Deep (@deepitreal) header
Deep (@deepitreal) profile picture

Risk, AI/ML, social coordination. Previously @chaoslabs @delphi_labs

NYC
Joined October 2017
Photo by Deep (@deepitreal)Photo by Deep (@deepitreal)Photo by Deep (@deepitreal)Photo by Deep (@deepitreal)Photo by Deep (@deepitreal)Photo by Deep (@deepitreal)Photo by Deep (@deepitreal)Photo by Deep (@deepitreal)Photo by Deep (@deepitreal)Photo by Deep (@deepitreal)
Another big reason I think this acquisition makes sense is that exposure to OpenRouter is almost like having exposure to the entire AI model ecosystem. It's almost like owning an index.
I can now officially say it: OpenRouter is being acquired by Stripe! Although we cannot comment on the price, this marks one of the fastest acquisitions of this scale in history: almost exactly 3 years after founding. In this blog post, we write about the founding story and history of OpenRouter, why we invested and why Stripe is the perfect acquirer. A few of the highlights: People love to say "OpenRouter has no moat". They think of moats primarily as technical difficulty. As a two-sided model marketplace, OpenRouter provides a really valuable service to model labs: predictability in capacity and distribution. And to user: a simple way to access all models. With scale, OpenRouter is able to provide better uptime, more models, with the cheapest discounts and have model labs prioritize launching on OR first! — We saw this with Zuck and Elon both tweeting about their models on OR first, all Chinese models announcing on OR and OpenAI doing their secret model launches on OR and having the largest, exclusive GPT 5.6 discounts too! — Andrej Karpathy distinctly calls out OR as the "Transfer switch for AI" — OR becomes the go to data source for multi-model usage with their iconic rankings page. — Tokens grew ~30,000x since first launch and as of this week is at a >4.5 quadrillion (that's 10^15!) run rate, a consistent 33% MoM growth for 3yrs! Founders Alex Atallah (@alexatallah), Chris Clark (@cclark) and Louis Vichy were at the center of it all. OpenRouter did it different: no hype about revenue, a completely remote workforce, no buzz around fundraising, just relentless focus on product. No doubt that the lessons Alex carried forward from OpenSea on token marketplaces played a role in making OR successful. Lastly, and definitely the least important, OpenRouter was very special to me as my first deal at Menlo: we partnered at the Seed, led by Anj (@AnjneyMidha) at a16z, and led the A when my partner Matt Murphy (@mmurph) and I joined the board. I feel very fortunate to have miraculously worked with such a special team!
1
1
91
I built JobStash to help people find crypto jobs. Now I need the Ethereum community’s help finding mine. My time at Nethermind has ended. I’m looking for an Engineering Manager, Head of Engineering or Technical Program Director role with ownership across people, technology and execution. At Nethermind, I led engineering delivery for Surge, a $1.6M program, and worked across business development, strategic accounts and incident response. Thanks @Smartprogrammer for leading the product and @tkstanczak for trusting me with Surge. Before that, I grew JobStash to 100k+ users, hired and managed a seven-person team, and owned its data and agentic AI systems. Former Principal Engineer. 20+ years in software. I care deeply about Ethereum and want my best work yet in Ethereum, infrastructure, AI or developer tooling. Amsterdam-based; U.S./EU citizen; open to remote. CV: docs.google.com/document/d/1… Know the right team? Please DM me or repost.
🤖 Made with AI
10
34
6
97
7,278
i see that many like to write comparisons between robinhood chain and base, so here's mine: * canonical bridge - robinhood: permissioned and trusted. only two EOAs can propose and challenge state roots. - base: permissioned TEE proposes state roots, but anyone can challenge with a ZK proof if compromised. If TEE is down, anyone can also propose state roots with ZK. * censorship resistance - robinhood: operator can both censor L2 txs and txs sent from L1. - base: forced txs from L1 can bypass sequencer censorship, only an upgrade can stop them. * upgradability - robinhood: 2/3 msig, all internal EOAs. - base: 2/2 between a base internal msig (3/6) and a security council (8/11, all external to base, full member list here docs.base.org/base-chain/sec…)
10
9
4
124
25,046
one of these days kalshi or polymarket should announce some really fucking awful idea and see how long it takes before the other one announces an exact copy.
8
4
3
99
5,417
Energized after the monthly progress call of the 9th Edition of the Kleros Fellowship of Justice! The breadth of research being conducted by this cohort reflects the maturing intersection of law, mechanism design, and technology. Some of the topics currently under development include: - AI Content Curation: A Chrome extension utilizing Kleros for filtering "AI slop" online. 🤖 - Legal Comparative Analysis: Researching e-commerce arbitration frameworks in Portugal. 🇵🇹 - Decentralized Refereeing: Exploring @Kleros_io as a crowdsourced VAR system for sports. ⚽️ - Mechanism Design: Formalizing Schelling point models to handle "no decision" outcomes. 🧮 - Institutional Outreach: Engaging with regulatory frameworks for arbitration in Bolivia. 🇧🇴 - Sociological Mapping: Analyzing the friction between decentralized jurors and traditional ADR in Nigeria. 🇳🇬 In the coming months, participants will submit their final papers. I look forward to seeing the results of this six-month cycle of work! kleros.io/fellowship/
2
9
1
28
1,121
1/ stETH CAPO Misconfiguration Today, a misconfiguration on Aave's CAPO oracle caused wstETH E-Mode liquidations, resulting in a loss of 345 ETH. No bad debt was incurred, and all affected users will be fully reimbursed. More below.
24
43
31
172
75,057
What does the token actually do? This has been one of the most fundamental unanswered questions in DeFi since the beginning. Does the token control governance? Does it have any claim on the treasury? Does it receive protocol revenue via buybacks or dividends? Today, we launched Token Rights on DefiLlama. Token Rights gives you a clear, standardized view of what a token entitles holders to: revenue, treasury, governance, or none of the above. We’ve rolled this out across two dozen protocols, including additional context like historical governance discussions around token rights and whether teams raised equity separately from the token.
170
183
84
1,487
216,041
“Ethereum was not created to make finance efficient or apps convenient. It was created to set people free” This was an important - and controversial - line from the Trustless Manifesto ( trustlessness.eth.limo ), and it is worth revisiting it and better understanding what it means. “efficient” and “convenient” have the connotation of improving the average case, in situations where it’s already pretty good. Efficiency is about telling the world's best engineers to put their souls into reducing latency from 473 ms to 368ms, or increasing yields from 4.5% APY to 5.3% APY. Convenience is about people making one click instead of three, and reducing signup times from 1 min to 20 sec. These things can be good to do. But we must do them under the understanding that we will never be as good at this game as the Silicon Valley corporate players. And so the primary underlying game that Ethereum plays must be a different game. What is the game? Resilience. Resilience is the game where it’s not about 4.5% APY vs 5.3% APY - rather, it’s about minimizing the chance that you get -100% APY. Resilience is the game where if you become politically unpopular and get deplatformed, or if a the developers of your application go bankrupt or disappear, or if Cloudflare goes down, or if an internet cyberwar breaks out, your 2000ms latency continues to be 2000ms. Resilience is the game where anyone, anywhere in the world will be able to access the network and be a first-class participant. Resilience is sovereignty. Not sovereignty in the sense of lobbying to become a UN member state and shaking hands at Davos in two weeks, but sovereignty in the sense that people talk about "digital sovereignty" or "food sovereignty" - aggressively reducing your vulnerabilities to external dependencies that can be taken away from you on a whim. This is the sense in which the world computer can be sovereign, and in doing so make its users also sovereign. This baseline is what enables interdependence as equals, and not as vassals of corporate overlords thousands of kilometers away. This is the game that Ethereum is suited to win, and it delivers a type of value that, in our increasingly unstable world, a lot of people are going to need. The fundamental DNA of web2 consumer tech is not suited to resilience. The fundamental DNA of _finance_ often spends considerable effort on resilience, but it is a very partial form of resilience, good at solving for some types of risks but not others. Blockspace is abundant. Decentralized, permissionless and resilient blockspace is not. Ethereum must first and foremost be decentralized, permissionless and resilient block space - and then make that abundant.
1,140
1,206
514
6,970
1,135,223
From Sam Harris HT @honest_math
We need to somehow hold the algo's / media / etc... accountable for making the population batshit crazy. Not sure how it's possible, but this is fucked and I'm far from immune to it. I see folk taking a break from this hell site... wish I wasn't addicted but will give it a go.
20
15
3
171
62,697
This is a solid demonstration of Railgun's privacy pools mechanism ( papers.ssrn.com/sol3/papers.… ) working in practice, allowing Railgun to avoid serving proceeds of crime without using any snooping / backdoors. How it works: * Anyone can deposit into Railgun. * After you deposit, there is a 1 hour period during which various algorithms detect whether or not the deposit likely came from what the algorithms consider to be criminal activity. * If your deposit passes the filter, then after 1 hour you can use ZKPs to withdraw privately (but ideally wait longer to get a good-enough anonymity set). * If your deposit fails the filter, then you can only withdraw back to your own address. There is no risk that your funds will get frozen/seized, you just can't benefit from the privacy pool. If you disagree with Railgun's filters, anyone is free to fork and make their own pool with their own rules, though if you can't get reasonably wide public support you're going to have a tiny anonymity set. railway.xyz/
Eventually, @zkLend has suffered a $9.5M exploit on the Starknet network. Stolen funds were bridged to Ethereum and laundered via Railgun, but due to protocol policies, the funds were returned to the original address by Railgun!
567
446
115
2,922
612,443
You know, in the midst of an unhinged public Nazi tirade, Kanye *still* decided that launching a shitcoin was way beneath him as a matter of ethics — makes you think huh
53
529
44
8,781
363,670
Reinforcement learning works well on domains that can be easily verified. This is why o1, o3, o3-mini, and r1 are so good at Leetcode and competition math. Ironically, being “easy to verify” is also what makes LC and math such a useful indicator of intelligence between humans
30
44
8
853
55,251
The good news? BTC hasn’t topped Alt season isn’t cancelled ETH may yet rise again We simply don’t know when. The best traders know this and will outperform the cycle-believers.
1
9
301
If anyone wants to build the ethereum community space Vitalik's asking for, DM me Social network @AgoraCitizenNet uses Rarimo's anti-sybil to prevent bots / multiple accounts distorting discourse Making a space within it exclusively for ETH-holders could also easily be done
We can make one based on proof-of-personhood ZKP too. @Rarimo_protocol has been recently working on cool stuff there, and we can make equivalents for @worldnetwork and @aboutcircles. It would be interesting to see where the average opinions on that, and the eth holder one, differ.
2
3
18
1,363
There should be a service that allows holders to pool their $AIXBT holdings and access the terminal for a tiny fee (to the service that enables this). Who’s building this?
1
1
5
724
GBhai 🙌🏾 A Bhai in need is a Bhai indeed. If you think you are worthy, DM to join.
79
5
8
120
11,887
Deep retweeted
Replying to @deepitreal
best r/r out there right now fartcoin = first organic meme conceived by an unprompted ai godflash = first organic meme conceived by multiple unprompted ai agents
1
1
3
465
A meme with a legit backstory, an AI agent called god, a game, and the ability to bet on the outcome. It doesn't get better than this!
Imagine @freysa_ai, but free to play and a much higher reward. That's the $GODFLASH challenge. And for the smooth brains at the back! $GODFLASH is now a live prediction market. The token price will eventually be reflective of someone’s ability to win this.
1
1
343
Deep retweeted
whatever man this is still how I'll remember iron mike got a massive paycheck at 58 and took a bunch of shots on the chin like a brick
43
33
4
482
56,754
So crazy to me that 6 months ago literally everyone serious AI researcher I talked to told me stuff like this wasn't feasible. Decentralized Training is gonna work - we're still at DDP stage and there's a long way to go for giant model runs, but no doubt where the vector of progress is pointing now. World is shifting beneath our feet and most people still don't know.
What if you could use all the computing power in the world to train a shared, open source AI model? Preliminary report: github.com/NousResearch/DisT… Nous Research is proud to release a preliminary report on DisTrO (Distributed Training Over-the-Internet) a family of architecture-agnostic and network-agnostic distributed optimizers that reduces the inter-GPU communication requirements by 1000x to 10,000x without relying on amortized analysis, and matches AdamW+All-Reduce in convergence rates. This enables low-latency training of large neural networks on slow internet bandwidths with heterogeneous networking hardware. DisTrO can increase the resilience and robustness of training LLMs by minimizing dependency on a single entity for computation. DisTrO is one step towards a more secure and equitable environment for all participants involved in building LLMs. Without relying on a single company to manage and control the training process, researchers and institutions can have more freedom to collaborate and experiment with new techniques, algorithms, and models. This increased competition fosters innovation, drives progress, and ultimately benefits society as a whole. This research is thanks to the hard work of @bloc97_ @theemozilla @apyh__ @UmerHAdil. We invite researchers interested in exploring this area to join us in our quest.
10
22
4
266
44,216