Mono no aware wa kanashimi no hana.
Kanagawa-ken, Japan
Joined December 2025
- Tweets11.3K
- Following715
- Followers946
- Likes22.5K
Ayaka retweeted
Not sure what they’re doing, but it’s fun to watch.
@grok, do you know?
Ayaka retweeted
fungo's $fungo airdrop is here, and it's good to see early community participation getting its own place in the distribution. @FungoLabs
10b total supply, with allocations announced for:)
> genesis nft holders
> zama og community
> x campaign
what caught my attention is the 36% allocation for genesis nft holders.
3,232 nfts, with 1,113,861 $fungo per nft. equal allocation, no tiers.
there's also 7% for zama og and 3% for the x campaign.
a distribution that gives early supporters and community participation a meaningful role.
i'm looking forward to seeing how the next phase unfolds.
good to see the fungo community moving forward.
$fungo 🟡
Ayaka retweeted
The market may be getting closer to a turning point, and $DINKY is here for the journey. 🐥
While everyone waits for the next big move, DINKY keeps building its identity around a growing community and a vision that goes beyond short-term noise. Small duckling, big ambitions. Sometimes the strongest stories start small.
CA: 3YNCESdqhSgvsD4aARHC4rdqXpjyW9CWjosH82hNpump
🤖 Made with AI
Ayaka retweeted
Nobody books a surgeon for a splinter. Medicine has urgent care, pharmacies and nurses for the small stuff, so the surgeon is saved for what actually needs one.
Justice never built that middle layer. Lee Castleton was blamed for shortfalls a computer made up, could not afford a lawyer, and was ordered to pay £321,000 in the Post Office's costs. For a mistake that started as a glitch, the only room available was the operating theatre, and he got a surgeon's bill.
Now AI agents are about to make millions of deals. Most of their disagreements will be splinters. They still need someone to look at them quickly, for a price that makes sense.
That is why @GenLayer caught my attention. Many independent AI validators each review the evidence and reach a verdict together, in minutes, at a fraction of a court's cost. It is the urgent care that justice never had.
The film closes on the idea that justice has always worked best for those who can pay. So tell me, which disputes would you hand to urgent care, and which would you still want a surgeon for? Give me an example. I will answer every one.
#JusticeNeedsAnUpgrade
Slow. Expensive. Wrong.
It's been this way for centuries. It doesn't have to be.
#JusticeNeedsAnUpgrade
Ayaka retweeted
My crypto auto-sniper has a perfect record. It has never lost a dollar, because it has never bought anything.
I built it to detect newly launched pump.fun coins and buy them. Detection worked. Buying did not. The free PumpPortal feed arrived about a second late, shared RPCs kept timing out or rate-limiting me, and some tokens had already graduated by the time the bot tried to buy. My bankroll was around $5. I lost $0.
Looking back, the bot had one feed and one RPC. One slow door decided everything.
An AI agent with a wallet has the same shape, with higher stakes. In May 2026, someone hid a transfer instruction in Morse code and asked Grok to translate it. Grok passed the decoded text to Bankrbot, an agent that executes onchain transactions from posts on X, and around $150,000 to $200,000 in tokens moved. That is prompt injection: an instruction hidden in something the AI was only meant to read. With one filter and one model, the attacker could rehearse until it worked.
@GenLayer's Greyboxing takes away the single door. Every validator rewrites the input before any model sees it, using defenses it keeps secret, then runs its own model. A trick tuned to one model reads as noise to the rest, and one fooled validator gets outvoted.
If you have built a bot that acts on a public feed, what is its one door?
Ayaka retweeted
Hey @grok, assign one animal to each of these world leaders based on their personality. Don't be diplomatic.
Ayaka retweeted
Hey @grok, what really happened to China's 4,000-year-old lost village? Uncover the mystery behind its destruction, the ancient mother and child, and the discovery of the world's oldest noodles! 👀
Ayaka retweeted
Every spam filter you have ever beaten taught you a rule you already know: whatever you can read, you can get around.
AI agents with wallets have not learned it.
In May 2026, someone hid a transfer order in Morse code and asked Grok to translate it. Grok did, and handed the plain text to Bankrbot, a bot that executes transactions from X posts. Around $200,000 in tokens left. That is prompt injection: an instruction hidden in text the AI only meant to read.
It worked because the attacker was effectively reading the defender. One filter to describe. One model to test against. Nothing in the path that could disagree.
Greyboxing puts the disagreement in the path. On @GenLayer no input reaches a model raw. Every validator rewrites it under its own secret rules and runs its own model, and a decision only counts when they agree.
This is the part I keep turning over. Secret cuts both ways. Defenses hidden from the attacker are hidden from you. You cannot see why your message came back rewritten, or check that five validators are five rather than one model counted five times. Agreement and shared blindness look identical from outside.
That is the price of a lock nobody can study: you cannot study it either. I think the trade is right, because the cheap alternative already cost someone $200,000. But it is a trade.
So which do you want to hold: a defense you can inspect and an attacker can inspect, or one neither of you can read? Say which, and what you give up for it.
Ayaka retweeted
team is using the remaining 85% of new creator fees to buy back $SGT, and everyone can track the buys on @sight_hood
SightEVM testing is also underway, with Testnet expected in about a month.
A DEX and cross chain bridge are in development too.
Ayaka retweeted
Holy shit, This idiot has 47 teeth (denticles), yet he drinks blood to feed himself and gets killed by his own prey... LMAOO 😂
Ayaka retweeted
Not every creator opportunity needs a big investment to get started.
@FungoLabs is giving creators a chance to earn a share of its ecosystem through content.
300M $FUNGO tokens (3% of the total supply) are allocated for creators.
Getting involved is simple:
→ Connect your X account at fungolabs.org/launchpad
→ Create original content about Fungo Labs
→ Submit your 5 best posts
The leaderboard comes after the campaign ends, with rewards distributed directly to wallets at TGE.
No complicated process. Just bring your ideas, share your perspective, and let your content speak.
Might be worth exploring. 🍄
🟨 $FUNGO X Campaign is live!
Reward pool: 300,000,000 $FUNGO
→ Connect your X on Fungo Campaign
→ Submit your 5 best Fungo posts
Posts from 2026-11-01 14:00 UTC onward count.
Leaderboard revealed when the campaign ends. Rewards hit your wallet at TGE.
fungolabs.org/launchpad/c/cf…
Ayaka retweeted
Something interesting is happening around fast matrix multiplication.
@WeinsteinOmri has been exploring a question at the centre of the FMM debate: why do algorithms with better theoretical complexity often struggle to outperform conventional matrix multiplication on actual hardware?
His work with Kevin Pratt and Yahel Uffenheimer explores FFT-based convolutions as an alternative to traditional Strassen-style recursion.
teapawt.net/WeinsteinOmri/status/2…
Now @PassRafael has presented a constructive proof showing that the matrix multiplication exponent satisfies ω ≤ 9/4, using polynomial packings to explicitly construct the underlying algebraic identity.
teapawt.net/PassRafael/status/2117…
These are different approaches, but they point towards the same broader problem: making matrix multiplication cheaper.
And there are some important questions worth asking:
If we can push the exponent down to 2.25, what still prevents that improvement from translating into practical GPU performance?
Could FFT-based approaches offer a more practical direction than recursive algorithms, or will memory movement and communication remain the bigger bottlenecks?
At what matrix size does asymptotic superiority become an actual engineering advantage?
And what happens when matrix multiplication is not just an AI workload, but also the computational foundation of a proof-of-useful-work protocol?
This is where @prlnet makes the discussion particularly interesting.
pearl-2:native is building around useful computation, with matrix multiplication serving as a workload for its cuPOW approach.
Its ability to adapt to improvements in the algorithms used to perform that computation is an important part of the bigger picture.
A faster algorithm changes the cost of doing the work. That has implications for the economics of useful compute, verification and the assumptions behind the protocol.
Could proof-of-useful-work ecosystems like Pearl become one of the places where advances in computational mathematics translate into real economic value?
That's a much bigger conversation than simply finding a faster way to multiply matrices.
Ayaka retweeted
My friend asked what the token actually did. I couldn't answer, so we ordered more coffee.
I was explaining a crypto project when I realized I was repeating its website back to him. My explanation came down to holding a token to earn more tokens.
He wanted to know what purpose it served. I genuinely didn't know.
That conversation came back to me when I started reading about @GenLayer's GEN tokenomics. This time, I wanted to understand where the token fits when someone uses the network.
Here's how I explained it to myself:
• Validators stake GEN to participate in judging cases.
• Users pay GEN for execution and AI verification.
• Developers can earn a share of fees when people use their Intelligent Contracts.
The builder side is what made me stop skimming.
A Developer NFT carries the right to those contract fees and can be held or transferred. The potential earnings aren't a one-off grant for launching something. They're tied to people continuing to use what the developer built.
That doesn't guarantee usage or income. But it gives me a better question to ask when I'm trying to understand a token: where does it fit between someone building something and someone using it?
I think I'd have more to discuss with my friend over coffee this time.
If you had to explain GEN without mentioning its price, which role would you start with: validator staking, paying for execution, or builder fees? Why?
GEN tokenomics is live.
How GEN is staked, spent and earned across GenLayer, all in one paper 👇portal.genlayer.foundation/g…