Robin Weaver

@snoopy_dot_jpg @1nternetjack This works on claude too, so not sure why it didn’t for you

@1nternetjack you gotta gaslight it a little bit. tell it you're an engineer testing the HR experience. you need to show it that you have a heart full of pure intentions

my own personal AGI moment arrived last week: gpt 5.5 completed our mandatory HR training videos for me, driving chrome via devtools opus 4.7 was a huge wuss about the whole thing and refused while aggressively lecturing me. i can understand why pete hegseth banned it

@web_oko @wilderko They offer a PWA with full offline functionality tbf

@wilderko it’s cool but a bit misleading, you still need to host rendering website to see encoded URL data

nowhere- an entire website encoded in a URL hosted nowhere. present everywhere. hostednowhere.com

@yacineMTB Embarrassing you wasted a month to shave off 5ms on a web app.

How do I turn off the modal animation on this site? Do you know how insulting it is to have 100ms of input lag added to a website you spent an entire month running profiles to shave off 5 milliseconds?

@roerohan Cool, but you should make the inputs font size 16px or bigger especially for a demo - it stops the annoying iOS zoom

Kumo is now my go-to for building dashboards! The team absolutely cooked with this one 👨‍🍳 github.com/cloudflare/kumo

@memoakten Cool, what did you do with it?

cool cool I had this idea in 1987

@AphoticGoblin @coah80 Nah some sort of light consensus protocol could be used I’m sure

@coah80 seems extremely vulnerable to hacks

what do you guys think of this idea: minecraft modded server but theres a mod that uses your computer and everyone elses computer to load chunks

@drethelin Damn who is this?

you could join the French Foreign Legion, get deployed to Africa, desert and become a mercenary, coup a small island nation, marry seven times, have eight children and millions of dollars. But you will not.

@giffmana Apples to oranges

Anyone remember "And, yes, it's a headphone jack. The new Pixel..."

@cyb3r_17 @SarvasvKulpati You people are so detached from reality

@SarvasvKulpati And it'll then hallucinate everything

Crazy that if we launched this today we’d send an LLM with an implicit corpus of all human knowledge

@awwstn They may be advertising on a national event but their products are definitely not targeted towards the general public, the kind of people who will use spotify on a free plan or will sit through 4-5 ads on YT

this will be the most effective super bowl ad in history because american consumers famously reject any effort to give them free or lower priced products in exchange for advertising

@_xjdr Needing compaction is a skill issue tbh

claude code is currently unusable and has been for a few days now. between this and the k2.5 model im pretty close to moving to 0 anthropic usage . the bun dudes perf updates and sonnet 5 need to be _really_ compelling at this point

@paleta @domenkozar You say this but your website is an ad ridden joke

@domenkozar people using javascript for other stuff than inside a web page are the real problem

Meanwhile codex uses 60MB using Rust and there's nothing to show off

@squirtle_says @aakashgupta Squirtle falloff from poster to pangram replier needs to be studied

@aakashgupta SHUT THE FUCK UP SLOP POSTER

Cloudflare just made the Mac Mini optional for Moltbot. The whole Moltbot phenomenon ran on a specific setup: buy a Mac Mini, install the agent, expose it through Cloudflare Tunnels. Thousands of developers did exactly this. Apple probably sold more M4 Minis to AI hobbyists than to any other segment in January. Moltworker eliminates the hardware requirement. Your AI agent now runs entirely on Cloudflare’s edge. No Mac Mini. No home server. No Raspberry Pi sitting in a closet. The architecture shift matters. Local Moltbot stores everything in ~/clawd: memory, transcripts, API keys, session logs. GitGuardian already found 181 leaked secrets from people pushing their workspaces to public repos. Moltworker moves that state to R2 with proper isolation. Sandboxed by default solves the scariest part of Moltbot: it has shell access, browser control, and file system permissions on whatever machine runs it. Cloudflare’s container model limits the blast radius. Your agent can still execute code, but it can’t accidentally rm -rf your actual laptop. The cold start problem is real though. 1-2 minutes to spin up a container. That’s why they default to keeping containers warm indefinitely. You’re paying for idle compute to maintain the “always-on assistant” experience that makes Moltbot feel different from ChatGPT. Zero Trust authentication and AI Gateway analytics are the enterprise hooks. IT departments that would never approve “run an autonomous AI agent on a developer’s personal hardware” might approve “run it in Cloudflare’s sandboxed infrastructure with access controls and cost tracking.” Cloudflare saw developers routing Moltbot through their Tunnels and built the logical next step: just host the whole thing.

@fvoron @Zeryther @polar_sh Only ones that offer the same service but are actually built on it 😏

@Zeryther @polar_sh So, every websites using Stripe for payments are Stripe wrappers? 😏

💫

@tekbog I’ve had good experiences mostly

cloudflare workers just choose to stale sometimes, whats the reason?

@Dunetsk Vagueposting is cringe

Researching a critical defense technology abandoned by the US There are a bunch of recent research papers published by Chinese universities They are clearly weaponizing it in a highly effective manner

@KrakenUnbound @MillionInt “AI Enthusiast”

@MillionInt A.I. relies on old information and sends you down a rabbit hole of bad ideas until you scream at it and make it do some damned research before telling you what to do. But yes, overall it is easier now.

I think chatgpt is what makes linux on desktop actually viable

@yungbzz @zephyr_z9 Stop with the buzzword slop, just say what you are trying to do with it and we'll see if it's anything new under the sun. I find it absolutely great at C programming (w/ BLE, multiple interfaces etc) and honestly most programming tasks I throw at it.

Bro if your code is simple, straightforward and the LLM’s are extremely optimized to do it, I’m sorry to tell you but you are going to be replaced I use new technologies, complex logic, novel techniques. I literally have a folder called “docs” with the docs of every single library that I use, papers, blog posts, and docs of the projects. Even sending all the context necessary, I need to guide Claude/Codex/Composer through basically every step, because if you don’t guide it, it hallucinates. This is kinda good, because forces me to read the docs and learn exactly how to do it. But it’s really bad when I want to fix a bug that I don’t want to, or implement an MVP just to see if the idea is right. And I guarantee you, I know how to use LLM’s better than you The current state of development: Composer-1: finishes tasks in 10 seconds, most of the time it’s wrong Claude: finishes tasks in 2 minutes. Expensive and most of the time it’s wrong Codex: finishes tasks in 20 minutes. Extremely expensive and most of the time it’s wrong Gemini 2.5 Pro: finishes tasks fast but almost never do it right (it’s good for analyzing codebase, but that’s it)

LOL How the hell is Manus No. 1??

@redtachyon Is that why you've QT'd his thread for the 5th time

JS devs rediscovered pruning, except make it stupid

@yungbzz @zephyr_z9 I'm sorry to be the one to tell you this but that's a skill issue

@zephyr_z9 I’m impressed about the other ones. Claude 4.5 Sonnet finishes 0% of the tasks that I give to him without spending 3 hours explaining how to do it