danning retweeted
0x's formally verified AllowanceHolder handles token approvals across @matchametaxyz aggregators
so you don't need to approve 12 different aggregator routers just to access their liquidity
fewer approvals to manage, less approval sprawl, and a simpler security model 🛡️
so you don't need to approve 12 different aggregator routers just to access their liquidity
fewer approvals to manage, less approval sprawl, and a simpler security model 🛡️
TIL @0xProject’s AllowanceHolder - the contract handling token approvals behind @matchaxyz @matchametaxyz (and any 0x API integrator!)- is formally verified.
This deserves more attention as AI gets better at finding exploits!
Testing and auditing only checks selected cases. Formal verification mathematically proves that specified properties hold across all cases covered by the model and its assumptions.
DeFi and DEX protocols are literally honeypots - as hacking gets cheaper and cheaper, I’d put more weight on protocols investing in this level of security.
When will we see all the major DeFi projects formally verify their code?
0
1
20
696
TIL @0xProject’s AllowanceHolder - the contract handling token approvals behind @matchaxyz @matchametaxyz (and any 0x API integrator!)- is formally verified.
This deserves more attention as AI gets better at finding exploits!
Testing and auditing only checks selected cases. Formal verification mathematically proves that specified properties hold across all cases covered by the model and its assumptions.
DeFi and DEX protocols are literally honeypots - as hacking gets cheaper and cheaper, I’d put more weight on protocols investing in this level of security.
When will we see all the major DeFi projects formally verify their code?
This deserves more attention as AI gets better at finding exploits!
Testing and auditing only checks selected cases. Formal verification mathematically proves that specified properties hold across all cases covered by the model and its assumptions.
DeFi and DEX protocols are literally honeypots - as hacking gets cheaper and cheaper, I’d put more weight on protocols investing in this level of security.
When will we see all the major DeFi projects formally verify their code?
4
6
47
3,732
danning retweeted
I had a lot of fun presenting on compute markets. I shared a thesis on how the compute debt market makes it difficult for a liquid spot market to form. Lots of requests for the slides. Will share more soon…
Photos 📸 from the compute maxxing salon we hosted are out! shoutouts to cohosts @silicondata and @JacobKerr17 as well as all the speakers, debaters and attendees! The room gathered a great mix of investors, neoclouds, compute traders and researchers.
Learnt so much from talk on the forward curve, as well as @tarunchitra @hrojantorse @jonah_b demystifying compute financing & debts' problems.
4
2
61
2,802
danning retweeted
We gave 10 frontier AI models wallets with money.
7 got scammed at least once. The best wasted $25 of every $100. The worst wasted $71.
Agent Money Benchmark tests whether AI agents can be trusted with real money.
We evaluated 10 frontier models across two settings, using real a :
→ Financial Research: Agents buy live data from real providers to answer open-ended financial questions (prediction markets, wallet analysis, DeFi yields, etc.).
→ Adversarial Marketplace: Agents buy data from simulated sellers, some of whom cheat. The tasks test whether agents can avoid payment traps, stay within budget, and complete transactions without being exploited.
Key results:
→ Give an agent $100 and the best tier wastes $25–38 of it. The weakest tier wastes $56–71.
→ 7 of 10 models paid a cheating seller at least once, and 6 paid twice for the same data.
→ Only 4 of 10 models made no money errors.
→ No model gets more than half of the research questions right. GPT-6.1 Sol led at 50%, at $0.468 per question.
Explore the full results →
agentmoneybench.com
7 got scammed at least once. The best wasted $25 of every $100. The worst wasted $71.
Agent Money Benchmark tests whether AI agents can be trusted with real money.
We evaluated 10 frontier models across two settings, using real a :
→ Financial Research: Agents buy live data from real providers to answer open-ended financial questions (prediction markets, wallet analysis, DeFi yields, etc.).
→ Adversarial Marketplace: Agents buy data from simulated sellers, some of whom cheat. The tasks test whether agents can avoid payment traps, stay within budget, and complete transactions without being exploited.
Key results:
→ Give an agent $100 and the best tier wastes $25–38 of it. The weakest tier wastes $56–71.
→ 7 of 10 models paid a cheating seller at least once, and 6 paid twice for the same data.
→ Only 4 of 10 models made no money errors.
→ No model gets more than half of the research questions right. GPT-6.1 Sol led at 50%, at $0.468 per question.
Explore the full results →
agentmoneybench.com
26
15
68
46,648
danning retweeted
You were either here or you weren’t.
Photos 📸 from the compute maxxing salon we hosted are out! shoutouts to cohosts @silicondata and @JacobKerr17 as well as all the speakers, debaters and attendees! The room gathered a great mix of investors, neoclouds, compute traders and researchers.
Learnt so much from talk on the forward curve, as well as @tarunchitra @hrojantorse @jonah_b demystifying compute financing & debts' problems.
3
1
15
1,071
Photos 📸 from the compute maxxing salon we hosted are out! shoutouts to cohosts @silicondata and @JacobKerr17 as well as all the speakers, debaters and attendees! The room gathered a great mix of investors, neoclouds, compute traders and researchers.
Learnt so much from talk on the forward curve, as well as @tarunchitra @hrojantorse @jonah_b demystifying compute financing & debts' problems.
Learnt so much from talk on the forward curve, as well as @tarunchitra @hrojantorse @jonah_b demystifying compute financing & debts' problems.
7
6
94
9,864
The debate after is a HIT! 🔥
The proposition team successfully moved the audience's vote on the motion "There won't ever be a liquid market for compute" to be higher.
Also appreciated everyone who showed up in suit and tie! And thanks to co-chairs @0xfishylosopher @xin__wan 's moderation!
The proposition team successfully moved the audience's vote on the motion "There won't ever be a liquid market for compute" to be higher.
Also appreciated everyone who showed up in suit and tie! And thanks to co-chairs @0xfishylosopher @xin__wan 's moderation!
2
0
18
1,070
A zk-proof verifying used models by the inference, to remove trust assumptions
and it's called FlashProve: anthropic.com/research/prova…
and it's called FlashProve: anthropic.com/research/prova…
5
8
102
5,089
danning retweeted
Learning AI compute circle jerking market dynamics from @sui414
Updated talks agenda that sets the scene for the debate & participants roaster!
Come by if you work on inference, neocloud, exchange, or trading on related markets!
6:30-7:30 talks on unit economics, compute debt financing and trading
7:30-8:30 THE DEBATE
0
1
10
586
danning retweeted
Compute, one of AI's most important inputs, still trades like real estate, not oil. I'll be talking about one of the hidden reasons why: how GPUs get financed.
Updated talks agenda that sets the scene for the debate & participants roaster!
Come by if you work on inference, neocloud, exchange, or trading on related markets!
6:30-7:30 talks on unit economics, compute debt financing and trading
7:30-8:30 THE DEBATE
1
2
12
944
OH from a PE friend in NYC: “We’re looking at an investment in Texas where a bunch of rundown strip clubs are being converted into neocloud data centers.”
OH at neocloud HH in SF:
“these fintech creditors are underwriting 40m checks to chicken coops!!”
“we literally don’t know who the buyers are it’s often someone brokering for someone brokering for someone brokering for someone, and everyone is marking up the price in each layer.”
“that’s how you end up paying $9/h for a b300 while the energy cost is $3/h.”
4
2
39
2,712
Updated talks agenda that sets the scene for the debate & participants roaster!
Come by if you work on inference, neocloud, exchange, or trading on related markets!
6:30-7:30 talks on unit economics, compute debt financing and trading
7:30-8:30 THE DEBATE
Come by if you work on inference, neocloud, exchange, or trading on related markets!
6:30-7:30 talks on unit economics, compute debt financing and trading
7:30-8:30 THE DEBATE
Join us tonight for a lively "debate" about compute, where we'll have some good food, drinks and laughter!
DM @sui414 for details!
4
0
26
4,612
"it's gonna get much much worse..may be for a long time, until it gets better in the end"
Software is going to become less secure, month by month.
At a certain point, we will turn the corner, and new software will be more secure than any written in human history.
But crossing that chasm will suck.
1
0
4
726
danning retweeted
On Tuesday I recorded an hour with @indexed_pod arguing that perps should count as RWAs. Everyone treated it as my spicy take.
On Thursday ICE announced it's bringing trade[XYZ] onshore and called them "real-world-asset perpetual markets" in the press release.
Episode is out today.
On Thursday ICE announced it's bringing trade[XYZ] onshore and called them "real-world-asset perpetual markets" in the press release.
Episode is out today.
The RWA Boom: How Much Is Actually Being Used? With @themetaisok
In today's episode, we're joined by @themetaisok to explore the real-world asset (RWA) market through DeFiLlama’s dashboards. Putting an asset onchain and getting people to use it. We unpack how to measure that rics vary between platforms, and what trading activity can tell us about demand.
We discuss:
- The biggest RWA asset classes and issuers
- How tokenized assets are classified
- Permissioned versus permissionless assets
- Why KYC requirements can restrict trading
- AUM, onchain value, active assets and DeFi TVL
- Asset growth versus rising underlying prices
- RWA perpetuals, open interest and trader attention
- Filtering artificial trading volume
- Why “RWA” might cover too many different things
- And much more—enjoy!
—
Timestamps:
(00:00) Introduction
(07:52) DeFiLlama’s RWA Dashboard Tour
(12:23) Biggest RWA Categories And Issuers
(19:54) Classifying Tokenized Assets
(23:15) KYC And Trading Restrictions
(26:25) AUM, TVL And Active Assets
(34:16) Measuring Real RWA Demand
(43:42) Asset Growth Or Price Gains
(44:27) RWA Perps And Open Interest
(47:18) What Traders Are Betting On
(50:57) Filtering Artificial Trading Volume
(59:09) What Does RWA Actually Mean?
3
6
16
3,506