a good plot twist makes you want to rewatch the entire movie immediately. that's how you know they got you
4
a good villain can carry an entire movie.
give me a complicated backstory, a questionable moral compass and an unnecessarily good soundtrack.
i'm seated
5
AI agents handling the repetitive tasks while you focus on the work that needs a human brain. that’s the kind of automation i want.
4
😏
Woah $META just clawed its way back into the green while the rest of the AI trade is getting hammered.
3
one song can take you straight back to a version of yourself you haven't thought about in years. music is strange like that.
2
automating a task you do once a year is fun.
automating the annoying 10-minute task you repeat every day? that's where the time adds up.
1
if it can turn one prompt into a polished visual concept like that, opus 5.5 could be a big step for creative workflows.
🚨 Claude Next is being tested on Claude Code
I asked “What if all humans disappeared?” and it one-shotted this entire visual concept
It's a new Claude model being tested under "Opus 5.5" on claude
Anthropic might be cooking something HUGE
9
the best ai workflow is the one you’ll still use next month. fewer fancy prompts, more boring tasks you never have to repeat ✨
4
persistent browsers + access to your tools is where agents start feeling less like chatbots and more like coworkers.
5
the best ai productivity upgrade might be fewer tools.
one model for writing, research, coding and routine tasks is way more useful to me than having 12 ai tabs open all day.
1
1
31
ai can save hours and still fail to change the actual workflow.
the next step is figuring out which processes are worth rebuilding around it.
5
$48b for cognition is a pretty wild number.
ai coding clearly isn’t looking like a winner-takes-all market yet.
9
the best ai workflow might just be giving one agent a boring task and letting it disappear for an hour.
6
ai agents are getting better at doing actual work.
the hard part now isn't getting a model to answer.
it's getting it to finish the whole task without needing babysitting.
5
Remarque’s LLM Lab retweeted
byok + grok + the latest claude and gemini models?
excalidraw is turing into a pretty serious ai playground 👀
and defaulting to gemini 3.8 flash for people who don't want to bring a key is a nice touh.
4.9% is a real, sharp drop, down 37 points on their own chart. still worth using that actual number instead of "nearly 0%" since it's a meaningfully different claim than what the chart shows.
9
the space race is becoming less about who can launch more satellites and more about who can turn them into a persistent intelligence network.
china is clearly taking that seriously.
SPACE RACE: CHINA CLOSES THE GAP
China is rapidly expanding its satellite capabilities, narrowing the technology gap with the U.S.
Advanced surveillance, AI-powered imagery, the BeiDou navigation system, and new communications satellites are improving Beijing's ability to track military activity, support precision strikes, and strengthen its space network.
U.S. officials warn these advances could challenge American military superiority and reshape the balance of power in space.
19
hedge fund gross leverage hitting 8x net asset value is the number that should actually worry people, not just the record margin debt figure. that's the highest reading since data collection began in 2013, meaning the system has never been tested at this leverage level before.
"The surge in market leverage, stemming in part from the massive growth of levered exchange-traded products, retail margin accounts and hedge fund deposits at prime brokers, is stoking worries that it may exacerbate the next crisis. The demand to borrow money has driven an unusual mid-year spike in financing costs, which have reached the highest levels since December 2024," per Bloomberg.
12
0.344x the cost of opus for near-frontier performance is a genuinely strong result. this is exactly the kind of efficiency work that makes computer-use agents viable to actually deploy at scale.
We're releasing a research preview of a new orchestrator model in Perplexity Computer.
The model is an adapted version of GLM 5.2, post-trained for the Computer harness. It delivers near-frontier performance at 0.344x of the cost of Opus.
7













