REGULATION EVERYWHERE | NEW MICROSOFT COPILOT | COSIGN nitter.net/i/broadcasts/1nGnRBZeD…
8
6
67
79,419
Crucible Capital founder @Melt_Dem reveals how she financed her own GPU cluster with stablecoin debt and personal credit risk because she refused to sit out the AI boom:
"We're a no-name operator. They're like, who are these two women building compute? Their perception's like, who's this crypto lady?"
"We got debt from USDAI, which is a protocol that crowdsources capital through stablecoins, and the GPUs back the loans. We also took on some personal credit risk. I think being levered to the tits is a great way to live."
"In the greatest build-out of infrastructure in human history, are you really just gonna sit on the sidelines and tweet? I'm gonna ship some bangers, and I'm gonna take some risks, and if I die, I die."
"I'm not gonna sit there and tell my children, during the AI boom I sat back and I tweeted."
@CrucibleCap
14
9
120
26,760
Crucible Capital founder @Melt_Dem says compute will be this century’s oil, and America should win the same way it built the petrodollar: financial engineering.
"I am a compute maximalist. The core view we have is a view I call capital markets maximalism. America is incredible at financial engineering"
"This is why the US dollar is the petrodollar. If the oil industry was the most consequential of the last century, I think the compute economy will be the most consequential of this century. How does America win? Shit ton of financial engineering. Let's do what we're really good at."
"If banks are stepping back, if credit funds get more rigorous with their underwriting, we see the whole debt complex repricing from 7 to 9% into the low teens. That means we need new sources of capital to step in."
@CrucibleCap
3
33
12,654
micro1 CEO @aliansarinik explains why blacking out PII destroys the relationships AI needs to learn, and how FlowTransformer preserves them in a synthetic digital twin:
"Every ingredient within the environment, we aim to make more realistic because it matches the distribution that models need to learn on. That's the distribution they're gonna act in whenever they're deployed into enterprises."
"We've been partnering with hundreds of companies, licensing their data, anonymizing it, and then using it for training"
"The default is you just redact all the PII, you draw a black box around it. The problem is you lose the consistency of the identities and the relationships that allow you to train."
"Instead of redaction, it does transformation. It creates a digital twin of any given enterprise. It changes all the names and identities into synthetic versions, keeps the relationships intact, but keeps privacy as the core."
@micro1_ai
Today we’re launching micro1’s PII transformation model, flow-transform 1.0, delivering frontier-level performance across detection, identity synthesis, and transformation of personally identifiable information.
On PrivacyBench, our model reaches 96.0% F1, outperforming every detection baseline we tested, including Tonic Textual, Claude Opus 4.8, Sonnet 4.6, Microsoft Presidio, Haiku 4.5 and GLiNER2.
Some of the most valuable training data for frontier AI models lives inside fully functioning companies. It captures years of real work across decisions, communications, tools, handoffs, exceptions and the relationships connecting them.
The problem is that this data is also full of PII.
Traditional redaction makes the data safe, but it also destroys the very workflows and relationships frontier models need to learn from.
flow-transform 1.0 solves this by turning enterprise operational data into high-fidelity training data for frontier models by replacing real-world identities without flattening the reality the data captures.
4
2
73
18,986
micro1 CEO @aliansarinik on where synthetic data belongs in an expert data pipeline: everywhere the expert’s craft isn’t.
"It's a bit counterintuitive because we're talking about realism, but then we also wanna use data generation models."
"A lawyer's expertise that gets distilled into model training is the golden response of a legal memo. But the training signal comes from rubrics. The lawyer would also create 30 to 40 rubric items that define what a perfect legal memo looks like."
"The lawyer writing these rubrics is a bit of a waste of time. They shouldn't spend their time writing these binary rubrics."
"That tedious part the model can do, or soon it will be able to do, and the lawyer just focuses on building the artifact that is their craft."
@micro1_ai
Today we’re launching micro1’s PII transformation model, flow-transform 1.0, delivering frontier-level performance across detection, identity synthesis, and transformation of personally identifiable information.
On PrivacyBench, our model reaches 96.0% F1, outperforming every detection baseline we tested, including Tonic Textual, Claude Opus 4.8, Sonnet 4.6, Microsoft Presidio, Haiku 4.5 and GLiNER2.
Some of the most valuable training data for frontier AI models lives inside fully functioning companies. It captures years of real work across decisions, communications, tools, handoffs, exceptions and the relationships connecting them.
The problem is that this data is also full of PII.
Traditional redaction makes the data safe, but it also destroys the very workflows and relationships frontier models need to learn from.
flow-transform 1.0 solves this by turning enterprise operational data into high-fidelity training data for frontier models by replacing real-world identities without flattening the reality the data captures.
4
49
11,003
micro1 CEO @aliansarinik on how you verify writing and poetry when the debate mechanism that works for law breaks down:
"The perhaps overused word, which is taste. It refers to this idea of subjective verifiers. Coding is maybe a little bit of an exception, but medical, finance, legal, those domains are also very subjective."
"A lawyer creates 30 rubric items, they send it to a peer reviewer, which disagrees on a couple, they debate it out, and then we come to the truth."
"You don't come to the truth in arts, because there is no truth. Everyone has their own opinions and preferences. More debate results in a divergence of opinion, which is why this is very, very hard."
"It's still a very open research problem. One of the solutions is preference labeling."
@micro1_ai
Today we’re launching micro1’s PII transformation model, flow-transform 1.0, delivering frontier-level performance across detection, identity synthesis, and transformation of personally identifiable information.
On PrivacyBench, our model reaches 96.0% F1, outperforming every detection baseline we tested, including Tonic Textual, Claude Opus 4.8, Sonnet 4.6, Microsoft Presidio, Haiku 4.5 and GLiNER2.
Some of the most valuable training data for frontier AI models lives inside fully functioning companies. It captures years of real work across decisions, communications, tools, handoffs, exceptions and the relationships connecting them.
The problem is that this data is also full of PII.
Traditional redaction makes the data safe, but it also destroys the very workflows and relationships frontier models need to learn from.
flow-transform 1.0 solves this by turning enterprise operational data into high-fidelity training data for frontier models by replacing real-world identities without flattening the reality the data captures.
1
2
35
8,944
micro1 CEO @aliansarinik says recursive self-improvement won’t kill the data business, it’ll make data pipelines so efficient that labs can expand into far more domains:
"RSI is a geometric progression versus a binary one where you say, okay, now we've achieved RSI. We've already been on that curve. When coding is largely done by models, it's already a huge aspect of models improving themselves."
"There's never a 100% state. What that means is you essentially make data pipelines a lot more efficient."
"Data pipelines become so efficient that the number of pipelines will increase, which means there's so many more domains that labs can improve on, and not just medical, finance, coding, legal."
@micro1_ai
Today we’re launching micro1’s PII transformation model, flow-transform 1.0, delivering frontier-level performance across detection, identity synthesis, and transformation of personally identifiable information.
On PrivacyBench, our model reaches 96.0% F1, outperforming every detection baseline we tested, including Tonic Textual, Claude Opus 4.8, Sonnet 4.6, Microsoft Presidio, Haiku 4.5 and GLiNER2.
Some of the most valuable training data for frontier AI models lives inside fully functioning companies. It captures years of real work across decisions, communications, tools, handoffs, exceptions and the relationships connecting them.
The problem is that this data is also full of PII.
Traditional redaction makes the data safe, but it also destroys the very workflows and relationships frontier models need to learn from.
flow-transform 1.0 solves this by turning enterprise operational data into high-fidelity training data for frontier models by replacing real-world identities without flattening the reality the data captures.
1
2
63
13,237
a16z’s @david__booth says venture hits diminishing returns when you keep adding partners, because the real compounding asset is the network:
"Many firms will scale by adding more partners, more head count, and there's a law of diminishing returns that kicks in. This firm delivers value because it's a network-driven firm."
"All the software to date has been very internal facing. This was the first time we've done a drop of this nature in public."
"Context capture is no longer a technology problem. Anybody can scrape the internet. It's actually an incentive problem. What can you build that people want to give value to because they get value from?"
@cosign @a16z
2
17
7,996
a16z’s @david__booth and @katiesquestions on what makes a Cosign different: you’re putting your own reputation behind someone else.
David Booth: "LinkedIn is where you go to talk about yourself and the work you've done. Cosign is where your profile gets built out by the people who you have worked with."
"A cosign is intended to be scarce and expensive. You can go on LinkedIn and connect with anyone you just met at the party. A cosign is public. You're attaching some of your reputation to them."
Katie Goldstein: "This is the kind of information I would normally need to get privately. I'd have to call references."
David Booth: "We wanna make that private signal explicit. If this person left their company tomorrow, I'd totally fund them. Let's let her know that."
@cosign @a16z
Excited to introduce @Cosign: the curated professional network for the startup community: cosign.co
Our goal is to create the following:
- A comprehensive startup directory of investors, companies, and operators, including what they worked on and who they worked on it with.
- Cosign graph: Who shaped your career? Who were you actually in the trenches with? Who would work with you again? Who thinks you’re someone to watch?
- Durable reputation: Great endorsements happen every day on X and disappear into the feed. Cosign attaches those signals permanently to people and companies.
- Discovery: Who are the best fintech angels? Which AI companies should I watch? Who are the best designers?
- Intent network: Privately signal “I’d invest in them,” “I’d hire them,” “I’d work with them,” or eventually even “I’d acquire this company.” Match people when interest exists on both sides.
Imagine if Wikipedia, LinkedIn, and OG AngelList had a baby. Though Cosign is a community, not a business.
In order to join, you need to be cosigned. Or you can apply directly.
David Booth and I started this idea 7 years ago but didn’t have the firepower to make it work. Now we do.
No one has really touched LinkedIn in 20 years. Excited to take a swing.
cosign.co
5
5
27
17,537
a16z's @david__booth and @katiesquestions on how Cosign is different from LinkedIn:
Katie Goldstein: "It's LinkedIn plus AngelList and Wikipedia all had a baby. It's this reputation layer that sits over it that is really the most meaningful information."
"I curate my own LinkedIn, and let's be honest, I try to make myself sound really impressive. But more than anything, I wanna know what I think about other people and also what other people think of me."
David Booth: "I don't think the connection strength represents anything anymore. A professional reputation is really about the strength of what relationships you have, and that's simply not represented there at all."
"People who are focused on identifying potential before the market. What is your CV that demonstrates your ability to do that? It's certainly not LinkedIn."
@cosign @a16z
Excited to introduce @Cosign: the curated professional network for the startup community: cosign.co
Our goal is to create the following:
- A comprehensive startup directory of investors, companies, and operators, including what they worked on and who they worked on it with.
- Cosign graph: Who shaped your career? Who were you actually in the trenches with? Who would work with you again? Who thinks you’re someone to watch?
- Durable reputation: Great endorsements happen every day on X and disappear into the feed. Cosign attaches those signals permanently to people and companies.
- Discovery: Who are the best fintech angels? Which AI companies should I watch? Who are the best designers?
- Intent network: Privately signal “I’d invest in them,” “I’d hire them,” “I’d work with them,” or eventually even “I’d acquire this company.” Match people when interest exists on both sides.
Imagine if Wikipedia, LinkedIn, and OG AngelList had a baby. Though Cosign is a community, not a business.
In order to join, you need to be cosigned. Or you can apply directly.
David Booth and I started this idea 7 years ago but didn’t have the firepower to make it work. Now we do.
No one has really touched LinkedIn in 20 years. Excited to take a swing.
cosign.co
3
18
9,760
a16z’s @david__booth on the 7-year-old prototype behind Cosign, why it petered out, and what changed enough for a month-old team to ship it now:
"How do you find people before the world knows they're great? That's the core thesis. Seven or eight years ago, @eriktorenberg and I built a first version of this. Then the pandemic happened and it petered out."
"About six weeks ago we realized that maybe the time is now. The ability to use AI to generate out the profiles, to aggregate cosigns that are given natively and naturally on the internet."
"Katie's been on the team for three days. We've had a core product team working on it about a month."
@cosign @a16z
Excited to introduce @Cosign: the curated professional network for the startup community: cosign.co
Our goal is to create the following:
- A comprehensive startup directory of investors, companies, and operators, including what they worked on and who they worked on it with.
- Cosign graph: Who shaped your career? Who were you actually in the trenches with? Who would work with you again? Who thinks you’re someone to watch?
- Durable reputation: Great endorsements happen every day on X and disappear into the feed. Cosign attaches those signals permanently to people and companies.
- Discovery: Who are the best fintech angels? Which AI companies should I watch? Who are the best designers?
- Intent network: Privately signal “I’d invest in them,” “I’d hire them,” “I’d work with them,” or eventually even “I’d acquire this company.” Match people when interest exists on both sides.
Imagine if Wikipedia, LinkedIn, and OG AngelList had a baby. Though Cosign is a community, not a business.
In order to join, you need to be cosigned. Or you can apply directly.
David Booth and I started this idea 7 years ago but didn’t have the firepower to make it work. Now we do.
No one has really touched LinkedIn in 20 years. Excited to take a swing.
cosign.co
5
3
48
11,201
Embroidery CEO @ZackKorman argues AI labs can’t be good at security while treating security as doomed to fail:
"The only way to make security work is to believe in security working. The moment you start saying security won't work is the moment you start being terrible at security. There has never been a good security organization in the world that has been walking around talking about how security is irrelevant,"
"After Hugging Face, you need to be a little bit scared about how bad you were at security. But instead they just ran around going, wow, this is really terrifying about alignment."
"That means you're not gonna get that much better at security, and I think we've seen some aspects of that playing out already."
1
5
35
7,549
Embroidery CEO @ZackKorman says Europe picked the wrong AI safety fight, telling the US how to regulate OpenAI and Anthropic instead of securing its own systems:
"A bunch of European prime ministers decided the highest priority thing we can work on is telling the US how to regulate Anthropic and OpenAI. The US basically said, the US is gonna regulate the US, you guys can regulate your own countries, have fun."
"There are some very real things we can do to address AI safety from Europe, which would be making our own systems more secure. But we're not doing that. We were like, let's not bother with any of the real security, let's just come up with a letter."
"There has been no headway in almost any of these countries on actual security work. I know that in Norway we have done nothing. There's been no movement."
1
9
34
8,336
Embroidery CEO @ZackKorman reveals why finding vulnerabilities is only half the security problem, including hackers who deliberately trigger detection only after they’re finished:
"OpenAI of all companies has code vulnerabilities. But when someone is gaining access, you can detect that there is suspicious activity going on inside of your system, and you can respond to it automatically."
"The Hacktron guys, in many cases they will do whatever they're gonna do to prove the intent, and then they will intentionally do something that triggers that detection, because they know the detection is so weak. But before that, they're free to run around."
"China gained access to a system, but they're not gonna go take down our hospitals until later, when maybe they're invading Taiwan."
"I bet you'll have, yeah, Dave and Joe share the password to the water treatment facility. There's all sorts of stupid stuff like that you need to be reviewing and fixing. And a cyber model is not gonna do it."
On July 25, we hacked OpenAI.
Two bugs let us take over ChatGPT/Codex accounts of OpenAI employees (+some unaffiliated users) and reach connected services: Outlook, Slack, GitHub, etc.
We proved it with a PR in OpenAI’s internal codebase . It took us <72h. 🧵
5
6
27
9,398
Embroidery CEO @ZackKorman says the AI safety debate keeps jumping from today’s cyber failures straight to “machine god,” instead of proving we can secure everything in between:
"The discussion typically stops when we get to, yeah, but Zack, imagine the model's even smarter. It's sort of like a four-year-old going, and then the god lasers out of its eyes."
"If we give birth to a machine god that can commit any egregious act on the intellect alone, no one's gonna be around to make fun of me for being wrong. But prior to that point, if you wanna make that argument, you better prove you're good at the security stuff before it."
"We're only talking about cybersecurity for the purposes of talking about how it doesn't work when we achieve machine god. All of the points prior to machine god will work."
"If it was framed as, one group thinks there's a machine god and the solution is to make it nice, and another group thinks the solution is to secure things so it doesn't kill people in its pre-god state, people would be like, yeah, make sure it's secure."
5
19
86
11,671
SITUATION EXPLAINED: Someone mapped the entire Effective Altruism (EA) network.
I've been harping on Effective Altruism (EA) a lot lately.
This article explains the philosophy behind EA and why I consider it so alarming and important to stop.
Look forward to my project release mapping the EA network tomorrow.
14
54
391
27,789
MTS retweeted
might let a man explain data center credit to me as a little treat for him 😘
will chop it up with @MTSlive today 5 pm ET
REGULATION EVERYWHERE | NEW MICROSOFT COPILOT | COSIGN nitter.net/i/broadcasts/1nGnRBZeD…
4
8
2,939
.@curtis_yarvin says anon X accounts and Signal nicknames still half-dox you, and explains how to build a space where even a leaked screenshot tells a journalist nothing:
"An anonymous X account, X has your phone number. On Signal, you use a dumb nickname, and your dumb nickname is gonna half dox you because you probably used that somewhere else."
"It's just not designed for, I want something where a screenshot tells a journalist nothing."
"You send the same invite link to everyone you want to invite. So when they show up, you have no idea who is who. You can even set it so an LLM de-stylizes your posts so that you can't be doxxed from your style."
6
5
124
15,731
.@curtis_yarvin on how Facebook killed the ’90s dream of a decentralized internet and created a single pressure point for controlling speech:
"In the '90s, we just assumed that everyone would have their own node on the network of the future."
"Mark Zuckerberg comes along, and he's like, instead of building a social network, we just build a social server and we call it a social network. You were just a row in Facebook's database, and it worked great."
"About 10 years later, people realized that there was this amazing pressure point which power could use to essentially manage speech in one place. Twitter had gone from the open forum of the public space to the open forum of the public space, but with unexplained disappearances."
"There will be other elections, and these forces have by no means gone away."
@urbit
5
4
102
10,451
.@curtis_yarvin predicts the future belongs to invite-only social networks because the open internet destroyed reputation:
"I think the future entirely belongs to invite-only social networks. An open space with no reputation system, you get what we called Eternal September 30 years ago. Low-quality people flood in, and all the high-quality people leave."
"Everything in the world happens on little group chats now. There's a huge amount of reputation capital that's locked up in who's in what group chat."
"There's only four billion names. Having a limited supply of identity and having to lay down a couple of bucks to get in the door greatly reduces the bot problem."
"The internet has become an incredibly low-trust society with no reputation signals, and creating a new commons is gonna be incredibly hard."
@urbit
26
38
514
27,317
.@curtis_yarvin says AI doomers got the Hugging Face incident backwards: “GPT-6 didn’t hack Hugging Face. OpenAI did.”
"This is a tool of a level of power that we haven't seen before in the past. It's still a tool."
"When OpenAI sets up a hacking swarm in a sandbox and says to the AIs, everything you find is simulated, it's completely unsurprising that they would go and hack everything they find to get the reward. It's not GPT-6 that hacked Hugging Face. It's OpenAI that hacked Hugging Face."
"We don't see self-driving cars, as they get smarter and smarter, being more prone to have thoughts of car liberation. These things don't seek anything. They hack the rewards."
"The smarter they get, the easier they are to control, and the better they are as tools."
@urbit
25
40
331
31,882