We can't find the internet
Attempting to reconnect
Something went wrong!
Hang in there while we get back on track
Top pick: Satya Nadella, "Models as Insider Risks in the Super Intelligence Era"
Microsoft's CEO published this essay on X. It argues that because we can't trace model behavior the way we trace code, we "need to separate the supply of intelligence from the authority over it" . The essay proposes treating closed and open-weight models as insider risks, "Not because they are necessarily malicious," but because any capable actor can make mistakes or be compromised . It lists principles including model diversity, tamper-proof logging, independent controls and audits, and containment. It closes with: "The most trustworthy Super Intelligence system will not be the one with the model we trust most. It will be the one that enables us to trust the model the least" .
Four people endorsed it, each for a different reason:
- Chamath Palihapitiya called it "very important observations." His takeaway: as models commoditize, "intent, verification and evidence" will decide how fast AI spreads through enterprises. He thinks a new system of record may emerge to map changes back to SOPs, policies and laws .
- Aaron Levie called it a "Good post" for thinking about enterprise safety. In his words, AI will need its own "zero trust" era like the one security went through, which is "a huge opportunity right now for those building such systems" .
- David Sacks wrote "Satya is right." He uses the essay to contrast engineering controls with Anthropic-style "alignment," arguing that models should be treated as "insider risks, not moral patients" .
- Elon Musk: "Interesting piece from CEO of Microsoft" .
Tobi Lütke: Patrick Collison's thread on agents in the economy
The last brief covered Matt Mullenweg's reply to this thread. This time Lütke called it a "Great read from Patrick, as always." He singled out one line, "Personal agents will, I think, act as a kind of structural subsidy for product quality," and added "Yes, and well put" . Collison's argument is that if agents research every purchase, improving the product becomes a more effective strategy than winning on distribution . Thread
Patrick Collison: Grossman's Life and Fate over The Brothers Karamazov
In a long post, Collison admits he "did not love" Dostoyevsky's Karamazov . He offers Vasily Grossman's Life and Fate as "maybe a useful counterpoint," because it "shows actual people confronting great wickedness" while the Karamazovs are "never tested" . In his reading, Grossman rejects any providential resolution and argues "humanistically, that we should be kind just because" .
Aaron Levie: Prime Intellect's agent-swarm post
Levie shared Prime Intellect's post with "If you want to know where all the tokens will get consumed in the future, here it is." His reasoning is that agents can parallelize work in ways human teams can't, "This is why we're going to need 1,000X more compute" .
Shorter mentions
- Sarah Guo promoted the new No Priors episode (her own podcast) with Reflection AI cofounder Misha Laskin. It covers China, safety, open-model business models, distillation and shipping frontier AI. Her pull quote: "Competitive tension is just good" . YouTube
- Elon Musk endorsed a16z's chart post, which argues "There are no low electricity, rich countries" . He added that electricity production is "the best metric for the true strength of any large-scale economy imo" .
- Jason Lemkin promoted SaaStr's own piece with Atlassian's Head of AI, with the takeaway "Juniors beat mids now" .
- Balaji Srinivasan, speaking with Vitalik at the Network State Conference, said he likes Gilbert Strang's linear algebra for people starting out. He then asked Vitalik which references he recommends for going "back to basics on cryptography, number theory." Vitalik answered that talking to an AI bot has been an effective way to learn . Watch