👻 THE REAPER'S CIRCLE: THE FOLLOW-UP Remember the Halloween Python puzzle? Nobody beat the Reaper, but @ExactlyJosh came the closest, and he did it with no AI. The thread voted, and the prize was his. Just wrapped our build. 🛠️ 🎯 WHAT WE BUILT A conversational AI agent for Josh's app that drafts and edits one-page documents: resumes, cover letters, one-pagers. You chat with it, it writes, and every document fits on one page. 🧠 HOW IT CAME TOGETHER • Built the agents as a backend his Android app plugs into • Tested on a local model, and one setting switches it to his own Anthropic API key (Sonnet writes the drafts, Haiku runs the chat) • Walked him through the flow and the repo, and added him to it 🚀 WHAT'S NEXT Josh is wiring it into his app himself now, and I'm on standby if he needs a hand. From "no AI allowed" to having his own AI agent in under a week. Love to see it. 👏 #AIagents #BuildInPublic

This is the function an LLM uses to turn its raw scores into next-word probabilities. What does it print, and why? A) [0.09, 0.24, 0.67] [0.0, 0.0, 1.0] B) [0.09, 0.24, 0.67] [0.09, 0.24, 0.67] C) [0.09, 0.24, 0.67] [0.33, 0.33, 0.33] D) OverflowError

Saturday coffee, new board game, quick thank you. To everyone who replies, reposts, DMs, or hops on a call to talk shop: I came from door-to-door sales, where most people shut the door. Here, people keep opening it. I don't take that for granted 🙏 Something I tell myself every morning: today is the greatest day in my life. Hope it's yours too 😄

In my goal to document and be transparent in my game dev experiences, here’s a funny playtest session of me finding almost 30 bugs! youtu.be/Vs0dK71Xzw8?is=7ILf… via @YouTube

Looking for best answer on X.

Watch Falcon 9 launch the USSF-432 mission from pad 4E in California x.com/i/broadcasts/1kKzDPYrZ…

I told my 9-year-old nephew he could build any game he wanted with Claude. He chose horror. I mostly stayed out of the way. When he got stuck, I'd ask him a question and let him figure it out from there. What he came up with: a haunted house where the book on the nightstand pulls you into the story, five difficulty levels called "Choose your dread," a weapon shop, light switches, and three chapters: the house, a forest, and Mars. And jumpscares. So many jumpscares. 😬😂

Does anyone know why I don't get notifications for all the comments / mentions? I keep finding one's that I never got notified on.

Read the agent's plan before you read its code. A wrong plan makes every line after it wrong too.

Alphabet says search queries are growing at their fastest rate in five years. Search revenue growth slowed to 12% in Q3 from 17% in Q2. Paid click growth went from 13% to 8%. My read for local shops... people are searching more than ever, but more of it is happening in AI Mode, which Google says earns less per query. Paid clicks are growing slower and costing more (CPC +4%). So track calls and booked jobs, not clicks. Make yourself easy for AI to recommend: a complete Google Business Profile, clear service pages, and a website that actually loads.

What does this print? A) 10 10 0 B) 10 null 0 C) 10 10 10 D) undefined null 0

If you had to explain what you do for a living without actually saying your job title, how would you describe it? 😂 Let’s see who can make their job sound the most interesting!

Four AI models give you four different answers to the same technical problem. What do you do first? 🤔

Hurray, 1500 milestone reached and beyond!! Thank you everyone, next target 2000. I know you will all help me to reach it with your kind reshare 💕

I checked 15,040 Charlotte-area business websites between Oct 7 and Nov 3, 2026. 407 of them (2.7%, roughly 1 in 37) greeted visitors with a full-page certificate warning. That means the very first thing a potential customer sees is their browser telling them your site isn’t safe. Most people won’t click past that. They’ll just hit back and call your competitor. The fix is usually quick and often free. Open your site in a private browser window today and make sure it loads cleanly.

Serving several customers from one Claude Agent SDK container? settingSources: [] skips settings files, but auto memory still loads into the system prompt. Set CLAUDE_CODE_DISABLE_AUTO_MEMORY=1 in env too, or one tenant's context can leak into another's.

🎃 THE REAPER'S CIRCLE: ANSWER 🎃 Nobody cracked it. The Reaper wins this year. 💀 Shoutout to @ExactlyJosh, who came the closest. Respect for showing your code. 👏 The last ghost standing is #624 👻 🔍 THE 3 HIDDEN NUMBERS 🟡 Start: the amber ghost was #6 (count clockwise from #1) 🦇 Bats: 11 (one on the moon, one on the prize banner, and two overlapping on the left) 🎃 Pumpkins: 7 🚫 The birds and moths didn't count 🧠 HOW THE SOLUTION WORKS • Put ghosts 1–1031 in a list and track your position and direction • Each round, move ahead (count − 1) spots and pop that ghost • If its number is a multiple of 7, flip direction for good • The next count is 11 + the digit sum of the ghost you just removed • Repeat 1,030 times. The ghost left over is #624 ⚠️ WHERE PEOPLE SLIPPED • The starting ghost counts as "1" • After a pop, the next ghost slides into the same spot • The turnaround lasts until the next curse, not just one step • Use the digit sum, not the ghost's number 🤖 WHY AI STRUGGLED • AI can write this code in seconds. The hard part was the image. • Paste just the text and there are no numbers to work with, so it has to guess. • With the image, it has to count small rotated bats, ignore the decoys, and find the amber ghost's exact spot on the ring. Exact counting and position are still weak spots for vision models. • Miss by one anywhere and you get a completely different ghost. The test example used different numbers, so there was no way to check. Full code is in the image. Thanks to everyone who played! 🪦 #Python #Halloween

Thank you we will win, just need the AI😂 not to feel to badly.

you wanted ai to do your work. now you’re managing a team of chat windows. deepseek explained how to run a team of ai agents without becoming their full-time manager. here’s the article in 6 points: 1. separate coordination from execution. one router talks to you and assigns tasks. workers handle them in parallel, each with its own context. 2. let events trigger work. messages, pr comments and ci results become inputs. code filters the noise before a model decides what needs action. 3. treat context like a cache. keep relevant details close. store the rest outside the window and retrieve them when needed. 4.keep the conversation moving. Long tool calls run in the background. you can change direction while a task is still running. 5. separate the agent from its computer. the agent loop runs in the cloud, state persists separately, and tools can run on your own machines. device-dependent work waits for them to reconnect. 6. check progress, not just uptime. a separate watchdog looks for stuck loops and agents drifting off task. a running process can still be accomplishing nothing. this is a research design with an open-source prototype. not every piece is implemented yet.

GPT-6.1 Terra, build me a 3D racing game in Codex! Dropping off the game made by GPT-6.1 Terra. Comment and like, and I will drop the full game link hosted by ChatGPT in a few moments if you want to try it out!