Overview
Today’s feed had two main threads: how we think (and how tools can either sharpen or blunt it), and what we choose to fund or neglect, from science budgets to biology moonshots. In between, there was a run of practical warnings, from toxic garden turf to misleading charts and sloppy headlines.
The big picture
There’s a quiet tension running through these posts: we want faster creation and easier automation, but we also want judgement, craft, and basic accuracy to hold steady. The best notes today weren’t anti-tech, they were pro-discipline, whether that means writing without crutches, building better knowledge systems, or refusing to be fooled by bad data presentation.
Write more, unassisted, if you want to keep thinking clearly
@thedankoe’s blunt point landed because it’s hard to argue with: if you always outsource the hard parts of thinking, you do less of it. Writing from a raw idea, without prompts or templates, forces you to find the shape of what you actually believe.
It’s also a useful counterweight to the growing habit of treating every blank page as a tooling problem. Sometimes the “setup” is just time and a bit of discomfort.
Stop searching your notes, compile them into a living wiki
Two posts circled the same idea from Andrej Karpathy: LLMs are more interesting as knowledge engineers than as chatty search boxes. Instead of asking your docs questions over and over, you keep a maintained wiki that gets compiled, cross-referenced, and checked for gaps.
The appealing bit is the division of labour: machines do the grinding upkeep, humans decide what matters and whether it’s correct.
AI adoption as a business skill, not a magic wand
@gdb shared results from a field experiment suggesting that simply showing founders concrete AI success stories nudges real behaviour: higher usage, higher revenue, and less capital needed. It’s a reminder that the bottleneck is often confidence and know-how, not access.
If this holds up across more settings, “learning to use AI well” starts looking less like a niche advantage and more like basic commercial literacy.
Building apps in minutes is getting normal, and that changes who gets to ship
@Scobleizer’s hackathon story is peak 2026: credits, an API, and two working apps in about 20 minutes, including one that turns his curated X lists into a live monitoring tool. The takeaway is not that every app will be good, but that the distance from idea to deployment keeps shrinking.
That pushes the centre of gravity towards taste, data sources, and distribution, and away from who can wrestle a framework into shape.
Open-source agent “super-setups” are here, with all the usual trade-offs
@RoundtableSpace promoted an open-source Claude coding setup with dozens of agents, commands, and security checks. For teams that know what they’re doing, a repo like this can save days of fiddly groundwork.
But it also raises the obvious question: how many moving parts do you want in the loop before you lose track of what is happening, and why?
Gene therapy for deafness, and a glimpse of medicine that fixes root causes
@Polymarket pointed at a small trial where an inner ear gene therapy helped all 10 patients with a specific OTOF mutation. That “10 out of 10” headline will tempt hype, but the real story is narrower and more exciting: targeted treatments that address the underlying genetic issue, not just the symptoms.
It’s early days, yet it’s hard not to read this as a preview of how more conditions could be tackled when delivery and safety keep improving.
Frontier biology wishlists meet the reality of science budgets
@brian_armstrong listed underfunded biology areas that could speed up progress, like cheap DNA synthesis and better sequencing. It sits awkwardly alongside the other conversation today: what happens when governments starve the research base that feeds these breakthroughs.
The gap between “we should do this” and “we are paying for it” feels like the story underneath the story.
Science cuts, as a policy choice with a long tail
@ylecun’s dry “Tired of winning” was aimed at proposed US budget cuts to major science agencies. The sarcasm works because the consequences are so predictable: fewer grants, fewer labs taking risks, fewer people trained, and fewer compounding gains.
Even if you only care about competitiveness, kneecapping research is an odd way to show it.
Toxic turf: the hazards you don’t notice because they’re marketed as tidy
@bryan_johnson admitted he’d been obsessing over health while sitting on artificial turf that can contain crumb rubber and an alphabet of chemicals. The wider point is familiar: modern “maintenance-free” choices often come with hidden exposure you only learn about once you go looking.
If you have kids, pets, or you spend time on it barefoot, it’s a topic worth a proper read, not a quick shrug.
Chart crime and headline errors, the small mistakes that steer big opinions
@fchollet called out a common stats trick: plotting autocorrelated time series as if they were independent points, which can make weak relationships look convincing. In a different corner, @JeffDean pointed to a major outlet mangling “North Atlantic” into “North American” in a NATO headline.
Different failures, same lesson: confidence is cheap, verification is work, and the audience pays when nobody does the work.
