Vroon retweeted
Rule induction at 0.6% pooled on Visual Cipher Bench.
The agent gets up to three drawings and an hour with tools and never finds out if it's on the right track, which is a lot closer to real work than the single-question visual reasoning evals I usually end up reading.
🤝 Paid partnership
Coding agents can now grind through long-horizon tasks, as long as they get a written spec and a test suite.
What happens if the spec is a drawing, and nothing tells if they're right?
We gave 8 frontier models 40 visual puzzles. The best configuration solved 26.7%. Three models solved none.
Introducing Visual Cipher Bench from Einsia AI, a benchmark for long-horizon visual reasoning. Agents get tools, then after 60 minutes, no written rules and no correctness feedback is given.
40 original puzzles · 67 drawings · 8 models · 29 configurations · 3,480 graded attempts
🌐Project: lab.einsia.ai/visual-cipher-…
💻Github: github.com/Einsia/Visual-Cip…
📄Paper: arxiv.org/abs/2611.16861
Vroon retweeted
Fresh drop for the AI community.
Karpo-Decision-31B is now the top-ranked model on the Decision Index.
• #1 out of 300 models
• Better than Gemini Decision 1 on agent decisions
• Calibrated confidence scores
• Free weights on Hugging Face
Grab it 🎁
karpo.ai/models/decision
🤝 Paid partnership
Picking the next tool shouldn’t take a paragraph.
We built Karpo Decision. A family of six open-weight models, from 0.8B to 31B.Give it the context, a question, and the options. It returns a probability for each option in one forward pass.
Karpo Decision 31B
🏆 73.9 on Decision Index, tied for #1 on the Nov 12 board
⚡ 94 ms median per request on a single RTX PRO 6000
📊 87.2 on JevBench Capability vs. 78.5 for Google’s Gemini Decision 1
🤗 Hugging Face
huggingface.co/MachinePulseA…
📝 Technical write-up
karpo.ai/blog/karpo-decision…
Here’s where I’d try it first: Search. Recommendations. Tool choice. Model routing. Browser actions. Point it at the slowest decision in your stack and tell us what breaks.
Vroon retweeted
Building AI agents often means managing separate models and services for every task.
Search, retrieval, OCR, structured outputs, classification, translation, vision and more.
Superlinked's SIE brings all these capabilities together in one self-hosted inference engine.
It supports 100+ models, loads them on demand, and offers an OpenAI-compatible API.
One cluster. Multiple AI workloads. Less infrastructure complexity.
Explore SIE: github.com/superlinked/sie
Vroon retweeted
Hey @grok , have the handsome guy in the left image strike the pose in the right image💅🏻
Vroon retweeted
Ranking the most beautiful actresses based on their overall contribution towards humanity
Vroon retweeted
Some ads make you remember the discount. Others make you remember the product. The best ones give you a reason to remember both.
For your next Black Friday project, eligible creators can get 90% OFF their first month of CapCut Pro with a monthly subscription. This desktop-only offer applies to eligible new users, users with expiring subscriptions, and former subscribers whose subscriptions ended more than 30 days ago.
The first month costs $3.99 or less. Some new users may also qualify for a free trial. Pricing varies by region.
A memorable campaign needs more than one repeated product shot. Seedance 2.5 4K in CapCut PC generates additional 4K AI video clips through AI Video in the multi-track timeline or Video Studio. It's available in select regions outside the US, with the US rollout coming soon.
Then, CapCut Agent EditPilot helps refine the timeline through natural-language instructions, making it easier to adjust the order and flow as your creative takes shape.
Give your campaign more visual moments to work with:
tinyurl.com/2v23rn22
#CapCutMarketing #CapCutPro90Off #CapCutPC #ChatwithCapCutAgent #CapCutDidThat #CapCutSeedance254k
🤝 Paid partnership · 🤖 Made with AI
Vroon retweeted
BREAKING🚨: These 8 careers will quietly dominate the next 10 years.
Most people won’t notice until it’s too late.
The people learning these skills today will be impossible to ignore by 2030:
Vroon retweeted
Please @grok this image isn't clear enough, make it realistic !
see grok results….🫦
Vroon retweeted
Hey @grok transform everyone in this picture into modern-day people.
Vroon retweeted
A boy becomes a man when he understands what girls go through during their periods
@grok is this true ??
Vroon retweeted
They were putting metal instruments inside women's vaginas in the name of MEDICINE?! 💀
Hey @grok , what exactly is happening in this historical photograph?
Vroon retweeted
Typing on a bigger screen isn't always convenient.
Typeless makes it easier with voice-first writing, now available on iPhone Duo.
Just speak naturally, and let AI turn your thoughts into clear, polished text.
Less typing. More creating. Any screen.
🤝 Paid partnership
Good morning, gVANGRID
Artificial intelligence systems are capable of ingesting and analyzing millions of visual assets every single second, but can they genuinely keep pace with urban environments and streets that shift and evolve daily?
That is precisely the complex hurdle @vangrid_io is actively solving. By turning standard, everyday smartphones into a powerful decentralized network, it enables continuous, crowdsourced capture of real-world locations in immersive 3D.
Instead of depending on costly, centralized mapping fleets, nearby community contributors simply record targeted locations whenever fresh, up-to-date data is required on the ground.
Privacy is also baked directly into the core architecture: sensitive details like human faces and vehicle license plates are automatically blurred on-device prior to any upload.
This is precisely where the DePIN ecosystem gets truly compelling: transforming practical, real-world data collection into an efficient, community-driven model.
The crucial test moving forward is whether this continuous stream of data can successfully remain accurate, verifiable, and commercially valuable at global scale.
Ultimately, advanced physical AI requires far more than raw data feeds alone.
It demands a living, trustworthy, and precise picture of our changing world.
looping used to be a whole afternoon of clicking
deposit, borrow, swap, deposit again, repeat until the gas bill looks like rent
most people quit after round two
@tori_finance just put the whole thing in one tab
quick refresher
$strUSD is tori's staked dollar
it earns yield from market neutral strategies and that yield shows up as $strUSD slowly getting worth more $trUSD over time
a loop uses your $strUSD as collateral, borrows dollars against it, buys more strUSD with them and repeats.
you end up earning that yield on a bigger number than you put in
in the app you pick a market first
@Morpho is on ethereum, monad and pharos and @Curvance is on monad
then you enter your usdc and drag a leverage slider
one transaction does the rest
a flash loan fronts the extra, everything gets swapped into $strUSD and posted as collateral
then a regular loan against it repays the flash loan
put in $1,000 at 3x and you hold roughly $3,000 of strUSD with about $2,000 borrowed, before fees
managing it is the same tab
you can add, withdraw, lower leverage or close
closing sells the strUSD, repays the loan and sends what's left back to your wallet
one thing before you touch it
a loop only pays while strUSD's yield is higher than your borrow rate and both move every day
crank leverage too high and a small move can get you liquidated
keep an eye on the health factor
🤝 Paid partnership
Vroon retweeted
GM, flock! ☀️🦆
Woke up with my head tilted, feathers looking good and a whole lot of life to enjoy. 🙃
Let’s make today a good one! 🧡
GM CT 🦆☀️
New day, same conviction. Dinky keeps moving, the community keeps growing, and the best chapters are still ahead.
Stay patient, stay focused, and keep believing in the little duck with a big story.
$DINKY is just getting started.
CA: 3YNCESdqhSgvsD4aARHC4rdqXpjyW9CWjosH82nH8pump
Vroon retweeted
GM CT!
One thing I find interesting about @sai_superai is that it gives AI creativity a competitive playground.
Think about it. Today, anyone can generate an AI image or video, but getting people to notice it is another challenge entirely.
That's the gap @NEXST_AI is exploring.
With $SAI, creators can publish AI-generated content, while the community gets to predict which creations will attract the most attention.
Three things stand out to me:
- CREATION: AI-generated content becomes the raw material.
- COMPETITION: Creations face off in prediction games.
- PARTICIPATION: Humans bring their judgment into deciding the winners.
I like this direction because it makes AI more interactive instead of another tool people open, use and forget.
The bigger opportunity is turning AI entertainment into a social experience people have a reason to return to.
AI CREATES. HUMANS DECIDE. THE GAME CONTINUES.
🤝 Paid partnership





































