husband” — Zucca Girl
Some data suggests that men doing more chores literally makes women more likely to have more kids ahah
Mind-Benders
“True wealth comes from ownership, not fees. The best investments are often in yourself—they compound forever and no one can dilute your stake.” — Peter Thiel
Larry Ellison is a similar story of betting on oneself
Ellison is also a major backer of Elon, showing the compounding value of doubling down on long-term partnerships with friends
The idea that you can purify training data using an AI trained on the same data is so laughable