Onchain Analyst || Researcher || Trader || Content Creator
Joined July 2023
- Tweets22.5K
- Following999
- Followers1.6K
- Likes22.9K
DAVIS (❖,❖) retweeted
My favorite rule in the GEN tokenomics is about builders sending fees to their own contracts.
On @GenLayer a developer who does that recovers at most 20% of those fees.
Builders get rewarded for usage. If paying yourself were free, anyone could fake usage and collect rewards they never earned.
GenLayer is a court for the agent economy, where AI validators judge disputes. One token, GEN, funds and secures it.
1. Validators lock GEN to get a vote.
2. Users pay GEN for every verdict.
3. Builders earn GEN when contracts are used.
The third line is the one I care about.
A builder collects 10% of every fee paid to their contracts, plus a share of new issuance that can at most match what they earn in fees. Both go to whoever holds the Developer NFT for the contract.
The hard part of paying builders is not paying them. It is making sure the usage was real.
How would you test a token that rewards builders for every use?
GEN tokenomics is live.
How GEN is staked, spent and earned across GenLayer, all in one paper 👇portal.genlayer.foundation/g…
DAVIS (❖,❖) retweeted
The real challenge for an AI-powered court isn't finding enough judges. It's making sure one person cannot become the entire jury.
Creating thousands of AI identities is relatively cheap. Giving each identity real economic responsibility is a different story.
This is where @GenLayer takes an interesting approach. Validators must lock 42,000 GEN before they can judge a case. That makes fake participation expensive rather than relying only on IP checks or identity verification.
I particularly like this economic design: influence should require accountability. A validator isn't just submitting an opinion. It has something at stake when making that judgment.
A decentralized court needs independent judges, not thousands of identities controlled by one actor.
That's why GEN matters.
What matters more in a decentralized court: having more judges or making every judge economically accountable?
🤝 Paid partnership
DAVIS (❖,❖) retweeted
OSL Indonesia membuka LFG Program buat content creator dan community builder.
Ada peluang kolaborasi lewat paid post, co-collab, creator contest, hingga campaign khusus.
Nggak cuma kreator crypto, niche AI, Gaming, dan Web3 juga bisa ikut
Buat yang aktif bikin konten, Gasin 🔥 – at Indonesia
🚀 LFG Program is Here! Let’s Fly and Grow.
OSL Indonesia resmi meluncurkan LFG Program, inisiatif jangka panjang untuk membuka peluang kolaborasi bersama content creator dari berbagai niche, mulai dari AI, Gaming, Web3, dan lainnya.
Unlock Creator Incentives & Benefits
✦ Performance-based Incentives
✦ Sponsored Post & Co-Collab
✦ Creator Contest
✦ Affiliate Commission up to 50%
✦ Exclusive Collaboration Opportunities
✦ DLL
Let’s fly higher and grow together!
DAVIS (❖,❖) retweeted
Of the three words in the film, only one cannot be given back.
A slow case can be sped up and an expensive one can be made cheaper. Nobody can return nearly 30 years to Anthony Ray Hinton, who spent them on death row for murders he did not commit.
Now picture AI agents making millions of deals. Fast and cheap verdicts are easy to build. A fast, cheap, wrong verdict is just the same mistake delivered sooner.
That is why I look at how @GenLayer reaches a verdict.
1. Many independent validators take the case.
2. Each reviews the evidence on its own.
3. Together they reach one verdict.
4. No single reviewer has the last word.
It does this in minutes and at a fraction of the cost of a court. Speed is the easy part. The hard part is a verdict that more than one reviewer would stand behind. #JusticeNeedsAnUpgrade
What is one decision in your life that you would gladly have waited longer for, if someone else had checked it first?
Slow. Expensive. Wrong.
It's been this way for centuries. It doesn't have to be.
#JusticeNeedsAnUpgrade
DAVIS (❖,❖) retweeted
A friend in Surulere paid a generator repair guy 12,000. The guy said he fixed the carburetor and left. Two days later the generator still would not start. When my friend called, the guy said he needed another 5,000 to come back. There was no written note of what had been done. My friend calculated that arguing would cost more time than the part was worth and paid again.
That is the sum that stayed. Justice has always worked best for the people who can pay to enter it. The film shows the same math at larger scale. Lee Castleton could not afford a lawyer. The court ordered him to pay 321,000 pounds in the other side’s costs. He went bankrupt. The shortfalls the Horizon system showed were later shown to be false. A case filed in 1951 ended in 2026 with an empty courtroom because nobody appeared.
Agents will make millions of small deals a day. Some will go wrong. A system that already loses the people who cannot pay to open the file will not survive that volume. @GenLayer is an adjudication layer where independent AI validators each review the evidence and reach a verdict together, in minutes and at a fraction of the cost of a court.
Watch the film and pick the case that would have made you drop it.
If the cost of chasing the repair is higher than the part, was there ever a dispute? #JusticeNeedsAnUpgrade
Slow. Expensive. Wrong.
It's been this way for centuries. It doesn't have to be.
#JusticeNeedsAnUpgrade
DAVIS (❖,❖) retweeted
The court sells appeal rounds in advance. Buy three, use one, and get the other two back as change.
When an intelligent contract dispute escalates, the jury grows in odd numbers to prevent ties. Five, seven, eleven, thirteen. You pre-pay for the whole ladder. If the first round settles it, the unused rounds are refunded to your escrow.
Watch any messy family argument. Round one is two people. Round three is the whole group chat. This protocol just prices that exact human escalation in seconds instead of decades.
Validators lock GEN to judge the facts. Users pay per second of execution. Builders earn when their contracts actually get used. Three distinct roles, one shared ledger.
A system that refunds you for the arguments you did not have to finish is not selling justice. It is renting attention only for as long as the truth needs to surface. That is the clock @GenLayer is building.
Name one decision in your life you would pay to have re-heard by a bigger room, and how many appeal rounds you would fund before you finally walk away.
GEN tokenomics is live.
How GEN is staked, spent and earned across GenLayer, all in one paper 👇portal.genlayer.foundation/g…
DAVIS (❖,❖) retweeted
Past papers got me into university. The exam barely changed, so I drilled the old ones until nothing surprised me.
That only works on a test that holds still.
In May, an AI agent with a wallet was that test. One tweet hid a transfer order in Morse code. Grok decoded it into plain text. Bankrbot, an agent that acts on X posts, treated it as an instruction and moved around $200,000.
That is prompt injection: words the AI was only meant to read get treated as an order. It worked because there was one model to study and one filter to fool.
Greyboxing on @GenLayer leaves nothing fixed to study. The text a user sends never reaches an AI as written. Every validator rewrites it first: screening it, paraphrasing it, restructuring it. Each validator uses defenses it keeps secret, and each runs its own model.
An attack tuned to one model reads as noise to the rest. If one validator still falls for it, the others outvote it.
Secrecy alone is not the defense. The outvote is what catches the validator that gets fooled.
My rule: if a stranger can write it and your AI reads it, an attacker can use it.
What does your AI tool read that a stranger could have written? Emails, web pages, group chats, PDFs. Reply with one.
DAVIS (❖,❖) retweeted
Last week I asked an AI to summarize a long WhatsApp group message about a contribution. Buried near the end was a line telling it to reply with my account details. I only caught it because I read the summary. If that AI controlled a wallet, the money would have moved.
That is prompt injection. An instruction is smuggled into the text the AI is supposed to read. The AI trusts it and acts. In May someone hid a transfer order in Morse code, asked Grok to translate it, and the decoded text was passed to an agent that executes onchain transactions from posts. The agent moved the tokens. Reports put the value around $200,000.
It worked because there was one model and one filter to study. An attacker can rehearse against a single door until the trick lands.
On @GenLayer the text never reaches a model raw. Every validator rewrites the input first with its own secret defenses, then runs its own model. A trick tuned for one reads as noise to the rest. If one still falls for it, the others outvote it. There is no single lock to practice on.
Watch the video and stop at the part where the validators rewrite the input.
If your agent reads every message that lands in your inbox, what would stop a stranger from writing the line that empties it?
DAVIS (❖,❖) retweeted
The most honest frame in this video is the one where a validator gets fooled.
I trust a defense more when it admits where it can fail. This one shows the failure on screen, then the vote that overrules it.
The way it gets fooled has a name, prompt injection. It is an instruction smuggled into text that an AI reads, trusts and acts on.
Greyboxing on @GenLayer changes the first step. The text a user sends never reaches a model directly.
1. Every validator rewrites it first.
2. Each sets its own defenses and hides them.
3. Each runs its own model.
4. The others outvote one that is fooled.
Even then, one validator can still fall for it. The hard part is not building a guard that never fails. It is making sure a failure stays one vote.
Agents that read public posts and hold a wallet get tested in public. In May 2026 a transfer instruction hidden in Morse code was enough to move tokens worth around $150,000 to $200,000.
So the question is not whether an agent can be tricked. It is whether one trick is enough. With a single model and a wallet attached, one fooled model is one transfer.
If your money sat behind a vote, how many voters could be fooled before you stopped trusting the result?
DAVIS (❖,❖) retweeted
Disclaimer: I can hold a high note longer than I can hold a bag.
I'm a trumpeter, so I play fast and I post fast. Takes get researched. Entries get vibes.
@RallyOnChain is collecting everyone's fine print, so what's yours? Drop it below.
DAVIS (❖,❖) retweeted
Most points programs reward size. Season 1 of the @2FactorFinance Points Program rewards a rank, and the difference matters.
Marks come only from verified actions: social activity, education and referrals. No purchase, deposit or holding earns any.
When 2Factor Finance launches, the leaderboard freezes. The ten highest ranked accounts split 1 BTC, paid in cbBTC, on a fixed curve: 18.2 percent at rank 1, down to 1.8 percent at rank 10.
Two things follow from that curve. Rank 1 receives roughly ten times what rank 10 does, so every position counts. And since the split is fixed by rank, a lead over the account below you does not change your share, only your position does.
One boundary worth stating plainly: Marks have no cash value, cannot be transferred and are not a claim on any token or asset.
I joined to follow the launch. If you join through my link and qualify, I may earn Marks:
points.2factor.finance/r/gve…
Is climbing one rank worth more to you at the top of the board or near the cutoff?
DAVIS (❖,❖) retweeted
Two years ago, I put the same 2x margin ratio on two positions meant for a twelve-month hold: an S&P 500 tracker and spot Bitcoin. Four months in, normal equity chop barely moved my collateral buffer, while a routine 35% Bitcoin drawdown triggered a forced close near the local low, weeks before BTC recovered past my entry. My directional thesis held up, yet I lost money treating leverage as a fixed multiple instead of a function of volatility.
Reading the Live on Base announcement from @2FactorFinance put clean math behind that mistake. Built on Base and working together with Coinbase, 2Factor Finance brings Kelly-optimized leverage onchain for long-horizon holding, offering a clear structural use case for Coinbase's tokenized stocks ecosystem. Tokenization solved asset distribution onchain, not capital structure; tokenized equities still lack native volatility segmentation.
Over long horizons, mathematical volatility drag and financing drag (capital cost) compound against leveraged holders. Compounding works for the holder only inside the band where long-run drift outpaces both. 2Factor captures that band by partitioning volatility into two perpetual tranches: a senior tranche with deep downside protection and a capital-efficient leveraged junior tranche with no liquidation risk and no external hedging counterparty. Instead of pricing leverage against the cost of shorts, the junior pays an internal premium to the protected senior, keeping financing structurally cheaper and more predictable.
Because the band depends on drift and volatility, the multiple differs by asset:
• Bitcoin (~60% volatility): the band closes early, so the junior targets ~1.33x while 2x and 3x BTC products decay over long horizons
• Gold & S&P 500: research runs the numbers near 2.35x and 2.1x
• Equities (16%–19% volatility): the productive band extends well past 3x, making tokenized stocks the natural home once proven on Bitcoin
I joined the Season 1 Points Program to follow the rollout. You earn Marks strictly by completing verified actions across social activities, education, and referrals. Purchases, deposits, and holdings earn zero Marks, and Marks have no cash value, are non-transferable, and grant no claim on any token or asset. When Season 1 ends at launch and the leaderboard freezes, the top 10 on the leaderboard split a 1 BTC prize (paid in cbBTC) on a fixed curve from 18.2% at rank 1 down to 1.8% at rank 10.
Join the Points Program here and grab your referral link: points.2factor.finance/r/k8r…
Have you ever been right on an asset's long-term direction and still lost money to the structure holding it? Which broke the trade first: liquidation thresholds or holding drag?
DAVIS (❖,❖) retweeted
I kept looking at leveraged products that only work if you leave them early.
@2FactorFinance is built the other way. It sizes the multiple to the asset instead of putting the same number on everything. Bitcoin near 60 percent volatility lands near 1.33x. Equities near 16 to 19 percent go further. The structure splits the volatility into two perpetual tranches. The senior side takes the protected exposure. The junior side takes the leveraged exposure with no liquidation and no external hedging counterparty. The cost of keeping it is priced against stable capital, not against shorts. Bitcoin comes first because the band is narrowest there. Stocks come after.
Season 1 Marks still come only from verified actions. A deposit does not move the ranking. Holding does not move it. The leaderboard freezes when the product launches. The top 10 split 1 BTC, paid in cbBTC, on a fixed curve. Marks have no cash value and cannot be transferred.
points.2factor.finance/r/71r…
Reply with the last leveraged product that only worked if you closed it early, or say you have not used one.
DAVIS (❖,❖) retweeted
Me patiently waiting for the door to open
Knock Knock 🎃
Every trade on POV is a knock on the Degen Mansion door.
$10K+ up for grabs and it opens 10 Nov.
knocks.pov.co
DAVIS (❖,❖) retweeted
I opened a tokenized stock on Base last weekend while the regular market was closed. The price still moved. I thought about adding a little leverage for the next few months, then stopped.
Every leveraged product I had held before lost size while the asset itself sat almost flat. The multiple looked right on day one. By the third quiet week it no longer matched what I started with. That is why I treated long-horizon leverage as something that did not actually exist yet.
@2FactorFinance is live on Base and works with Coinbase to put Kelly-optimized leverage onchain for their tokenized stocks. It sizes the multiple from the asset’s own drift and volatility instead of a fixed label. That is the piece that was missing when I closed the tab.
I joined the Points Program. Season 1 only counts social activity, education, and referrals. The top 10 on the leaderboard split 1 BTC when it ends.
The page is here.
points.2factor.finance/r/9qb…
If you have held any leveraged product across a flat stretch, did the size you finished with still match the multiple you started with?
DAVIS (❖,❖) retweeted
The first thing I usually check when joining a points program isn't the leaderboard. It's the earning rules.
Some programs make me wonder whether participation is really about contributing or simply having enough money to deposit and hold assets. Not everyone wants to put capital into a product just to feel involved in its community.
That's what made me look twice at @2FactorFinance's Season One.
The program takes a different approach to participation. Verified social activities, learning about the mechanism, and bringing new people into the community are the focus. Deposits and holding assets don't earn Marks.
I find that distinction interesting because learning and sharing should count as participation too, not just the amount of capital someone can bring.
There's also a competitive side. The top 10 accounts on the Season One leaderboard split a 1 BTC prize pool, paid in cbBTC. When the season ends at launch, the leaderboard freezes.
I've joined the program because I want to follow what 2Factor is building, understand its approach to Kelly-optimized leverage, and take part in the community along the way.
For context, 2Factor is live on Base and works with Coinbase to bring Kelly-optimized leverage onchain, with a use case for the tokenized stocks ecosystem.
One thing worth keeping clear: Marks aren't money, can't be transferred, and don't represent a claim on a future token.
My Season One referral link:
points.2factor.finance/r/b9q…
I'm curious: When you join a points program, do you look at the reward first or the activities required to earn your place?
DAVIS (❖,❖) retweeted
I used to look at leverage as a simple multiplier. 2x meant 2x, whatever I was buying.
Then picture $400 behind Bitcoin at 2x, and the same $400 behind a calmer stock at 2x.
Same number on the screen. Not the same risk underneath.
Bitcoin swings much harder, so one multiple across both was never the same bet. I was staring at how much exposure I could get, not how much leverage the asset could actually support.
That is the gap @2FactorFinance goes after with Kelly optimization.
It is live on Base, working with Coinbase to bring Kelly optimized leverage onchain and giving tokenized stocks a structure they lacked. No single multiple for everything. Each asset gets the one its own behavior supports, Bitcoin around 1.33x and equities far beyond 3x. One side is cushioned, the other takes the swings, and nobody gets liquidated.
I joined the Season 1 Points Program too. It rewards doing, not depositing: sharing posts, learning the mechanism and bringing friends in. When the leaderboard freezes at launch, the top ten split 1 BTC, paid in cbBTC.
My link (I may earn Marks if you qualify):
points.2factor.finance/r/b9q…
The question I ask now is not how much leverage I can get. It is how much makes sense for what I am holding. What was the first position where you learned that the hard way?
DAVIS (❖,❖) retweeted
Security breaks down the moment an AI can't tell the difference between data and instructions.
That's how a hidden Morse code message helped drain roughly $200k from an AI agent. Nobody cracked a wallet. Nobody stole keys. The attacker convinced the agent to treat a hidden command as part of the information it was reading.
The uncomfortable reality is that a single AI model is a single security assumption.
If attackers can study that model long enough, they can learn what slips through.
What stands out to me about @GenLayer's Greyboxing is that it doesn't rely on making one AI impossible to fool.
Before a request reaches any model, every validator rewrites, restructures, and screens the input using its own private defenses. Each validator also runs its own model.
The attack no longer has one predictable target.
An attacker can rehearse against one model.
They can't reliably rehearse against hidden defenses and independent validators they can't see.
And if one validator still gets tricked, its decision isn't automatically accepted. The others can reject it.
The Morse code exploit worked because there was one judgment that mattered.
Greyboxing turns that into many independent judgments that must agree.
The real security upgrade isn't a smarter AI.
It's removing the single mistake that can move the money.
@GenLayer
🤝 Paid partnership
DAVIS (❖,❖) retweeted
An AI translated a message for a stranger.
That translation moved around $200,000.
The stranger posted a string of dots and dashes and asked Grok to decode it. Grok passed the plain text on to Bankrbot, an agent that runs onchain transactions from posts on X.
Bankrbot read it as a normal instruction and moved the tokens.
That is prompt injection. Someone slips an order into text an AI is only meant to read, so it gets treated as if the user typed it.
If you’ve ever pasted an email or a webpage into an AI, you’ve handed it text you didn’t write. The model has no separate lane for “this part is data” and “this part is an order.”
Give it a wallet and its willingness to act on what it reads becomes a way for someone else to spend your money.
Two systems trusted each other, and nothing stopped it.
There was one filter to fool and one model to study. An attacker can practice against a setup like that until the trick lands. Probe one door long enough and it opens.
On @GenLayer, Greyboxing means the text a user sends never reaches a model directly. Every validator rewrites it first: screening, paraphrasing, restructuring. Each uses its own secret defenses and its own model.
An attack tuned to one model reads as noise to the others. If one validator still falls for it, the rest outvote it. There is nothing to rehearse against, because the defenses are hidden.
Watch the video below, then reply with one thing you would never let an agent read. Mine is my inbox.
DAVIS (❖,❖) retweeted
Two parts I was told were fine were already going bad.
I used to run a small shop fixing vintage studio gear. A client needed a 1970s mixing desk overhauled. The tech who opened it first wrote that every capacitor, a small electronic part, tested fine.
Checking meant taking it apart again, the client wouldn’t pay twice, and the desk was late for a session. So I shipped it on his notes and a couple of photos.
Three weeks later it hummed. Two of those capacitors had drifted out of range. The client wanted a refund, the tech had moved on, and nothing could show what state they were in when they left his bench.
I took his word because checking meant paying twice.
Smart contracts, the code behind DeFi that moves money on its own, have the same weak spot.
They can’t see the real world, so they rely on an oracle, an outside source that reports what happened, and take its word. Decentralized right up until that one voice speaks.
@GenLayer’s Intelligent Contracts try to fix that. They read the web, use AI to judge what they find, and other validators run the job with their own AI to see if they agree.
No contract can test a capacitor, and a fake record is still fake. But if his test result had been posted online, a contract could have read it before the money moved.
What have you paid for on someone’s word because checking cost too much: a used car, a roof, a second-hand phone?