dirtydata

@doodlestein when you hear the word "over-engineered" this pops up

So nuts! My longest codex /goal marathon ever: gpt-5.5 xhigh · ~/projects/storage_ballast_helper Goal achieved (5d 13h 40m)

Hermes Agent is evolving FAST. In just the past week, Nous Research added: - A full WebUI/Desktop App - Background Computer Use on macOS - Multi-agent orchestration - Hermes Kanban upgrades - Lightpanda browser backend support - Qwen3.6-Plus FREE in Nous Portal - Better autonomous workflows - Persistent long-term memory systems Hermes is starting to feel less like an AI tool and more like a true open-source Agentic AI Operating System. Full breakdown/demo: youtu.be/Gx2joHxUhgg

@dair_ai I feel like it's a number of factors. #1 the server is out of your control for optimal experience. Best tested on when they're actually giving you compute. #2 It seems like incremental steering along the way or none at all... it's hard to pinpoint exactly when.

Cool paper from PwC. "Earlier is always better" is the default intuition for agent clarification. New paper claims that's mostly wrong. Goal clarification loses nearly all of its value after just 10% of execution. The team built a forced-injection framework that drops ground-truth clarifications at controlled points along a long-horizon agent's trajectory, across 4 information dimensions (goal, input, constraint, context), 3 benchmarks, and 4 frontier models. 84 task variants, 6,000+ runs. Pass@3 falls from 0.78 back to baseline. Input clarification keeps value through roughly 50%. Past mid-trajectory, asking any clarification at all performs worse than never asking. A complementary study of 300 unscripted sessions shows no current frontier model asks within the empirically optimal window. 52% of sessions over-ask. Others never ask at all. Why it matters: clarification has been treated as a binary capability, does the agent ask or not. This is the first quantitative demand curve for *when* the question is worth asking. Paper: arxiv.org/abs/2605.07937 Learn to build effective AI agents in our academy: academy.dair.ai

Today marks the beginning of a new era. Introducing: Cowboy Space Corporation. We are building orbital infrastructure for the AI era: a fully integrated system of rockets and satellites designed to deliver high-performance compute and optical data transmission directly from Low Earth Orbit.

@pretty_sarlin if you use your whole brain you'll see a picasso piece

If you use your right brain, you’ll see a kitten; if you use your left brain, you’ll see a baby rabbit. What do you see?

@steipete i'll take credit for this since one of my replies mentioned doing this awhile back and now here we are XD

Can highly recommend running a claw cron job that sweeps through mentions. GPT is really good at detecting shills and AI reply guy slop.

@steipete @ahmedgagan11 cavemen

@ahmedgagan11 Who's still using Claude?

We are stuck at claude limits and this guy is hitting github rate limits 😅

@Jordan456257099 And this is why you do not do drugs kids... perfect example of how it fries your brain

Best Halo muliplayer to the worst Halo muliplayer overall 1: Halo 3 2: Halo Reach 3: Halo 2 4: Halo 5 5: Halo Infinite 6: Halo CE 7: Halo 4

@0xRemedy what a crackhead

Streamer 9trevv calls out Chudthebuilder for being a "pussy" after her stole his hat on stream 🤯 "You go up to innocent people and harass them but when things don't go your way you call the cops like a little girl."

@WileyCoyoteGG you're just bad

It’s a hot take but beat downs gotta go. I genuinely think they are the most troll shit in Halo. They are also so broken now

@FlameKaizerX english dubs over subs weebs

the pause between “ban” and “kai” is everything. but bleach eng dub ruins this. it straight up says “bankai”.

Name a video game that you've easily put 1,000 hours into Gifs only

@NotWeb3liveNews @9trevv @ChudTheBuilder wonder why he's missing that tooth... hmmm

🚨 BREAKING: @9trevv allegedly snatches @ChudTheBuilder the Builder’s hat LIVE on #PumpFun stream 😭 🎩 Absolute chaos breaking out in the trenches right now as viewers watch the hat heist unfold in real time. Clip below of live stream mins ago

@dhruvtwt_ sad to see so many people who claim to be smart but cant seem to figure it out XD

Unpopular opinion: People telling everyone to switch from Claude Code to Codex right now will be the same people telling everyone to switch back from Codex to Claude Code again in a few weeks.

@dhruvtwt_ rage bait... you're lost

@SStricklandMMA next time just make fun of his voice, GG's you're in his head

Exactly what I expected a coward to do.

@ChampRDS just make fun of his voice and GG's you're in his head

@VBarsoum @karpathy juggling 10, 20, 30, 40, heck maybe 50 unfinished projects sucks... could be more though honestly XD

hey @karpathy i think we need a new term: Chill Coding. It's vibe coding just one app at a time, and youtube or netflix is on.

@dhruvtwt_ yes, codex/codex-app are clearly better models but the harness sucks usage so fast it's not worth it

codex has gotten insanely polished over the past few weeks and gpt-5.5 just get things done wild that people still think claude code with opus-4.7 is better

@tonysimons_ feels bad when you're not aware of the best hack for infinite usage XD

My Codex quota when using GPT 5.5 with xHigh reasoning...