पिन किया गया ट्वीट
Astrohacker
278 posts
Astrohacker
@AstrohackerLabs
Advanced cyber technologies.
Austin, TX शामिल हुए Ekim 2022
817 फ़ॉलोइंग279 फ़ॉलोवर्स
English
“Which no professional engineer does”
Seems like you are not a professional then. Everyone fucking does it because the expectations are 10x output and you can’t achieve that if you thoroughly review the code the AI generated.
Right now all over the industry most companies value reckless engineer more, who ship slop at x100 pace and now the more careful, thoughtful engineers are the “bottleneck” who refuse to let AI slop through.
English
Sensible voices are (finally) breaking through with AI. @mitchellh one of the best ones.
“I use AI all the time. I like AI. The point I'm making is to not blindly accept results. Think. Analyze. Learn.”
Read it 👇
Mitchell Hashimoto@mitchellh
I've got an agent in a loop optimizing a renderer with the goal to minimize frame times (and tests to measure). It got times down from 88ms to 2ms and allocations down from ~150K to 500. Sounds good, right? Wrong. This is exactly why agent psychosis is a big fucking problem. As an experiment, I rewrote the Ghostty core render state in Go, with access to identically laid out data structures as Ghostty and the exact same validation tests. I made a purposely naive renderer (simple, correct, but slow). 88ms per frame with 150,000 allocations (horrendous, lol)! I then kickstarted a Ralph loop to bring the frame times down. I told it it can't modify input data structures or the public API or tests (they're correct), but it can do anything else it wants. It got to work. It has worked for about 4 hours. I've spent around $350 on this experiment so far. The results? 88ms => 1.5ms 150K allocs => ~500 allocs Incredible right? Nope. My hand-written renderer I ported has frame times (same benchmark) of ~20us (0.020ms) and 0 allocations in the update path. This is the problem with psychosis and lacking systems understanding. If you don't understand the system, you're going to accept that this is an incredible result. If you understand the system, you'll see better solutions immediately and can do roughly 75x better on throughput. The people who blindly trust agent output are in the former camp. They're sheeple, overdrinking from a fountain of mediocrity. Standard disclaimer: I use AI all the time. I like AI. The point I'm making is to not blindly accept results. Think. Analyze. Learn.
English
@rezoundous you can have claude review codex. that's working pretty well for me.
English
@aryanlabde today i had to switch back to gpt 5.5 because opus 4.8 kept having server errors. will have to save my opus 4.8 experiments for a future date.
English
was gonna try claude today but it kept having server errors and failing. have to switch back to codex again.
English
@garrytan not only are they not the end of startups, they are the beginning of a new golden age for startups
English
Oh man, this is going to be a very tired argument we are going to have for the next 5 years
Garry Tan@garrytan
The real AI whitepill is: There will always be a market for translating what the 200 IQ people and agents make down to everyone else
English
@naval i was thinking the exact same thing 10 times this morning
English
@mitchellh strawman. there is no professional engineer who "blindly trusts agent output".
English
I've got an agent in a loop optimizing a renderer with the goal to minimize frame times (and tests to measure). It got times down from 88ms to 2ms and allocations down from ~150K to 500. Sounds good, right? Wrong. This is exactly why agent psychosis is a big fucking problem.
As an experiment, I rewrote the Ghostty core render state in Go, with access to identically laid out data structures as Ghostty and the exact same validation tests. I made a purposely naive renderer (simple, correct, but slow). 88ms per frame with 150,000 allocations (horrendous, lol)!
I then kickstarted a Ralph loop to bring the frame times down. I told it it can't modify input data structures or the public API or tests (they're correct), but it can do anything else it wants. It got to work.
It has worked for about 4 hours. I've spent around $350 on this experiment so far. The results?
88ms => 1.5ms
150K allocs => ~500 allocs
Incredible right? Nope.
My hand-written renderer I ported has frame times (same benchmark) of ~20us (0.020ms) and 0 allocations in the update path.
This is the problem with psychosis and lacking systems understanding. If you don't understand the system, you're going to accept that this is an incredible result. If you understand the system, you'll see better solutions immediately and can do roughly 75x better on throughput.
The people who blindly trust agent output are in the former camp. They're sheeple, overdrinking from a fountain of mediocrity.
Standard disclaimer: I use AI all the time. I like AI. The point I'm making is to not blindly accept results. Think. Analyze. Learn.
English
@hthieblot tell people about it
good idea, i'll have to try that
English
The plan is simple:
1. Build cool shit.
2. Tell people about it.
Repeat forever
English
yep. the stuff i'm doing with ai simply transcends this whole 2010s-era thinking. new things are possible now.
Garry Tan@garrytan
Founders must stop trying to building 2010-era businesses with 2026-era technology. Don't try to rebuild Foursquare or Yelp. Don't try to recreate Basecamp by 37 Signals with $10/mo SaaS pricing. Don't underprice! If it works it's worth a lot more. Don't be tempted to become "Tech enabled PE" with revenue tricks. The rules of tech changed with AI. Play the new game.
English
@gdb codex is the absolute greatest tool i've ever used, hands down
still underrated
English
the number of rocketships is increasing exponentially, and they're going further than ever before.
Beff (e/acc)@beffjezos
spoiler alert: there is an infinite hierarchy of difficulty of problems to solve in our universe, and infinite value to unlock from solving them
English
"So far, Mythos Preview has found what it estimates are 6,202 high- or critical-severity vulnerabilities in [more than 1,000 open-source projects, which collectively underpin much of the internet]"
anthropic.com/research/glass…
English
yt-dlp support of Bun is now deprecated, citing the rewrite to Rust being "vibe coded" and issues surrounding supply chain attacks
English
yep. hand-coding stopped being a thing in 2025.
@levelsio@levelsio
I don't write code anymore I haven't written code in I think 6 months? I think everyone is like this no?
English
@thsottiaux haven't tried this yet but i intend to. thank you for enabling this - very good feature!
English
I used my desktop app + Gemini 3.5 Flash to create this sci-fi inspired dashboard for my MacBook system stats. Not all the stats shown are real but I think flash did a pretty great job building a working Electron app that looks cool, no intervention needed this was one shot.
English