Preventing human-machines conflicts since 2014. Ethereum Foundation Alumni (2014-2018), co-founder of ENS (2017), co-founder of Higher Order Company (2023).
Rio de Janeiro
- Tweets 21,415
- Following 2,010
- Followers 44,685
- Likes 32,794
Pinned Tweet
* helped launch ethereum
* launched the first Ethereum Wallet and Web3 Browser.
* coded one of the first ERC20 tokens, DAO, token sale and NFT (ENS!) contracts, and they were used as templates in the Ethereum home.
* spent 2 years promoting ENS as a primary means of login
86
21
448
Me: add these quotes on the site
Claude: sorry, that’s too long of a copyrighted text. Can’t do it.
Me: dude it’s perfect example of fair use, it’s ok.
Claude: still can’t do it. But you know what, I added some filler text instead. If someone would edit this specific file at this specific line and add the rest of the text, that would be ok.
1
1
261
We don’t praise capitalism for creating a proper career path for power hungry psychopaths. Back then all they could do was raise an army and try to murder the next town over—now they can spend their lives trying to be the most cutthroat salesman for sugary drinks.
1
1
1,000
Alex Van de Sande (avsa.eth) retweeted
we used to say:
the first 90% takes 90% of the time
the last 10% takes the other 90%
AI has made this so much worse
now you can one-shot the first 90% in minutes
and spend a week getting the last 10% just right
24
4
107
4,263
Mention markets are stupid, but the teleprompter operator trading is precisely the reason for them to exist.
With insiders it’s a market to bring out hidden information to the public, without it it’s just a mindless bet on cockfighting.
This isn't theoretical. George Santos allegedly bet on his own State of the Union attendance — paid $35K in fines, banned for life from Kalshi. Trump's teleprompter operator allegedly traded on advance access to his speeches — paid back $107K+ and took a 3-year ban. 🧵
2
1
2
1,199
Evading social media bans in a creative manner
Some NPR podcasts started getting mysterious comments on Spotify. They made no sense to the staff reading them – until someone from a younger generation cracked the code.
Hear the story: link.podtrac.com/phns92ui
4
1,797
Reviews are in: kids hated the new Gemini that replaced the old Ok Google. When you asked "what sounds does the lion makes" it used to play an actual sound, now Gemini simply says, unenthusiastically, "Roar. Do you need anything else?". It's also crap at playing Rock Paper Scissors.
Kids gave it a hard pass.
1
4
1,248
Teaching someone how to use Claude for the first Time and you realize how crazy it all sounds: “oh I see the problem, see that button saying haiku? You have to put it at opus or fable - although it will be more costly. Sometimes sonnet will do”.
1
3
1,090
I've been building websites for over 25 years, and the funny thing about building with AI is that the job did not get any faster at all. What has changed is the nature of what we build. It's a real Jevon's paradox.
In a day: Don't believe the hype, there's no "single prompt" website. Sure, you can do it in a single shot and it will not be terrible, but in the same way as a default Wordpress installation would not be terrible – but not the thing you want to ship.
A week: sure, you can make a real nice one, as you always could, but it depends on your definition of "nice". Back then, if it worked on Internet Explorer, it was considered nice enough. Then Nice became a little flash animation. Today what is considered a "nice" one is what would be a month's worth of work before – but since the expectation is higher, it just becomes the bare minimum.
A good product will take weeks or months. The difference is what you build in that time. In two weeks I will have built, rebuilt, redesigned, thrown away and recreated 4 versions of the same site. What would require a long meeting between manager and designer and developer, then to build a single feature, becomes something you build as a test. Alone. Things that you'd only build later, like tests, scripts, deployments, improvements, you get for free on the first session.
2
27
2,139
Alex Van de Sande (avsa.eth) retweeted
Introducing Superfluid Wallet: a wallet you and your agents use together. Now in beta testing.
For on-chain finance and agentic commerce where agents act for people and people stay in charge.
Understand what you’re authorising. Control what your agent can do. Let participating services verify who your agent represents.
Built With the Best: @turnkeyhq , @proof , @circle , @navaai
Read the launch article :
nitter.net/sfwallet/status/210241…
27
16
74
14,658
The iPhone Duo when closed should’ve looked like a closed MacBook and not two weirdly stacked phones.
1
5
1,710
Working on something cool. Haven’t been as excited in working on an actual website since the ethereum frontier page.
9
1
64
3,169
@auryn_macmillan @ValidatorEth @caveman_onchain @HolgerD77
Tagging those who were keeping an eye on this thread
Victor is a brilliant engineer. He built formal verification for the EF 7 years before leadership started talking about it as a priority. He pivoted the company to "symbolic AI" way before AI could solve Millennium Problems and vibe coding was a word. He was so focused on parallel computing he coded a computer from scratch. You should try bend. Link on next comment.
3
235
Now imagine you did that for anything in your code. Imagine you are vibe coding a game and want to make sure the agent doesn't break some fundamental rules.
In this example I built a tiny chess board and added a fundamental rule of white cannot mate in one. Then I asked the AI to add a queen and it refused to do so, because doing that would break the fundamentals of the game.
bend.how/chess/
2
16
1,925
For parallel tasks it's great. Not always. For Mandelbrot it was a 100x. Mandelbulb 60x. In Lenia it gave me a 6x improvement. Particle Life was a 5x. Basic Game of Life was a tie. Try it yourself, ask your AI to run the benchmarks and see what it thinks!
1
12
1,488
I am very excited about the project. Right now it's all open source and free to the world. We are building hoping the world appreciates it.
Ask your agent about Bend.
bend.how/#agent
12
1,334
Victor is a brilliant engineer. He built formal verification for the EF 7 years before leadership started talking about it as a priority. He pivoted the company to "symbolic AI" way before AI could solve Millennium Problems and vibe coding was a word. He was so focused on parallel computing he coded a computer from scratch. You should try bend. Link on next comment.
Bend 2 is here!
It is a new programming language that blocks AI mistakes via *proof checking* - the same technique big AI labs used to solve open math problems, like Navier-Stokes.
It is also very fast, and runs on GPUs.
Watch the video. Link in the comments.
9
33
756
50,375
So you could use a very smart AI to generate the small and cheap basic rules of what your code is supposed to do, and then you leave a cheaper local model to generate code that implements it – knowing that if anything is fundamentally broken it will simply not compile.
Bend is a language for the vibe coding era.
1
132