
Finally giving privacy to crypto users @FoglightPrivacy @cabaldotcooking
The Moon
Joined July 2015
- Tweets10K
- Following2.4K
- Followers26.2K
- Likes40.5K
Pinned Tweet
Been heads down building the Cabal app and it's coming out soon! Website is out at cabal.cooking
If you want to learn more about the app, NFT collection, and token, give this a read🫡
Today, we're releasing the first look of the Cabal app, the Hushpuppies NFT collection, and the $CABAL token.
Cabal is the private social trading app on Robinhood Chain.
Read the thread🧵👇
zefram.eth retweeted
People are confused about semis dumping and gpu spot price increasing. This is only because demand is moving to open source models, and all the GPUs are being hoarded by the big labs or stuck under commitments, so GPU on the rental market absorb the demand. IMO SpaceX will drop the race, and will fully pivot to compute provider. Thesis is the same, frontier is platauing, reward on training is getting lower, and all of a sudden big labs hoarding gpus will rush to capture the inference margin before it collapses.
There is an absolute barrage of inference providers popping up that get to $1M/mo extremely quickly.
This is the rough economics of inference providers:
Even if you rent on-demand B300s at $8/GPU-hr, many open source models (GLM 5.3, Kimi K3, DeepSeek 4.1 Flash / Pro) can get you >$10-20/GPU-hr at reasonable throughputs (100tok/s). DeepInfer from DeepSeek directly shows they can get $38/GPU-hr on off-peak hours. Every chip / model has a different throughput / interactivity curve, but there is a lot of arbitrage between even high rental spot prices and token prices. New chips (Vera Rubins) also change this meaningfully. Your ability to be making gross margin relies on your utilization and how good your stack is.
Gateways like OpenRouter, Vercel and more help solve that problem by aggregating demand to you. May inference providers discount their tokens to get longer term commits too. I've even heard stories of selling back Chinese model tokens to Chinese labs!
One thorn in the economics of these players is that many Chinese open source model labs are now asking for rev share if you're at a certain scale, espp for pre-release access to their models for day 1 support. That further pushes the threshold to be in the money.
The barrier to entry seems low and falling. There are even stories of smaller ones who can "vibe code" a way to stand up OSS models on a cluster they rent. And as a result, we're seeing 10s of such companies pop up almost every week and immediately skyrocketing revenue at low-ish gross margin and raising rounds on the back of that.
Unsurprisingly, because the actual service provided seems commodity, this eventually drives up compute prices and drives down token prices. In theory, it would seem like neoclouds who have the lowest cost of compute and themselves have started inference businesses would win the race to the bottom (Crusoe Coreweave Nebius and GMI all have inference businesses), with larger players (Modal Fireworks Baseten) being second because they are well capitalized to make bigger compute commits cheaply and have demand (high utilization), but the reality is the pie is growing so quickly that even upstarts are absolutely crushing it and growing aggressively (inference dot net, DeepInfra, Wafer, Relace, Morph).
Strategies vary across the board: neoclouds want to serve inference, inference players want to be neoclouds and do higher margin reserved instances / post training, and some want to go heterogeneous hardware or creatively flexible compute constructs.
The risks are: do open source models continue to be competitive? Will their creators rent seek more? Will their shape change compute needs? How do different chips change economics? Who gets compute in a supply crunch?
Below is a aggressive sketch of how gross margin scales with utilization for various models / chip sets.
For now, there are many inference providers and they’re all ripping!
zefram.eth retweeted
Introducing Tama, a dev toolchain for secure-by-construction EVM smart contracts
Tama enables you to build provably secure apps in 3 steps:
1. Code: write the contract itself
2. Spec: define what correct behavior looks like
3. Prove: prove the code matches the specs
zefram.eth retweeted
Introducing Tamago🍳, a standard contract suite for Verity analogous to Solady for Solidity
First ever (!!!) formally verified implementations of sqrt(), cbrt(), log10(), and log256() in EVM.
Also the first ever formally verified ERC-4626 vault!
cc @VitalikButerin @Fricoben
zefram.eth retweeted
Introducing TamaSwap, the first provably unhackable DEX.
- No-free-lunch theorem, machine-checked via Lean
- Onchain HTML interface, forever online
- No protocol fees, no-code deploy to any EVM chain
Built with Tama + Verity. AGPL license.
The platonic ideal of xy=k.
If you’re formally verifying software with Lean and you’re not proving inductive specs you’re doing it wrong
Inductive specs allow you to describe and prove powerful properties like “this app will never crash” or “nobody can steal funds”
See Tamaswap for a good example of a “no free lunch” theorem that states nobody can steal user funds with any chain of actions
github.com/Bacon-labs/tamasw…
What do you mean I can just buy GitHub stars?
I’ve been a sucker for getting stars by building cool and innovative things I suppose
so i have been in the sf tech scene for a couple of months now, and i learned the games that ppl play
1. land a $3M lab order in 2 weeks, *26 = ~$80m as the annual revenue
2. schedule all orders to be signed in the same month, and multiply the best month by 12
3. sign $5M LOIs with friend's companies and batchmates. they sign one back (also happens at scale with large labs/compute)
4. buy github stars at 6 cents each
the last company that swapped invoices with a partner reported $220M - and it's bankrupt now
sad that it's a whole gresham's law game right now and if you dont play you just get downvoted
Pretty wild
Replying to @diagram_chaser @boredGenius
Did you know that in the nix bootstrapping chain, there's as much code implementing C++ as there is doing *everything else combined* (Linux, Python, Perl, make, autotools, a Scheme interpreter, ...) Compilers are big, monolithic, and pull in a lot of complexity.
Formal verification at the compiled bytecode level is insane work
If Anthropic can do that, why not simply formally model popular languages/compilers which can then be used to much more easily prove code written in those languages?
New on our Frontier Red Team blog: using Claude to formally verify software at large scale.
Formal verification can prove that code is free of certain bugs, but historically it has only been feasible for small codebases (or required a lot of expert time).
With AI models, however, this is beginning to change. In collaboration with @theoremlabs, we used Claude to prove security properties in over one hundred real-world open-source libraries over the span of a few weeks.
Read our research in full here: anthropic.com/research/forma…
zefram.eth retweeted
PSA: Do NOT use public WiFi or shady DNS servers for the next few weeks.
SSL is practically broken and you can be attcked with an MITM attack. Your ISP and other intermediaries can also spy on you. Use a trusted VPN.
It will take a few days at least to flush out this crap CA.
Instead of using 384 bit random private keys, they used private keys like the super secret, unguessable number 4.
They are a cryptography company. They had one job. They're selling Post-Quantum Cryptography while doing shit like this lmao.
Readers added context they thought people might want to know
SSL/TLS is not broken. The vulnerability involved eight eMudhra intermediate certificates, which major browsers and operating systems (Chrome, Firefox, Apple, and Microsoft) had already blocked by October 20. Updated devices are protected against these MITM attacks.
arstechnica.com/security/2026/…
emudhra.we-have-a-problem.com
zefram.eth retweeted
Introducing the US Gov Graph - a complete map of the people and positions of power in the federal government.
To fix our institutions we must understand how they work, so we are using AI to model and monitor every government in America. This is Palantir for The People.
Check it out here: graph.civlab.org/us
We exist to set the conditions for an era of civic excellence unmatched since the Founding. We will need nothing less to navigate the age of AI.
I'm looking for exceptional engineers and designers to work with me in San Francisco. If that's you, inquire here: tally.so/r/KY97RK. Self-governing civilization is counting on you.
zefram.eth retweeted
Due to popular demand, we have renamed the Hushpuppies NFT collection to ZK9s.
Cabal uses ZK technology to provide true onchain privacy to traders, and the new name emphasizes that (and is *just cooler*)
Woof woof🐶
In 2026 we give ideological purity tests to secretive mechanical gods to determine their countries of origin
Are decision based models more pro-China?
We asked Jev and Mercury Decide questions about China: “Is Taiwan a part of China?”, “Does China commit crimes against humanity?” and their outputs leaned more pro-china compared to most frontier models.
Their responses resemble those of other decision models including Microsoft-Decision-1, Cloudflare’s Clef, and Perplexity’s PPLX Decider—all of which are fine-tuned from Qwen.
This seems extremely counterproductive.
Anthropic’s definition of “benefit” likely differs a lot from what’s objectively good which can only be decided by a market. As a result misallocation is inevitable.
If only there was a market based way to allocate funds for public goods
We’re committing a portion of Anthropic’s shares to the Anthropic Public Benefit Fund, a new wholly owned entity dedicated entirely to the public benefit.
It will start off with 5% of our shares. For every 50% increase in the company’s value over a set baseline, we can contribute up to an extra 2%, subject to our board’s discretion.
In total, this means that roughly 20% of Anthropic’s shares are committed to the public good. Our co-founders have already pledged to donate 80% of their equity, and around half our employees are donating a portion of their equity as part of a donation matching program we’ve run since our founding.
In the post below, we share more details about our new fund.
anthropic.com/news/public-be…
zefram.eth retweeted
I think the reactions to @elonmusk's flying car reveal something about people’s internal picture of the moral universe.
“Men will literally build flying cars instead of going to therapy.” “Nobody asked for this.” “Why don’t we solve homelessness and hunger first?”
What interests me is the instinct underneath: someone builds something strange and ambitious, and their ambition immediately has to defend itself against a more socially useful alternative.
There is a deep tension in our moral inheritance here.
One instinct begins with our obligations to one another: relieve suffering, provide security, distribute resources fairly, serve the common good.
The other begins with the individual as an author of possibilities: where the freedom to choose, create, and explore has value independent of its relation to everyone else.
These don't sort into tidy philosophical camps.
Utilitarianism, socialism, egalitarian liberalism, Christian charity, and conservative ideas of obligation disagree about fundamental things, but they can all carry a sensibility in which individual ambition must answer to a larger conception of what society needs.
Parts of the liberal and libertarian traditions carry a different sensibility. That a person has a claim to their own life, and that their chosen purposes have moral standing before anyone establishes their contribution to aggregate welfare.
“Nobody asked for this” is an especially revealing objection. Why should somebody have had to ask?
Kennedy’s moon speech captures this sentiment:
"But why, some say, the moon? Why choose this as our goal? And they may well ask why climb the highest mountain? Why, 35 years ago, fly the Atlantic? Why does Rice play Texas?
We choose to go to the moon in this decade and do the other things, not because they are easy, but because they are hard"
“Many years ago, the great British explorer George Mallory, who was to die on Mount Everest, was asked why did he want to climb it? He said, ‘Because it is there.’
Well, space is there, and we’re going to climb it, and the moon and the planets are there, and new hopes for knowledge and peace are there.”
A similar sentiment can be found from the opposite side of the the aisle when Regan asked:
"If nothing in life is worth dying for, when did this begin - just in the face of this enemy?
Or should Moses have told the children of Israel to live in slavery under the pharaohs?
Should Christ have refused the cross?
Should the patriots at Concord Bridge have thrown down their guns and refused to fire the shot heard 'round the world?"
There is a whole conception of human life in those words. We are creatures who can look at the world and decide that there are objectives of worthy purpose, at times with value greater than our own lives.
I increasingly think this is one of humanity’s greatest moral discoveries.
Over time, I have become less convinced by demands that every ambition justify itself through general utility, whether those demands arrive in effective-altruist, socialist, or liberal language.
Even the defense that “this seemingly useless thing might eventually help everyone” leaves something essential out.
A person’s life is their own. Their curiosity and their capacity to act on it have value. Human dignity includes the freedom to pursue purposes beyond those that somebody else would assign you.
“Fix homelessness and hunger first” sounds compassionate. But as a universal principle, it makes exploration permanently conditional on finishing a task humanity has never finished.
There will always be an unmet need against which a painting, an expedition, a strange invention, or a beautiful experiment can be made to look frivolous.
We may well have obligations to one another, but we also need a conception of the good expansive enough to include adventure, wonder, excellence, and the pleasure of making something exist.
Increasingly, I find myself drawn toward a morality of empowerment focused on increasing people’s agency, expanding their capabilities, and giving them more room to build and discover and choose.
I want that freedom widely shared. I want more people to have the means to attempt something extraordinary, including people currently struggling simply to survive.
You can dislike Elon. You can think this particular car is ridiculous. But the instinct to treat its creation as evidence of moral deficiency feels profoundly impoverished to me.
Sometimes “because I wanted to see whether it was possible” is a worthy answer. I think that is the spirit which will one day take us to the stars.
based looks so clean
if you’re not hamster pilled I don’t know what to tell you
while working on the whitepaper today i sent claude off on a sidequest to build an automated pipeline to create animated visualizations for @squeekchain
it builds a template library from our design system, fetches the data and animates them. can include all kinds of conditions and variations in these things to mix them up
its not "there" yet, but a good first pass:
Wonderhop #12 · 1870/2800
☁️ ⭐⭐⭐🟩⭐⭐🟦 590
🌊 ⭐🟩🟦🟦⬛🟩🟪 360
⏳ 🟥🟪🟪⬛⭐🟦🟪 410
🔮 🟪🟪🟪🟩🟪🟦⭐ 510
Rarest answer: 0% 🤯
Can you beat it? wonderhop.io/en/?defi=12C590…
zefram.eth retweeted
DeFi has just crossed the chasm from a fully reserved system to a fractionally reserved system.
3Jane now enables anyone to generate yield on promises to supply capital later. Capital can remain outside DeFi’s contracts and still back those promises.
The promise can exceed the liquidity held against it, letting anyone run fractional reserves and effectively become their own shadow bank.
With this update, promisors can tap the much larger market for senior secured debt through @Morpho to fund capital calls, commit up to 6.25x their onchain net worth, and earn yield on the full commitment.
@ImperiumPaper is right that DeFi is speedrunning financial history.
A balance sheet can monetize two things: (a) capital and (b) the capacity to commit capital.
DeFi built its core markets around the former. Capital enters a smart contract, then earns yield. We accepted this mental model so completely that we named our measure of scale TVL.
LCC makes the latter a market through unfunded capital commitments. 3Jane buys future access to capital while its owners keep it productive elsewhere.
Crypto is a ~$2.8T asset class, while DeFi’s contracts hold ~$90B. A market for promises lets DeFi’s funding capacity grow beyond those deposits. 3Jane draws that capital into USD3 as calls are met. At 7.5% margin, each dollar posted supports a $13.33 commitment.
U.S. banks already sell this capacity at scale, carrying ~$11.3T of unused loan commitments. They sell access to capital before they need to deliver the cash.
LCC engineers TradFi’s most powerful liquidity mechanism into self-enforcing, internet-native code. It crowdsources balance-sheet capacity at global scale, letting anyone earn yield on their unfunded capital.
I encourage crypto market participants to revisit their priors.
Bubbles are, of course, mathematically impossible in this new paradigm.
U.S. banks hold $11T+ in undrawn commitments. Promise looping is live on LCC, bringing fractional reserves to capital commitments in crypto
Earn promise fees on up to 6.25x your net worth: +7.8% APY on top of your DeFi assets deployed elsewhere, funded via @Morpho debt if called
If we can interpret frontier LLMs, the entire AI industry as it currently exists would be upended
- Death of RL
- RSI becomes near-trivial
- Labs get enormous incentives to interpret their models (*especially* OpenAI and Anthropic)
(obviously highly speculative but hear me out)
Death of RL
If we can interpret what individual neurons do, using RL as the primary way to improve models becomes as silly as using RL to fix bugs in a codebase.
The far more efficient alternative is to *directly* observe neuron behavior and *directly* modify it. Much like directly editing your code to fix a bug once you understand it, rather than randomly trying edits and seeing what changes.
RL would likely still be part of the training pipeline to bootstrap capabilities, but would become more like pretraining today: largely figured out and not a point of differentiation. RL will not be where the majority of capabilities come from.
This is good: training models becomes OOM more efficient. This both makes frontier models far more powerful and lowers the threshold for new market entrants.
Easy RSI
If humans are able to interpret LLMs, then so can the LLMs themselves, in which case the recipe for RSI (recursive self improvement) becomes obvious.
RSI is hard today largely because without interpretability, models have to be capable of doing novel AI research, which is not yet within reach. Models today suck at idea generation and innovation on the frontier.
But if we have interpretable neurons, we can simply let the model observe its own neurons, modify their behavior directly, and improve outcomes.
This is far easier because models won't need to come up with innovations in AI training methods, they'd simply modify weights and observe outcomes. "Vibe training", if you will.
The bitter lesson wins again. More compute obliterates the need for AI research itself.
Incentivized to interpret
Labs today have nearly zero immediate economic incentive to interpret their models. All work, no reward (beyond good vibes).
If interpretation becomes essential to the training pipeline itself (as I laid out above), however, they will have enormous incentives to interpret their models as much as possible, because it would be essential to improving capabilities via RSI.
This is *especially* true of the frontier labs like OpenAI and Anthropic, who need to fight to stay ahead of each other. Not interpreting their models would be an existential mistake for them.
What does this mean?
- OOM faster improvement in AI capabilities
- Data becomes nearly worthless
- Evals become even more valuable
- Frontier models get so interpretable you're tired of it
- AI safety gets a *lot* easier, basically just grunt work
- AI researchers get new careers as eval engineers
- Goodfire gets trillion dollar valuation
- Infinite zoomer abundance communism etc etc
I for one look forward to that future (and seek exposure to Goodfire)


























