@Tezos co-founder w/ wife @breitwoman & Agitprop founder. Aligning ASI to not zap everyone is unsolved, that's bad. Tezos stuff, high context humor & more.

Joined March 2008
1,664 Photos and videos
1
2
680
Arthur B. retweeted
Big thank you to all @tezos bakers. It's comforting to know we can gather support for fast upgrades on L2! No multisigs here.
📣 Etherlink 6.4 Upgrade Status Etherlink 6.4 was successfully activated on block #45,018,602 🗳️ Thanks to all Tezos bakers who participated in the governance vote 🧑‍🍳 A full post-mortem covering the reported vulnerabilities and fixes will be published soon.
1
5
24
1,291
Arthur B. retweeted
Jun 10
know the Claude rules
71
867
16,023
397,424
Arthur B. retweeted
complaining has been unfairly demonized because the market for complaint hearers has always been vastly smaller than the market for complainers. now we can complain as much as we want in whatever detail we want and hear well reasoned feedback on it, and that’s broadly good
2
2
73
4,276
Clever.
NEW: malware developers added nuclear & biological weapons text to to their spyware. Goal? To trigger LLM safety refusals... so that their spyware wouldn't be analyzed by an AI security scanner. Cleanest practical example I can think of for why over-indexing on first order safety alignment is risky. When closed (and open) models ship with aggressive refusals, they will be sprinkled with second-order blindspots that attackers will discover...and exploit. We are only in the earliest days of attackers leveraging these features, and it wouldn't surprise me if users systems that need to handle complex cybersecurity issues demand that models be less safety-blunted. In the weeds: @SocketSecurity's post also shows why intention matters in how you design a malware analysis pipeline to avoid prompt manipulation. H/T to colleagues that shared this with me socket.dev/blog/mini-shai-hu…
1
9
164
33,206
Claude Fable aces it.
LLMs are pretty bad at coming up with original jokes, but often they can at least explain why some jokes are funny. I was playing around with a dataset of jokes and noticed one that LLMs tend not to get at all. It's not particularly funny or clever, but it relies heavily on a good theory of mind. On one run, Gemini Pro did manage to get it, but it was the only one. GPT 5.4 Pro, Opus 4.6, and Grok consistently miss it.
4
1,358
A tragicomedy in two acts I. Fable is out, yay I can audit my code for security issues! II. "Fable 5's safety measures flagged this message for cybersecurity or biology topics. They may flag safe, normal content as well. These measures let us bring you Mythos-level capability in other areas sooner, and we're working to refine them. Switched to Opus 4.8. Send feedback with /feedback or learn more"
1
7
1,042
Arthur B. retweeted
welcome to the world, Claude Fable 5!
11
21
282
108,131
Two cognitive biases: - opportunity cost neglect - loss aversion Fully internalizing the first without internalizing the second is a really bad combo.
1
1
2
740
Arthur B. retweeted
📣 @tezos bakers assemble! We need your support to pass Etherlink 6.4 during governance today. Please vote STAT. And please vote later today/tonight.
📣 Etherlink 6.4 has been injected. This is a security and liveness hardening upgrade based on issues found during internal stress tests and audits ahead of the first Tezos X kernel release. 📍 Proposal period: Fast Governance Period 401 From L1 levels #13,565,089 (June 9 11:25 UTC) and #13,569,888 (June 9 19:25 UTC). Bakers: Please check participation instructions below ⬇️
9
22
1,004
New credentials attack on Google, it's a subtle one. The email actually comes from account.google.com. It informs you that your recovery contact (an email you don't recognize) is about to reset your password and prompts you to take action. There's a link that appears to point to accounts.google.com but uses the continue URL parameter to redirect you to sites.google.com, which hosts the attacker's site. If you scroll down, you can see the bottom of the email, which just shows that someone is asking to add you as their recovery contact. The entire first part of the email is a user-controlled field in Google's system that the attacker controlled to include the malicious link and text.
27
294
1,097
85,538
Arthur B. retweeted
OpenAI joins Anthropic in thinking pausing may be needed 👀 "there should be an international organization that helps [...] make it possible for the world to take coordinated action, including slowing frontier development when needed"
Here is our current plan for OpenAI: openai.com/index/built-to-be…
16
35
348
57,706
Focus on the positive, it's important to teach kids early about going around government censors.
Keir Starmer set to impose internet restrictions on millions of Britons within days gbnews.com/politics/social-m…
3
10
53
2,652
Arthur B. retweeted
Given enough GPUs, all bugs are shallow.
7
29
5,578
Arthur B. retweeted
Tokenized commodities give investors something the commodity markets have never seen before. Exposure and access to real assets with supply constraints, industrial demand, and decades of price history… Accessible 24/7, without purchase minimums, without storage logistics. → Gold helps hedge inflation. → Uranium powers nuclear energy. → Rare-earths drive new gen technology.
5
12
137
2,552
The Jones Act, also known as the Merchant Marine Act of 1920, was signed into law by President Woodrow Wilson on June 5, 1920, 106 years ago to this day.
They finally hit pause on the Jones Act for 76 days during the energy crunch, and the numbers that came out are straight-up nuclear. In that short window, more clean petroleum products moved by tanker from the Gulf Coast to the West Coast than in the entire previous eleven years combined. The domestic Jones Act fleet stayed fully employed the whole time the foreign ships didn’t steal their work, they took the surge demand the protected fleet couldn’t handle. New routes appeared overnight that the law had blocked for decades, including direct Houston motor fuel to Hawaii and Strategic Petroleum Reserve crude reaching California for the first time in history. And when you look at who actually showed up to move the barrels? Between 74 and 85 percent were U.S. companies or vessels from close allies and trusted trading partners. The big scary China takeover everyone’s been warned about for years? Roughly 8 percent. The data ripped the mask off a century-old protection racket that’s been raising fuel prices in Hawaii, Alaska, Puerto Rico, and everywhere else while pretending to protect national security and American jobs. But here’s where it gets dangerous for the people still defending this thing the part that shows the entire insurance policy argument was never real in the first place. Full breakdown, the charts that make the old excuses look ridiculous, and what this actually means for reform? It’s all on the Substack. Link in bio. Go read it before the usual suspects try to memory-hole the numbers.
2
4
10
1,732
It's going to get worse before it gets better in terms of security bugs discovery in blockchains and defi, but I think it'll be over in about a year and the perception will lag for a while after that.
1
2
13
1,286
Armando Iannucci (‘The Death of Stalin’) and Simon Blackwell (‘Veep’) are writing the script for ‘PADDINGTON 4’. (Source: variety.com/2026/film/global…)
2
1
6
1,927
I bought my first Apple computer after being prejudiced against them for many years. I'm now postjudiced against them.
5
2
39
2,915
As to the why: a) you can't easily rent them on the cloud, the minimum time is a day, because Apple's license is fundamentally cloud hostile, so a Mac mini is cheaper to own pretty quickly b) I have been benching a lot of stark proofs as part of the R&D effort to make Tezos quantum resistant, and the very high memory bandwidth of the M4 is really well suited for this c) it's fun to build phone apps today now that you can whip them up in a day, and I wanted to share some with my friends on iOS, and Apple makes you buy Apple hardware if you ever want to do that
1
9
869
(the hardware is fine, OSX is a horrible annoyance)
2
9
688