Raphael Spannocchi

The true test of any organizational paradigm or collective consensus building effort is this: Does it make the whole more intelligent than the one, or less? Do people get better when interacting with the system? Or worse? It's really that simple, and that hard.

one thing I love about AI agents is that you never have to tell them "hold that thought"... lol

The spirit of most banking regulation like the 1940 Investment Company Act, Basel and much of IOSCO guidance is simply about transparency and disclosure. DeFi prides itself as transparent, because everything on-chain is open to read for anyone. In practice this is exceedingly hard to do, and many decisions get made off-chain. DeFi would do very well to embrace many of the principles of banking regulations and improve its documentation and legibility. TL;DR: Just say it like it is. We're doing a workshop on this with @MCooperResearch on Wednesday here on X. Feel free to join if this interests you. x.com/i/spaces/1nJOL…

Excellent. Documentation goes such a long way. Just having a place where you spell out your assumptions in a legible, coherent way. Instead of hand waving in 25 different forum posts and burying the actual risk profile in tech docs that are hard to find. Some form of standardized risk testing is also noteably absent.

When you step back, I think it’s clear that one of the lessons in the Kelp Incident is that DeFi needs to account better for priority of claims. No, this doesn’t mean slap a tranching layer on top - many have defective features that will break them in my opinion, and they only address part of the problem space. In no particular order, here are some pain points and my recommendations as a starting point: 1) Clarify that lending is non-recourse. If you’re a borrower, this means your maximum risk is your collateral securing the loan. This is EXTREMELY valuable as a feature, and DeFi needs a way to codify or formalize this so institutions believe they won’t be identified and pursued if some collateral becomes insufficient to cover losses. 2) Build in mechanisms where the lenders simply foreclose on the collateral assets if they cannot be liquidated. This allows illiquid markets to clear, but also needs to be understood by lenders as a possibility. This would, coincidentally, make lending on RWAs, semi-liquid assets, and permissioned assets easier. To my knowledge, only Silo and maybe Ajna have this as a mechanism. 3) Understand that liquidity creates de facto different seniorities amongst ostensibly pari passu lenders. This is an urgent problem amongst all of the major lending protocols, and is a real problem for any evergreen fund out in TradFi as well. In the real world, this is why painful practices like clawback windows exist in the run up to bankruptcy, but that’s neither possible nor desirable in DeFi. In theory, some protocols offer immediate pro rata haircuts, but in practice this is hard to execute well, and generally at the discretion of someone with an incentive to drag their feet. This creates race conditions where de facto losses tend to accrue to the last out, creating the incentive for bank runs, as we were most recently reminded. 4) Software or financial intermediary. This is a big one and has large implications for things like recovery claims. Way back in the mists of time, DeFi was supposed to be software (we call these projects protocols for a reason), like SMTP or POP3. Devs weren’t touching your funds (in theory), but were updating software to support various digital assets. It’s pretty easy to argue that software like Ajna or Uniswap are indeed software, not financial institutions. Morpho also has positioned itself well, albeit with the wrinkle that curators are now where the buck stops. Contrast with many CeFi (often branded as CeDeFi) companies or token-projects where the team makes investment allocations and decisions rather than expanding or curtailing support for an asset and then seeing whether users react with utilization. If I’m a lender who loses money and feels litigious, these distinctions matter a lot, because it affects who my ambulance-chasing, Dollar Store attorney will name in a lawsuit. Clearly delineating between “dumb pipes” (can’t upgrade), “smart pipes” (expand/curtailing support but specific users are not targetable), and onchain asset manager will better let every one of those categories lean into their comparative advantages. And even the onchain asset manager would be less exposed if they make sure to address Number 1 (clarify non-recourse status of DeFi lending), although they may themselves have to put up collateral instead of just opening a vault.

@bluestats_one @reserveprotocol @bluestats @nnevvinn how did you model it?

Hey @reserveprotocol 👋✅ reference [CF Large Cap Index] $LCAP Check your DM, we modeled the risk of DTF cc @bluestats @raphbaph @nnevvinn

@chacon I would argue that comments on Word are en par to anything Github has to offer.

The big problem with everything legal I’ve ever done is MS Word and redlines. Legal needs a github - markdown, diffable, mergeable, etc. I’m sure everything changes with AI, but if legal collaboration is still emailing fucking docx files around, I tell you thats not the answer.

@benedictk__ Entrepreneurs in Germany are often more concerned with the law than the actual government. It's a weird form of masochism

San Francisco is great because everyone is excited about AI and nobody opens the conversation with „Aber was ist mit der DSGVO?“ Bro build your chatty wrapper worry later.

I think @Meta Muse Spark is really good. Very well tuned for it's purpose. But: They're swapping something they dominated (open source models) for something they're playing catch up at. Not sure it's a great move. They could be American posterboy OSS champions. Would help them a lot vs being their own closed source slop bazooka

@0xdoug biggest user of OAI tokens was duolingo by a mile. ...

Sure. I’m not saying non engineers don’t use Claude. But do they generate significant revenue? Unless there are like 100 million Claude subscribers, the only way to get to $44b ARR is finding users who will pay $1k/month or more. And (afaik) the only way to spend $1k on Claude is to use API billing and consume on the order of 10 million tokens per day. And the only workstreams I’ve really seen using millions of tokens per day is coding and technical work. I could be wrong, but it’s hard for me to imagine lawyers or product managers or investment bankers using 10 million tokens per day in their normal work.

@beffjezos blood panels are very lagging indicators of health.

Traditional healthcare is too expensive and isn't made for health optimization. You can get a full closed cybernetic loop now (perception, inference, control) with biomarker testing, AI, and peptides. This is the future of health optimization

@0xdoug there can't be that much claude use outside the US? brother.. I have a flight ticket to sell :)

I’m really struggling to see how the back of the envelope math on this works out… There are generously 4 million characterized “software workers” in America. That’s pretty broad and includes a lot of people who aren’t really classical engineers don’t produce that much code. That comes out to nearly $1k per month of average Claude spend across every dev in America. Yes, there’s some international usage, but it can’t be that much. Yes there is some non software Cowork usage, but that doesn’t use that many tokens. Yes, some non engineers are using Claude to vibe code, but I really doubt many are spending hundreds per month on. Even if we assume 50% of all software workers are using Claude, that comes out to $2k spend per month per Claude user. Thats 10X more than the highest tier Max subscription. So almost all of Anthropics revenue has to be API billing So the only explanation is that something like 20%+ of software engineers are not only Claude users but on API billing and regularly spending thousands per month. At $5/m Opus tokens that means the average API user has to be going through something like 25 million tokens per day. *OR* the other possibility is API revenue is heavily power law dominated. Maybe there’s just something like 100k super users who are making up the majority of the revenue. For that to work the typical super user would have to be spending on the order of $50k/month and guzzling nearly 1 billion tokens per day.

@J_E_J_H @WojPawelczyk yeah these tend to be hard to keep clean

@raphbaph @WojPawelczyk Solved. Just use a pop open locker. Similar to a an Amazon locker. Glass cover and let you select your specific egg.

Egg vending machines (jajomaty) are becoming more popular in Poland. They have transitioned from a niche curiosity to a common sight in both large cities (like Warsaw, Kraków, Poznań, Wrocław) and smaller towns. The primary driver is the desire for freshness. These machines are usually owned by local farmers, allowing them to sell directly to customers without a middleman. They operate 24/7, which appeals to people who want farm-fresh eggs outside of traditional market or grocery store hours. While older models were simple lockers, newer machines often feature touchscreens and support for contactless payments, fitting well into Poland's highly digitized payment culture.

fun

@Whachazuck @WojPawelczyk tiny robot elevator. quite common here in Germany too

@raphbaph @WojPawelczyk Really, I’d have thought it’d be like the Dutch machines with crokets 😋, a little door for each one, how do they lower the eggs?

@urbLon @WojPawelczyk hope some boomer invents better vending machines and absolutely kills it :)

@raphbaph @WojPawelczyk Ok boomer

GPT 5.5 is a really good model, ngl.... Mythos better be cooking, or it's the foot gun of the AI race.

@ryancarson doesn't AFS have versions?

This makes a lot of sense

@dhh xcode is pretty bad too

Ternus has a unique chance to reset the relationship between developers and Apple: 1) Tear down the App Store tollbooth, 2) reintroduce Boot Camp for M macs so we can run Linux on them, and 3) fix the infuriating 500ms workspace animation delay. Approval rating +50% instantly.

@DFinsterwalder Lem is the GOAT

@raphbaph I won’t change your mind and rather recommend a fitting book to read while your Golem does the work for you.

One of the best things to do while waiting for your agent to finish work is reading a science fiction book. Change my mind!