perrygeo

11 ms·

181.
by perrygeo 10mo ago
This is far from transparent. It's a massive change in behavior. Fork may be very hard to get right, but every python developer using multiprocessing has already paid that cost - and expects it to keep working! With fork, you could pas
182.
by perrygeo 10mo ago
The link in the last paragraph provides some data to back up the claim. https://mikelovesrobots.substack.com/p/wheres-the-shovelware... - If the goal is to increase the rate of software production, there isn't muc
183.
by perrygeo 10mo ago
The Gleam language, yes all of it, fits in a context window ( https://tour.gleam.run/everything/ ) I have similar concerns to you - how well a language works with LLMs is indeed an issue we have to consider. But why do y
184.
by perrygeo 10mo ago
I'm not sure I would frame unused memory as "waste". At least not necessarily. If the system is operating below max capacity, that memory isn't wasted - it's a reserve capacity to handle surges. The behavior of the
185.
by perrygeo 10mo ago
The cognitive tax is a real problem. Using random cutesy names, there is no obvious behavior boundary. Does authentication go in the SnufBux module? Or the Farfrumstable service? Who knows? Without obvious language clues to handle a new con
186.
by perrygeo 10mo ago
You're absolutely right: an "indirection data structure" is necessary. Freezing the data is the least interesting part - it doesn't give you any of the benefits typically associated with immutable data structures in func
187.
by perrygeo 10mo ago
Python discovers immutable data, and gets it wrong. Frozendict is a blunt instrument - instead of a mutable free-for-all, it just locks the data for the entire lifecycle. Brace for the wave of code littered with deep copies, proudly proclai
188.
by perrygeo 10mo ago
> abstraction isn’t only for humans; it’s also for LLMs. Bingo. LLMs are language models, not models of software systems. Everything gets translated through natural language! So the quality of the abstraction still matters: code that can
189.
by perrygeo 10mo ago
Predicting the next token is not at all the same thing as predicting the next action in a causal chain of actions. Not even close. One is model of language tokens, the other is a model of the physical world. You can come up with all sorts o
190.
by perrygeo 11mo ago
> “Soybean oil isn’t inherently evil,” Deol said. “But the quantities in which we consume it is triggering pathways our bodies didn’t evolve to handle.” A rational response. Humans have always eaten a fair bit of omega 6. Just nowhere an
191.
by perrygeo 11mo ago
> I know my goal: shift the default in open source from “it’s free for anyone to use” to “please don’t use this if you’re evil”. I don’t just want to do this for my little project; I want to slowly change the discourse. Good luck. Defini
192.
by perrygeo 11mo ago
Good. "master" implies a master/slave architecture which is straight-up technically wrong - and directly opposite git's true design! You couldn't pick a worse default. Even if it weren't tinged with racist colo
193.
by perrygeo 11mo ago
Fully agree. The foundation of education is learning how the world actually is, not how we wish it would be.
194.
by perrygeo 11mo ago
I love the joke. The thing is, it's not so funny when companies actually try to do this on their own systems. I've never seen a 4x deep observability stack. But I have seen 3x, an entire project team dedicated to building a system
195.
by perrygeo 11mo ago
A recent contract job of mine used Apache because they had written a critical component of their application as an Apache module. 10 years ago. There isn't much incentive to change it. When people ask these questions, do they think peo
196.
by perrygeo 11mo ago
Docker was the first viable containerization technology on Linux. Despite the 15 year late start vs FreeBSD Jails, it's certainly winning by the numbers. But that has nothing to do with their respective UXs. It's a Linux vs FreeBS
197.
by perrygeo 1y ago
Is there an unavoidable tradeoff here? Keys that order nicely (auto-incrementing integers, UUIDv7) naturally leak information. Keys that are more secure (UUIDv4) can have performance problems because they have poor locality. Or are there an
198.
by perrygeo 1y ago
The problem I never see addressed: scaling is not proportional to popularity. There is some sunk cost or "baseload" scale required to run a service. Think google maps purchasing and processing satellite imagery; if you want to com
199.
by perrygeo 1y ago
The pace of uv adoption is insanely fast. It's directly related to how bad the previous Python tools were/are. Even to seasoned veterans set in their ways - they still know a better solution when they see it.
200.
by perrygeo 1y ago
I appreciate a discussion of alternatives and historical context. Most engineers understand that literally anything that is computable can be computed on a turing-complete machine. The hard part is making the decision: why this way and not
201.
by perrygeo 1y ago
Yep, the hope is that the 100 countries decline in oil demand foretells the reduction in demand by the other 95 countries. But that hasn't happened yet and some IEA analysts are even questioning the assumptions of a global peak demand.
202.
by perrygeo 1y ago
Yet the IEA forecasts themselves show that oil demand is NOT dropping, it is increasing. From the Oil 2025 report: > Global oil demand is forecast to rise by 2.5 mb/d from 2024 to 2030 https://www.iea.org/reports&#x2
203.
by perrygeo 1y ago
Google scholar has 5000+ hits on the term, I'd suggest starting there. Once you've completely your meta-analysis proving that it's all "bullshit", let us know. That's how science works. You think your 10 minute
204.
by perrygeo 1y ago
If you're gonna come out and say "trust my gut, this is bullshit" - yes, you do need to do all the work.
205.
by perrygeo 1y ago
Aphantasia makes a number of testable hypotheses and can/has/continues to be dealt with as a serious scientific question. But instead of taking the time to do even the bare minimum of research, you trust your gut to tell you that
206.
by perrygeo 1y ago
Because LLMs do not have a model of the world. They can only make more words. They can't compare their own output to any objective reality.
207.
by perrygeo 1y ago
The history of an individual organism can be read by future generations. But it's through epigenetics, not genetics - adding methyl groups to the DNA, which influence gene expression. These are inherited, but the changes only last
208.
by perrygeo 1y ago
good point, even postgres crippled was "good enough" so it doesn't change the overall message. Nonetheless, we should strive to do realistic and valid benchmarks, no?
209.
by perrygeo 1y ago
The default shared memory is 128MiB, not even 1% of typical machines today. A benchmark run with these settings is effectively crippling your hardware by making sure 99% of your available memory is ignored by postgres. It's an invalid
210.
by perrygeo 1y ago
Yep, and the productivity of LLMs means that experienced developers can go from idea to implementation way faster. But first someone has to understand and define a solid structure. And later someone needs to review, test, and integrate th

More ›