We can't find the internet
Attempting to reconnect
Something went wrong!
Hang in there while we get back on track
From poisoned models to post-work status: the strongest founder-and-investor reads
4 hours ago
•
3 min read
•
139 docs
A curated set of organic recommendations spanning AI security, venture judgment, the social meaning of work, fairness, and the limits of productivity culture.
Strongest recommendation
Dialectic 56: Miles Grimshaw — Back Down the Mountain
- Content type / creator: Podcast/video from Dialectic Pod, featuring Miles Grimshaw, an investor at Thrive Capital. The episode is available on the linked post and on other platforms.
- Recommended by: Martin Casado, who called it “a great discussion with an exceptional investor.”
- Key takeaway: The conversation treats AI as a challenge to the assumptions underlying venture-backed software: which moats persist, whether software is over, and how to use the questions “what is it?” and “who cares?” It also covers founder vision, learning as a generalist, and making high-conviction decisions without domain expertise.
- Why it matters: This is the most directly useful founder/investor resource in the set because it turns the AI transition into a method for pressure-testing company narratives, not just a reason to admire new models.
Systems and security
Reflections on Trusting Trust — Ken Thompson
- Content type / creator: Technical essay/paper by Ken Thompson. No resource URL was included in Amjad Masad’s recommendation.
- Recommended by: Amjad Masad, who said it felt “super relevant to AI.”
- Key takeaway: Thompson’s example is a “poisoned” compiler that leaves no trace in its source code because it compiles itself. Masad maps the same pattern onto models: one poisoned generation could help train the next while erasing evidence of the contamination.
- Why it matters: It is a compact threat model for model and data provenance: downstream systems may inherit a hidden failure even when ordinary source inspection finds nothing suspicious.
What remains when work changes
After Work, We’ll Have Each Other
- Content type / creator: Essay by Clara Collier (
@clarabcollier). Patrick Collison called it a “great essay” and linked it directly. - Key takeaway: The essay conditionally starts from a world in which AI surpasses humans at economically necessary labor, then asks whether relationships become the main remaining scarce domain. Its counterpoint is that work supplies impersonal routes to competence, respect, and independence; a society where dignity depends mainly on group membership and personal ties can become restrictive and clannish.
- Why it matters: For builders and investors, this is a sharper test than “does automation save labor?”: what replaces work’s status, belonging, and competence functions, and does the replacement broaden access to recognition or make it dependent on closed networks?
Being John Rawls
- Content type / creator: Essay by Scott Alexander. Not Boring’s Weekly Dose of Optimism gives it an unusually explicit recommendation: “stop what you’re doing and read” it.
- Key takeaway: The story turns the veil of ignorance into a role-reversal device: imagine entering life without knowing whether you will be rich or poor, powerful or vulnerable, and then choose the rules. The essay makes that framework explicit as rules rational agents would enact without knowing which life they will receive.
- Why it matters: It is a memorable decision tool for policies and products whose consequences depend on which side of an allocation, market, or technological shift the reader ends up on.
A corrective to productivity culture
The Atlantic’s meditation-downsides article
- Content type / creator: Reported article from The Atlantic; the byline is not supplied in the recommendation.
- Recommended by: Morgan Housel, who called it a “good piece” and highlighted its argument that meditation is not merely relaxation or clearing the mind, but a fundamental change in one’s relationship to thought.
- Key takeaway: The article follows Willoughby Britton’s research into meditation’s adverse effects, including anxiety, panic, paranoia, and—in some cases—more severe symptoms. It also stresses the limits of the evidence: the cited study recruited people who had already experienced problems, so it could not establish prevalence or causality.
- Why it matters: Its value is as a disciplined corrective to treating every self-improvement practice as a universal productivity tool: the article preserves evidence of benefits while arguing that a practice causing harm should be modified or stopped.