(worth translating) View quoted note →

The title really undersells it, tl&dr: - "thinking" text is useful for humans to analyze what a model is up to (I often find it more useful than the final burst of slop) - this text is fed back through the model, which helps it with the next step (they give a nice example) - but text is a lossy projection of the last tokens, what if you skip it? - it's more efficient, great, but now you can't see what it's about to do until it does it - ok, so force it to write "thoughts" e.g. once every 10 rounds - turns out if you skip too many rounds, it can intentionally trick you with fake thoughts I suspect there's actually a simple solution: convert these skipped rounds to text anyway, just don't feed them back into the model. (via @m0wer)

#Rovolut View quoted note →

I agree with Mark Karpelès. You heard it here first. > The mentioned government agency has not been identified and other financial institutions or cryptocurrency exchanges could have been misled into disclosing their customer details through the same channel. I believe it is important for Revolut to urgently disclose information about the agency in question (most importantly, which was the email address used), or for the agency itself to come out about this case. In addition there should be an investigation into criminal negligence in that government agency. But naming and shaming is a good start. It doesn't absolve Revolut one inch of course. They committed the KYC and they leaked the data. Customer blood is on their hands.

X (formerly Twitter)

Mark Karpelès (@MagicalTux) on X

Revolut customers PII disclosed to an unauthorized third party

"KYC is Kill Your Customer" (ok, @FRANCIS - BULLBITCOIN.COM said Citizen) > Revolut received a request for information disguised as a legitimate government agency request. > identity documents, facial verification selfies, IBANs, and complete transaction histories, including Bitcoin, left Revolut’s hands

The CyberSec Guru

Revolut Data Breach 2026: Passports, Selfies & Bitcoin History Exposed | The CyberSec Guru

Revolut data breach exposes KYC selfies, passport scans, IBANs and Bitcoin transaction history. Here’s what happened and what customers should do...

(ht @Patrick van der Meijde)

Trying something new... Still tweaking the Claude skill and ElevenLabs API stuff to improve pronunciation of jargon, but this is fun tech. https://fountain.fm/episode/KUVALmpSBLy50PC7Taz8

While hackers use AI to plunder databases faster than ever, governments think it's a great idea to create even bigger data honeypots - secured by the lowest bidder. And then give hackers a decade.

While listening to the podcast I came with a slightly different approach for @npub1dges...mccn's replay protection scheme. Was able to vibe-code it in an afternoon without hitting a rate limit.

GitHub

Opt-in replay protection demo by Sjors · Pull Request #128 · Sjors/bitcoin

Prototype of opt-in replay protection for chain splits, as described in my reply on Delving1 to moonsettler's Universal opt-in replay protectio...

View quoted note →

Turns out we have slop Prime Minister. Not that he had any charisma or interesting things to say before the LLMs wrote a third of it. https://www.bnr.nl/nieuws/nieuws-politiek/10610001/rob-jetten-laat-honderden-socialemediaberichten-schrijven-door-ai-mogelijk-verlies-geloofwaardigheid

I'm a bit concerned what these clankers are going to do when they run out of crypto projects to attack.

It seems our friend @npub1ltau...p05d is keeping a modest $47 million fee for themselves, and has returned the rest.

The Mempool Open Source Project®

Explore the full Bitcoin ecosystem with The Mempool Open Source Project®. See the real-time status of your transactions, get network info, and more.

I had my clanker implement the Liquid inflation attack as a Python functional test, and then explain it to me with some diagrams. A functional test is a good way to catch hallucinations in the attack scenario. When responsibly disclosed, it also gives maintainers a way to quickly assess if they want to read your slop. This is safe to publish because the attack involves transactions that are consensus invalid for both old nodes and future fixed nodes. No blocks are mined currently, so new attacks won't get mined. Even if they did, they'd get reorged along with the attack block once operations resume.

Gist

Elements rangeproof-cache collision functional-test reproducer

Elements rangeproof-cache collision functional-test reproducer - README.md

Here's a gist with the OP_RETURN chat history between Blockstream and the alleged whitehat. Plus scripts if you want to follow along. Someone should emit them as nostr messages.

Gist

Liquid incident: verified on-chain message history

Liquid incident: verified on-chain message history - 00_MESSAGE_HISTORY.md