Sovereign AI - AI should be yours. https://teapawt.net/t.co/9MNxjvBYrn
Milpitas, CA
Joined July 2008
- Tweets416
- Following153
- Followers133
- Likes1K
Ubuntu had just promoted their CVE-2026-98432 Linux kernel fix to 24.04 LTS and 26.04 LTS to regular production updates. They haven't updated their own CVE page yet.
However, the fixed kernel packages are now available via apt without needing to use the -proposed pocket.
If any company gets hacked because of Ubuntu insisting on not fixing CVE-2026-98432 on time (and no, having a fix in -proposed doesn't count) - there needs to be a class action lawsuit against Canonical.
Random thoughts before going to sleep
I was talking to an old pastor after going to church last Sunday. The pastor used to be a tech executive - and so he's extremely well versed in technology, despite his age. One thing he mentioned, was how AI generated sermons could easily be more logically correct and referentially consistent than what he or other pastors can write or speak. Then, what is the value of the pastor? What is the value of the AI?
I gave him perhaps half an answer - not from a religious perspective, but from a product design perspective. But what I was internally thinking, was where did the creative spark really came in.
AI can generate tons and tons of designs these days - no problem. But whenever you drill down into the details, you'll find that the AI would generate something but it doesn't really understand how a user is supposed to feel about something, or what's the most sensible way for a user to build his relationship to it. AI can replicate known best practices - but when you give your product or thing enough details, you'll start to see it doesn't really understand humans, especially human feelings.
Let's say you ask an AI to generate a kitchen. If your prompt is just "generate a kitchen" - you'll get something half decent. However, if you want a specific kitchen that you want to use - e.g. you want a certain set of spices, certain set of utensils, certain types of pots and pans, certain types of food, certain kind of cooking and washing arrangement for your own cooking process, etc. You give it enough descriptions... and... the AI will cram **everything** into the generated kitchen. No organization. No space for you to walk around. No space for you to sit with your friends at the kitchen island to enjoy a sensible view.
A human designer, using the AI - would work much quicker than without AI. He would think about how a family is supposed to interact in the kitchen. How the utensils, pot and pans are supposed to be used, washed and stored. He would sculpt out the non-sensible suggestions from the AI generations, and come up with something a family would feel happy to be in.
So, if you think about it - what is really the difference between the AI designer model, and a human designer? What is the keyword?
"Something a family would feel happy to be in". Great design is not just about aesthetics - it is about building relationships.
And among all the relationships that people can build in this world - e.g. exploitative, transactional, argumentative, cooperative, etc. What is the highest form of relationship you can make in this world?
Love.
A design that can make a family feel happy to be in, is a design of love. That is the creative spark.
Yes, AI can learn and emulate it, in the form of another set of best practices. But as the world changes, how we care about people also changes. The spark of love and care for each other, is the source of it all.
I... wasn't able to give such a full answer to the pastor back in Sunday. I was only able to give him, maybe 50% of it. But, it doesn't matter. It's more than enough.
What the red button vs. blue button meme highlights is how everyday people focuses on answers, even though they're given a contrived premise that's engineered for engagement only.
I'll just make random reference to HxH - "A lethal pop quiz doesn't need to be answered".
Great summary of the current state of the Clarity act, being marked up in Senate right now.
martinkou retweeted
The new Senate Banking draft of market structure just was published and here is where ChatGPT says it draws the regulatory perimeter when it comes to self custody interfaces (This is quick - a deep dive is required):
This is the crux of this bill — and the answer is yes, it really does protect non-custodial web interfaces, but only up to a very sharp line.
Here’s the clean way to understand it.
⸻
The bill creates a three-layer stack
The bill doesn’t say “wallets good, websites bad.”
It says:
There are three legally different things:
1.Protocol layer
Smart contracts, blockchains, bridges, DEX contracts
2.Control layer
Who can stop, change, route, or custody transactions?
3.Interface layer
The website or app users click on
Regulation attaches only if control exists.
⸻
When a web interface is PROTECTED
A web interface is legally treated as mere software if:
•It does not hold user funds
•It does not control private keys
•It does not have authority to approve, block, reorder, or reroute transactions
•It does not have discretion over execution
•It does not act as an agent for the user
In that case, even if the website:
•Shows prices
•Builds transactions
•Suggests routes
•Broadcasts transactions
•Charges UI fees
…it is legally a non-custodial publisher of software, not a broker or exchange.
That covers:
•Uniswap web UI
•1inch
•MetaMask swap UI
•Non-custodial bridges
•Wallet-embedded DEXs
They are statutorily carved out of financial regulation.
⸻
When a web interface becomes REGULATED
The moment the operator has control, it flips.
If the website operator can:
•Hold assets
•Move funds without user signature
•Batch or net trades
•Route orders through proprietary liquidity
•Pause, censor, or reverse transactions
•Decide whether a transaction executes
then the interface becomes a:
Broker / dealer / exchange
even if it still uses blockchains underneath.
That captures:
•Coinbase
•Binance
•Robinhood Crypto
•Custodial bridges
•Hosted wallets
•CeFi yield platforms
⸻
The key test CLARITY uses
The bill’s rule is basically:
Who has the power to make the transaction happen?
If the user’s private key is the only thing that can move funds → protected
If the website operator can move or stop funds → regulated
The web UI is irrelevant by itself.
Control is everything.
⸻
Why this is a real safe harbor
CLARITY doesn’t just say “we won’t regulate you.”
It says agencies may not re-classify non-custodial interfaces as intermediaries just because they:
•Are popular
•Earn fees
•Look like an exchange
•Are easy to use
This blocks the SEC/FinCEN theory that:
“If users trade through your website, you’re an exchange.”
Under CLARITY:
“If users trade through their own keys, you’re software.”
⸻
Bottom line
Yes — the bill really does protect non-custodial trading interfaces.
But it mercilessly regulates anything that:
•Touches custody
•Touches execution
•Touches control
That bright-line is what makes the bill powerful.
There needs to be a class action lawsuit for all crypto companies and holders w.r.t. arbitrary de-banking. Banks must pay for the business and personal impacts caused by their discriminatory practices.
decrypt.co/351895/crypto-deb…
reddit.com/r/ethereum/s/BO1a…
Trojan horse discovered in VSCode Solidity plugin. Possible rootkit or crypto stealer.
martinkou retweeted
We had a bug in Reth's state root computation today in Ethereum mainnet, which caused multiple nodes to stall.
Recover by running the following:
```
1. reth stage drop --datadir DATADIR merkle
2. reth stage unwind --datadir DATADIR to-block 23272426
3. reth node --datadir DATADIR --debug.tip 0x2eb1fcafd864aafe21f2cb66310a869b8945231330f0da80c9e9b77861b56fca
```
We are still investigating root cause, but the above gets you past the issue. This is safe to run on both pruned & archive nodes, won't lose you any RPC information, and will take ~45m to rebuild the trie before it continues syncing.
Brutal moment, but ultimately one we learn from and get stronger as we break through performance barriers.
github.com/paradigmxyz/reth/…
martinkou retweeted
The new Hummingbot MCP Server is now available in @Docker MCP Hub!
One click adds the MCP Server to your Docker Desktop, and then it will expose a comprehensive set of tools to your AI agent of choice, turning a local or remote Hummingbot API into a powerful trading agent.
Wow, apparently you can add prompt injection attacks in public forums and randomly attack insecure agentic crawlers, including Perplexity.
brave.com/blog/comet-prompt-…
martinkou retweeted
Now you can just talk to your bots!
Check out @cardosofede's livestream on how to use the new Hummingbot MCP server to add credentials, analyze your portfolio, execute trade, deploy bots, & more!
youtube.com/live/Tvun-Zn-ii0…
martinkou retweeted
this Friday I will be streaming again! let's build some cool stuff with Hummingbot API and Hummingbot MCP...
let me know what do you want to see ;)
martinkou retweeted
like or retweet this post if you didn’t turn bearish on Ethereum, ETH, and its developers and community at the bottom
because you deeply understood that Ethereum is the future of global settlement and trust
and you weren’t going let someone psyop you out of it