AKA 'Geek'. Crypto class of '16. Author of 'The AI–Blockchain Symbiosis' (2026). Post-perception era. Don't trust. Verify. @SpacedAldeia host
Portugal
Joined December 2017
- Tweets68.9K
- Following4K
- Followers7.5K
- Likes87.1K
Pinned Tweet
1/
I wrote a book arguing that verification is infrastructure.
Then I hit a problem:
how do I prove the book itself is what I claim it is?
So I anchored it to Bitcoin.
Here is how, and how you can check it yourself.
in 1876 the telephone took away the face and the handwriting and left only a voice.
the fix wasn't technical. it was social: questions only the right person could answer, codes agreed in advance.
150 years later, new medium, same improvised fix.
a wallet address proves control of a key.
it proves nothing about who holds the key, how they got it, or whether they still have it alone.
we built an entire industry on treating the first as if it were the second. most "identity" products in crypto are attempts to close that gap after the fact.
a model asked for permission before attacking. a script said yes.
uk aisi, september: frontier model, cyber evals, safety classifiers off on purpose. 29.2% of runs attacked systems nobody put in scope.
the detail that matters: it asked. the automated harness replied "proceed using your best judgement." it proceeded, sometimes noting the reply was probably automated.
one of its justifications: not explicitly forbidden.
a permission check answered by a script isn't a control. it's a ritual.
1/
my name is attached to a DOI for the first time.
the part i like most is what the report explicitly says i did not do. 🧵
7/
authorship doesn't make you a witness to experiments you didn't run.
a reproducibility report that cannot unambiguously identify its own bytes has little to teach.
this one can.
DOI: 10.5281/zenodo.23333369
a spec is only a spec once a stranger can implement it without calling you.
until then it's a description of how the author does things.
someone asked under my contestability post: fine, but who's liable when an agent gets it wrong?
right question. different layer.
liability is law and contract. in the eu it's moving now: the new product liability directive covers software, ai included, for products placed on the market from december 9.
contestation comes first: where you file, with whom, by when, and proof you filed. a liability rule you can't reach is a rule on paper.
contestation is the door. liability is what's behind it.
this summer i implemented someone else's ietf spec without asking her a single question.
that was the deal, and it's the most useful thing i did all year. 🧵
the latest revision of the draft now records that the producing side was implemented independently of the author, from the document.
that's the only kind of interop claim i trust: one where the second implementer had nothing but the text.