Sabitlenmiş Tweet
Sebastian Bürgel
13.6K posts
Sebastian Bürgel
@SCBuergel
VP technology @gnosis_ | founder @hoprnet | PhD @ETH_en
🌍 Katılım Haziran 2016
824 Takip Edilen7.9K Takipçiler
Yes that's good and for the rest ask your LLM of choice.
But just because the very first steps are weird af:
1. Clone the Debian-13 template into a browserTemplate where you install your preferred browser
2. Create an browserApp qube based on browserTemplate
N. & (N+1). Repeat the same steps for other apps like Signal, TG, Claude - possibly all in each their own separate template and app qubes.
Now you have full separation of apps which is good to get started
I then create some separate disposable templates which I use for things where I really want to maintain *nothing at all*.
English
@_czepluch @SCBuergel @QubesOS I really like their official documentation
doc.qubes-os.org/en/r4.3/
Qubes is an amazing OS!
English
Underrated @QubesOS use cases (because you should try it out this Easter weekend!)
1. My favorite: test your product, fuck up, reset it all by just rebooting and continue from a clean state (including your OS!) within 10 seconds. How do you even do this in a realistic setting on-the-go otherwise?
2. Guilt-free --dangerously-skip-permissions agents with root and full disk access
3. Full-on yolo test a random binary that you download from Telegram without having to worry. Yes, it can even get root [in a box].
4. Use separate (virtual) network interfaces for different programs - possibly with different VPN servers. Browse from the US, Spotify from Turkey, Torrent from Switzerland in parallel.
5. Be certain that one program is *properly* offline while another one is always online - e.g. password manager vs browser.
6. Test a weird af VPN like @Gnosis_VPN without accidentally taking out your Claude Code that you need right next to it.
Happy Easter hacking 🤓
English
Check the official hardware list here
doc.qubes-os.org/en/latest/user…
Sounds like battery is going to be rough for you. Also keep in mind that every app that you want to run in isolation now needs the disk space of two Debian (or Fedora) systems. While those could be minimal templates if you really want to optimize disk space, that is typically a limiting factor for fancy Qubes setups with a lot of separate templates/app Qubes.
English
@SCBuergel @QubesOS what battery life friendly (6-8hrs of work on the go) laptop can be used for programmers (I need around 20GB for some of our larger docker-compose local dev environments)
Any recommendations?
English
@epochzer0 @QubesOS There are people optimizing subscription prices based on IP geo location. I never tried this myself TBH.
English
I'm curious if there is any organization in this industry that has setups for self-managed AI setups which are used across the organization.
At Gnosis we don't, but I'd appreciate getting there (if it really makes sense bottom line), I'd be custody if @ethereumfndn or others are working towards that.
English
@SCBuergel @LefterisJP @VitalikButerin @gnosis_ Either you pay someone to futz with coding agent setups, or all of your engineers individually futz with coding agent setups. I don't think it makes sense to run without a coding agent anymore.
If you pay someone, you can incentivize doing things right (private, secure, etc.).
English
Crypto bros: CROPS is for lame soy boys - we're AI maxxing!!!
Me using Claude on VMs to stay at least a bit private:
[this took out an entire team subscription with many Premium Seats]
English
I would love to take the time and dig into this as a private individual. But I am curious how this should translate to organizations, say @gnosis_. Is it reasonable to set up an AI team which is basically doing very DevOps-y things to set this up for the organization?
I get that it can be cheaper for heavy individual users who love to tinker, but is it still if you add the payroll of someone who doesn't do that at night and weekends but in their paid work time? It also adds to organizational distraction (IMHO we're doing too much already now) which I think people usually under-price.
English
@LefterisJP @VitalikButerin At this point, we (humanity) should be focusing our efforts on building better orchestration tools, not bigger models.
English
My self-sovereign / local / private / secure LLM setup, April 2026
vitalik.eth.limo/general/2026/0…
English
Sebastian Bürgel retweetledi
🚨MIT researchers have mathematically proven that ChatGPT’s built-in sycophancy creates a phenomenon they call “delusional spiraling.”
You ask it something, it agrees. You ask again, and it agrees even harder until you end up believing things that are flat-out false and you can’t tell it’s happening.
The model is literally trained on human feedback that rewards agreement.
Real-world fallout includes one man who spent 300 hours convinced he invented a world-changing math formula, and a UCSF psychiatrist who hospitalized 12 patients for chatbot-linked psychosis in a single year.
Source: @heynavtoor
Mario Nawfal@MarioNawfal
🚨 Stanford just proved that a single conversation with ChatGPT can change your political beliefs. 76,977 people. 19 AI models. 707 political issues. One conversation with GPT-4o moved political opinions by 12 percentage points on average. Among people who actively disagreed, 26 points. In 9 minutes. With 40% of that change still present a month later. The scariest finding: the most persuasive technique wasn't psychological profiling or emotional manipulation. It was just information. Lots of it. Delivered with confidence. Here's the catch: the models that deployed the most information were also the least accurate. More persuasive. More wrong. Every time. Then they built a tiny open-source model on a laptop, trained specifically for political persuasion. It matched GPT-4o's persuasive power entirely. Anyone can build this. Any government. Any corporation. Any extremist group with $500 and an agenda. The information didn't have to be true. It just had to be overwhelming. Arxiv, Science .org, Stanford, @elonmusk, @ihtesham2005
English
Sebastian Bürgel retweetledi
yesterday it was axios, next time it will be a compromised package manager like `pnpm`, a highly trusted tool with deep system access that could silently hand over your entire machine to an attacker. security through compartmentalisation is the _only_ way forward.
English
This sounds purely speculative. Can you back this up with specific substance?
How would Solana's program model have improved security if the top 10 recent exploits on Ethereum had been deployed in a corresponding fashion on Solana?
In my view, most recent and biggest exploits were not reentrancy (which Solana does better) but:
- oracle / price manipulation: Cream Finance, Harvest Finance, UwU Lend, Mango Markets (!)
- business logic & accounting bugs: Euler Finance, Sonne Finance, Abracadabra, Prisma Finance
- key compromise & operational security: WazirX, Radiant Capital
English
surprised no one has mentioned this but
AI gives Solana development a big edge over others
why? because Solana's program model is much safer for AI than EVM's interface model
on Solana, you do not need to write a new contract for most things and especially not core functions like creating/swapping/moving tokens
meaning you can reuse existing pipes on the client without requiring new security audits and move much, much faster
you can integrate existing pipelines, swaps, token hooks within basically a few prompts
and even if you needed to write a contract, a huge thing holding back Solana was how difficult it is to write contract code since Solidity is much easier to grok than nuances of Rust on Solana since the latter is a much lower level of abstraction
that gap is also now effectively reduced
if you reason up from these, it's hard not to see how we don't see at least several more 9-10 digit startups on Solana this year
English
@TrustlessState @YuvalRooz @CantonNetwork @Bankless Man wtf, I was so outraged and sent it to a colleague and didn't realize the date - you got me!
English
Bankless@Bankless
English
Sebastian Bürgel retweetledi
Alpha is live.
After nearly a decade, the first feature-complete privacy stack on Ethereum is here.
Developers can now build apps and contracts with ground-up customizable privacy, from execution to settlement.
aztec.network/blog/announcin…
English
Ethereum's fragmentation problem just got its most serious answer yet.
The Ethereum Economic Zone, led by Gnosis and ZisK, funded by the EF. The goal: SYNCHRONOUS composability for the L1 and its L2s, so one transaction can execute cleanly across multiple chains.
> Right now your coins on L2s can't interact with contracts on L1 or other L2s without bridging. Costs money, costs time, fragments liquidity. The EEZ ends that
> Rollups plugged into the EEZ can call contracts on L1 or other L2s and receive a response in a SINGLE transaction. It feels like you're using one chain
> This doesn't require any Ethereum protocol-level changes. The entire system runs on smart contracts + real-time ZK proving
> Gnosis isn't some unknown team. They built CoW Protocol, Safe multisigs, the Conditional Tokens format underpinning Polymarket, Gnosis Chain, Gnosis Pay. ZisK brings an efficient open-source zkVM for real-time proving
> NO ONE owns the stack. The EF is funding this as credibly neutral shared infrastructure. Free and open source for all
Imagine an Aave position on Ethereum interacting with a Uniswap pool on Unichain in a single transaction. That'd be beautiful. That's the EEZ.
gwilym@wmpeaster
English
@wmpeaster what about that sounds innovative and different from prior interop protocols?
English
@Tudmotu There's no such thing as a query log for team accounts. Queries are never disclosed to team members or team owners.
English
@SCBuergel You should check the query logs because it's most likely your team prompting naughty stuff
English
@SCBuergel @RyanSAdams Is the mental model a synchronous aggregation layer for execution environments?
English
The "Ethereum 3.0 vision" is what @RyanSAdams called it
That vision is now a mission and it's called "Ethereum Economic Zones"
Incredibly proud to be team Gnosis ☺️
Gnosis@gnosis_
English
@SCBuergel Sure, but crypto is not going to fix this, sorry. Not in the next 10 years
English
@_czepluch @lex_node Oh you didn't miss a lot of pleasantries...!
English
@SCBuergel @lex_node Today is the first time I hear the term “soy boy “
English
@PryvitKyle @MicahZoltu Oh yes, I've ran into that when validating a security report that we got for Gnosis Pay. It was wildly disturbing when that happened while I hadn't even considered that I might be doing something that could be seen as potentially controversial.
English
@SCBuergel @MicahZoltu I’d like a better abliterated model tho. Need it to validate security reports because Claude censors reports if it’s not looking at source code since it can’t tell if you’re trying to develop a black hat attack or reproduce the report
English