pphysch

Nope, negative feedback is dampening and stabilizing. Antitrust enforcement represents a healthy negative feedback for the natural runaway concentration of wealth and power.

This proposal is literally about making it easier for ordinary code to be written without defining new generic types or iterators.

At this point, the "damage" has been done and even skeptics of generics/iterators should applaud this proposal since it means burying some of that added complexity behind straightforward stdlib interfaces.

Unless they are just here to complain, of course.

Exactly. If you ask anyone on the street, "is it easier for a fast food worker or a business-owning millionaire to make their next million dollars?", common sense will prevail.

Money begets money (and power). It doesn't require an economics degree to understand this positive feedback loop.

I got a cold email recently and was happy to engage, but I warned them upfront that LLMs change the game, and asked for more info about their background (to authenticate them as a human). They replied with another unnecessarily long email that was light on details and reeked of AI Voice, so I ghosted at that point, unconvinced they were acting in good faith.

Demonstrating authenticity and effort are super important in this space.

> For us, that means interactive work, automatic work, and production work would share one well-crafted underlying system instead of living in separate tools.

Unless this is only for the narrow band of purely local work, I would be surprised if k8s didn't play a big role in the backend solution. It is the standard "well-crafted underlying system" for automatic production work. But, it isn't strong in the interactive-work/frontend aspect or other aspects, so it certainly wouldn't be the main focus of the final product.

Random example: you want to multiplex some work across multiple (self-hosted) agents, using niche fine-tuned models. It would make good sense if those agents were automatically provisioned and deprovisioned (and isolated to your subtenant), and k8s is a great underlying system for that sort of thing.

People clown on Alex Karp for his unedited maniacal "crashouts", but this is a real public crashout that made it past a team of publicists.

It's standard regulatory capture.

You don't say "let's ban my competitor".

You say "let's create laws that make it uneconomical for my competitor to access the market".

Personally I don't trust the (real, actual) barbarian horde when it tries the ole switcheroo and projects barbarism, without evidence, onto its victims.

"The baby I just sniped in the head was a confirmed Hamazbollah IRGC terrorist." Sure, buddy. Hasbara is for losers.

Completely different geopolitical scenario. If USA exterminated Germans/Japanese then their Soviet/Communist neighbors would fill in the power/population gap.

It's also worth mentioning that the Germans and Japanese themselves were extremely brutal to their defenseless, surrendered neighbors that they attempted to colonize.

Iran will never surrender because US/Israel proved themselves unequivocally to be brutal savages that will torture, rape and kill you if you surrender. There is literally no point in surrendering to a barbarian horde.

I did not first bring up fairness in this thread.

And what does Iran get in return for America butchering 160 innocent kids, or assassinating their leader? Or does "fairness" only flow in one direction?

pgdump / pgrestore, using native binary format

Meta commentary: this is clearly a vibe-coded frontend to a low quality "dataset" with substantial omissions.

The real value and work would be someone actually compiling and creating a high-quality castle dataset, in which case anyone could vibe-code a similar frontend to it.

Potentially a agent (swarm) could do this, if properly integrated into OSM, wikis, libraries, photographs etc. But it wouldn't be quick and would still benefit from a human-in-the-loop verifying each entry.

Do you think the average vibe coder or AI-smitten suit cares about slim code? A lot of folks still think "LOC = good" in 2026. It will be a hard sell for the practical engineer.

Even if they understand "slim code = fewer tokens = less spend" they still have to grok that less code is not less functionality.

> Any agent, any model, any scaffold: the bill is essentially O(reads)

Therefore token sellers have incentive to produce verbose, unreadable code

How are roving mobs of luddites going to burn down your datacenter in deepest Antarctica?

Well, at least the organizers that care. There's definitely a class of grifter organizers that view events as an opportunity to profit from high entry fees and low production quality / relying on volunteers.

Datacenters in Antarctica or deep ocean also fulfill that.

This is the kind of thing that is totally technically plausible, and useful to intelligence groups but not the general public, and therefore I would assume it is 100% in production in international intelligence agencies. But nothing we can "prove" without getting shot.

Similar to ultra-accurate multimedia geolocation models.

I'm not saying "only use Postgres", but you need to explain why core graph data needs your listed features.

They should rebrand it as "Praise the Sun" mode. We are sorry, GOES-19 is temporarily unavailable during a planned solar worship break of indefinite length.

Nice basic advice but ironic coming from him

This AGI silliness is predicated on a flawed "Platonic" view of epistemology. The notion that there is some well-defined "idea space" and therefore superintelligence can explore that space faster than any human. In fact, there is no such space. There is a "token space" that can be explored, but that has only fleeting overlaps with reality.

This is an interesting perspective on the concept of "atomization of the individual", which the interview interestingly does not refer to. Both atomization and the "uni-context" arise from the destruction of an individual's local contexts, like family and community, leaving them an individual in a global, maximally-large pool of individuals, the so-called uni-context.

Atomization has clear motivations: increasing the individual consumer base (no, you shouldn't share your car or lawnmower with your block, you need your own), suppressing democracy, and generally making a population more predictable and easy to manage.

Graph query languages like Cypher are great, but I am wholly unconvinced by the concept of a dedicated, general-purpose "graph database".

IMO, you're better off just using Postgres/etc, modeling your graph data there, and pulling in subsets of your graph for in-memory analysis. This is for the 99% of enterprises that aren't doing online streaming graph analysis on TB-scale graphs, and the other 1% should probably figure out something tailored* to their specific business model.

* Graph algorithms are more accessible than ever with GenAI code, and efficiently modeling a graph in memory is trivial (it's just structs with pointers to other edges/nodes, plus its nice to have full control over the memory layout).

Yeah and if this were taken seriously, you would have Mechanical Turk style services where poors are paid pennies to hand-write submitted/generated slop, defeating the purpose.

I feel like you're splitting hairs. Geohot is/was selling shovels during a gold rush, full stop.

Maybe he had a recent change of heart, but his public actions (including Elon stuff) clearly put him in the "monetize hype" camp rather than the "quietly build" camp.

Geohot is one of the (attempted) merchants, but maybe that is not going so well and he is changing his tune.