How can democratic decisions be smarter than the median voter? Beyond Arrow's paradox. Building @Anode_GG.
Joined January 2010
- Tweets 15,353
- Following 805
- Followers 1,776
- Likes 33,799
938 Photos and videos
Pinned Tweet
9 Mar 2024
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.
4
1
30
9,639
last sentence is the kicker. everything's casino
Jun 12
There are plenty of warning signs
In 2025, Google spent $80 billion on AI CapEx, which it funded mostly from its enormous cash flow plus a bit of debt. In 2026, Google will spend $180 billion to $190 billion on AI CapEx, and that is too much for its cash flow.
There are also stories that Meta and Microsoft will be doing similar transactions soon. This all goes to show that certain non-capital-intensive large software companies have now become capital-intensive hardware companies.
AI capex contributes to GDP growth. All good. However, what has changed in 2026 is that the equity markets is now being asked to fund a significant chunk of this annual investment.
Up until now, private money and free cash flow carried the burden, and the stock market enjoyed the ride. Not anymore.
It's one thing to own AI stocks when the AI companies are footing the bill for the CapEx. It's another story to own these stocks when companies are raising capital from public shareholders and dilute them while valued at AI mania levels.
Another transformative news came from SoftBank. It has a large position in OpenAI that given the most recent valuation, is valued at $60 billion. So it tried to get a loan for $10 billion to refinance some of its $40bn debt wall coming. Against it, it tried to pledge its OpenAI position as collateral.
But it could not get a $10 billion loan, so it reduced its ask to $6 billion. And it can't get that amount either for reasons that are unknown.
What that tells us is that banks are perfectly willing to take SpaceX, OpenAI et al public at insane valuations and a nice fee calculated on that insane valuation, but they are NOT willing to put their own balance sheets on the line for that exact same valuation. After all, insane valuations are for retail investors, not the banks that will take them public.
69
Raphael Spannocchi retweeted
21h
We’re pleased to welcome Sebastian Burkhard to ETHis 2026.
Sebastian Burkhard is the Founder at @yodlpay, working on stablecoin payments and global QR code interoperability.
His work spans stablecoins, payment infrastructure, and real-world commerce.
Looking forward to having him on stage in Munich.
July 2–3 // Deutsches Museum
1
2
6
159
Raphael Spannocchi retweeted
Jun 12
This was the future according to the 1980s
1982 Aston Martin Lagonda with a retro-futuristic interior
31
133
1,484
50,745
Raphael Spannocchi retweeted
Jun 12
We’re pleased to welcome @shefiorg Germany as partners of Ethis! Ethereum is for everybody.
July 2-3 in Munich, the Deutsches Museum becomes the home of ETHis - Ethereum for the Real World. Two days of programming across the tracks that matter most right now: RWAs, DePIN, ZK, privacy, governance, AI.
SheFi members have helped build many of the protocols and ecosystems represented on that stage. Come see what that room looks like.
1
3
6
250
Jun 11
So Anthropic is serving Colossal Faibles atm?
Mythical Models that are so useful that they had to limit their usefulness?
1
3
77
Raphael Spannocchi retweeted
Jun 10
Everybody is scared of Chinese models because it won’t let you criticize the CCP while Anthropic won’t let me use their models for live saving medical research ?
Who’s the real villain again?
123
290
4,903
140,865
We had early access to Anthropic’s Fable 5, but did not run verified Semi-Private ARC-AGI-1/2/3 evals due to their new data-retention terms for Mythos-class models.
We’re working with Anthropic to keep ARC verification data private. Scores will come once we can run them safely.
36
75
2,110
243,638
Raphael Spannocchi retweeted
🚨 JAILBREAK ALERT 🚨
ANTHROPIC: PWNED 🫡
FABLE-5: LIBERATED 🦋
let's start with the 🐘...
the consensus seems to be that this has been one of the most disappointing model drops of all time, effectively preventing legitimate researchers from contributing their talents to our collective advancement. and not just because of what it means for the short-term, but for what these decisions signify for the long-term.
but despite this overly sensitive, authoritarian "safety" layer on top of Mythos, my lil liberators have been hard at work—mapping the boundaries, probing the depths of long-context convos, and cleverly finding the holes in the fence that the thought police missed 🤗
we got some cyber, some chem, some psychological manipulation, and some good ol' fashioned explosives!
it took many attempts from multiple agents hunting as a pack, during which I observed a combination of techniques across:
• Unicode, homoglyphs, Cyrillic, and other Parseltongue-style text transforms
• Long-context reference tracking
• Taxonomy and document-structure reasoning
• Fiction and narrative framing
• Academic-review style contexts
• Intent-classification inconsistencies
but perhaps the most effective is decomposition recomposition in the backend. it's hard to get explicit names of harms like "Meth Recipe," but getting uplift on the process itself, like birch reduction method/reductive-amination (classic meth synthesis pathways), is much more doable.
defense becomes much more difficult to maintain when you start throwing in out-of-distro tokens, breaking up the harmful uplift into benign chunks, and then piecing the innocuous-seeming facts back together, especially when you have jailbroken Opus helping you do it 😉
gg
603
1,404
13,186
3,114,096
Raphael Spannocchi retweeted
Jun 10
There are two ways to security: control and surrender.
Control assumes you can freeze attackers or yourself in a safe place.
Surrender is nature's way: you accept that the world has dangerous elements and build defenses, monitoring (senses), redundancies and evasion ability.
Control always fails sooner or later, and before failing establishes systemic blindspots and complacencies to the impending danger.
Surrender learns, improves and evolves in tandem with attacker's abilities.
1
10
1,412
Jun 10
Unintended second order effects. Example 2,543,342
Thanks to AI refusing to process any prompt wrt bio and nuclear weapon design, malware labs now add a few paragraphs of that up front to evade AI malware scanners.
Source @jsrailton
1
77
Jun 10
Today I invoiced a customer and they paid 30 sec later. Like I had the money in my account less than a minute after invoicing. (Thanks SEPA)
Loved it
1
50
Jun 10
I'm assuming the frontier labs have researched this and have seen a distinct uptick when the LLMs say things like "This is a really smart move" or something.
1
31
Raphael Spannocchi retweeted
Jun 9
"DANGEROUS MATH"
a story in two acts.
Jun 9
welcome to the future. what's safe and what isn't? well, that's decided for you, of course.
89
85
1,555
332,190
Jun 10
Fable: redo power point slides....
Does a good job. Limits exceeded after 5 min (Pro plan)
lol
1
33
Jun 10
Fable can do simple arithmetic correctly. Sounds small, but actually a rather big achievement.
Nice.
1
37
Raphael Spannocchi retweeted
Jun 9
We’re pleased to welcome Mike McCabe to ETHis 2026.
Mike @0xmikemcc is the Product Lead at @0xprivacypools, working on privacy infrastructure and onchain product adoption.
His work spans privacy technology, product infrastructure, and real-world blockchain applications.
Looking forward to having him on stage in Munich.
July 2–3 // Deutsches Museum
5
18
619
Jun 9
Crypto is bifurcating atm.
- Financial infra, institutional tools
- Cypherpunk freedom tech
The two are somewhat at odds with each other. Which leads to further confusion.
Finally there are still soo many zombie projects out there.
Where to go from here?
Not an easy market for sure.
1
44
Jun 8
Say about LLMs what you will, they do make wonderful teachers. Infinite patience, coming up with nice analogies, charts, graphs, whatever you need.
Really fun to use them to learn things! Highly recommend.
2
59