Shaad

The funniest part about “treating yourself” culture is that people are stressed about money while constantly rewarding themselves with things they don’t actually need.

The craziest thing about modern spending habits is that people don’t even wait to have money anymore before buying things. Everything is: pay later, EMI, credit, “future me will handle it”. Could you imagine the concept of taking an EMI to buy expensive shoes 10 years ago ?

@rupyaoff I wish I was tracking my expenses. Now I'm homeless. Please release this asap

Modern spending habits are crazy. Food delivery, Blinkit, random online purchases, “small” UPI spends that somehow end up draining your salary before month-end. That’s exactly why we started RupyaOff. Trying to make spending online a little smarter. More to follow.

Really need this right now 😭😭

@Rukia1994443 I parry rational thought

The fact he underestimates himself so much is crazy. But being called talentless as a kid clearly left a mark.

@AzurLane_EN Aha?

【Preview: Mogador's Dual-Form (L2D) skin - Diving for Desire】 We're jumping! Ahah♥ Isn't this feeling of weightlessness just the best? You and me, Commaaander, falling and pressed together... You want this to last forever too, don't you? *Content is in development and may not reflect the final version. #AzurLane #Yostar

this video kills me all the time why does he have a 3 second realisation of his existence and forgets it a frame after

Remembering Sayuri 🖤 The voice behind so many unforgettable anime songs, like Erased, Scum’s Wish and Lycoris Recoil, passed away in September 2024 at just 28 years old. Her music had this raw, emotional feeling that hit differently. Even while dealing with a chronic illness, she kept creating and connecting with fans through her songs. Gone too soon, but her voice and impact will always stay with the anime community.

@thekitze @benjiDotSo lol letting it cook on a full trpc ->elysia migration unattended is brave… we tried something similar and Hud.io ended up being the only way to see where it drifted from parity mid run

trying komposer 2 to migrate @benjiDotSo from trpc to elysia, let's see how it will do (i like that kursor can open projects via ssh, most agent orchestrators can't do that and it's annoying)

@ferron_web yeah it’s kinda nuts, 5.4 just casually doing full refactors now… we let it loose once and Hud.io showed a bunch of side effects we def didn’t ask for lol

Codex with GPT-5.4 is doing wonders here... 😄

@maria_rcks came back to a full PR + postmortem is wild lol, we had codex do something similar and Hud.io was the only way we figured out what actually ran vs what it thought it did

left codex to do some stuff and came back to this

yeah this is the stuff people underestimate, shaving latency + cost off “boring” paths compounds like crazy we had a similar phase and Hud.io made it obvious which changes actually moved p95 vs just felt like improvements, some tweaks look great in theory and do nothing in real traffic

I recommend working with @nateberkopec if you have a Rails app that is doing decent revenue/mo. Just from engaging with him I'm spending a lot of time revisiting old code, optimising it, and cutting costs. Old code I wrote years ago is so easily improved with all the new models we have at our fingertips. One day it's saving money with how we use S3. Another day it's planning migrations of key services. Today it's cutting job queue times in half. Tomorrow it's tweaking @Judoscale to juice more utilisation. Everyone needs a personal trainer for app performance.

this is actually such a smart way to fight codebase drift… forcing small “anti pattern cleanup” loops instead of letting the entropy win we tried something similar and Hud.io made it obvious over time which patterns were actually improving runtime vs just looking cleaner, some of those refactors sneak in weird behavior changes if you’re not watching closely

at @humanlayer_dev we are running a ralph loop that runs one iteration per day. it starts with two things in context: 1. the "you might not need an effect" post 2. "Find a useEffect and fix it" We have a couple others: - "find a useEffect and fix it" - fix the highest-priority finding from react-doctor (by @milliondotjs ) - find a smell and fix it with a refactoring LLMs are auto-regressive and follow your codebase's patterns, so you have two options: 1. allow your codebase to get worse over time as they get more-cemented in place by agents 2. find bad patterns and fix them. Make your codebase 1% better every day.

yeah frontend-only “entitlement checks” are basically just suggestions to users we had a similar leak and Hud.io made it obvious some clients were just skipping validation paths entirely, everything looked fine until you saw what actually executed vs what you assumed was enforced

When I launched ChapterPal, I offered all the subscribers of my AI newsletter free access to a paid plan for the first three months. Because this was a one-time event, I decided to keep it simple and put the validation of whether three months had expired on the frontend. Once a day, a script would check if the returning user was officially subscribed to a paid plan, and if not, this complimentary subscription would be cancelled. A couple of days ago, I noticed that the Premium OCR (where expensive LLMs are involved) was used by users who aren't on the Stripe subscribers list. After an investigation, I found 9 users who had been using their complimentary paid subscription 3 months after it was supposed to be cancelled. Most of them used ChapterPal daily. These users fooled the frontend verification script by pretending that they had already been tested earlier today, so there was no need to verify their subscription status again. Two days ago, I manually downgraded these users to the free plan. None of them have paid to subscribe.

Kyoto, Japan 🇯🇵

桔梗和傘の色の変化

@cyber_razz lol “ping works” has probably wasted more engineering hours than any real outage we had a similar case where everything “looked up” until Hud.io showed all the actual TCP calls just silently failing while ICMP was happily green

I watched a junior admin insist the VPN was fine. He logged into a workstation. He ran ping 10.0.0.45. It replied. He shrugged and said the VPN was up. I asked him if he thought ICMP was the new TLS handshake. He looked confused and said ping was standard. I told him ping tests ICMP, not TCP. The VPN could allow pings but block everything else. He wouldn't know the difference. I showed him nc -zv 10.0.0.45 443. It timed out. I told him ping costs us an hour of false confidence per incident. I disabled ICMP on the test server and forced him to write a health check that tries three different TCP ports.

@ZixuanLi_ lol that “thinking:” turning into eldritch unicode is always a great sign we saw similar corruption once and Hud.io made it obvious the model output stream was getting mangled mid token, otherwise it just looked like the model completely lost its mind

If you’ve encountered garbled output like this while using GLM-5 or GLM-5.1 on our official service, the issue is now resolved. We've patched the underlying inference-side bugs and will be releasing an article shortly to dive into the technical details and our specific fixes.

クセが強すぎる天才ピアノ少年