design, open source, music
London
Joined March 2009
- Tweets920
- Following7.5K
- Followers1.5K
- Likes27.7K
Jouni Helminen retweeted
interpretability might be the most important scientific problem in the world right now, as I don't think we can solve alignment if we don't know what we're building. I think I have good news about it. I co-authored early work in mechanistic interpretability at OpenAI and came away pessimistic, because neural networks appeared to fold into arbitrary tangled knots. years later, I started openai's first large-scale automated interpretability project, finding about 2% of neurons legible with the primitive tools we had
with modern agentic tools and a much closer look, the picture looks totally different. the vast majority in models up to trillions of parameters are legible, and their circuits beautiful. many famous examples of tangled neurons turned out to be misunderstood. The roadblocks I expected mostly aren't there. I now think ambitious interpretability will be solved by the time we need it most, and possibly quite soon, as automated scientists keep improving. it may be a bitter lesson for interpretability: the simple thing, at absurd scale, works
already-clean.goodfire.com/
made a wasm modular synth app and a bunch of analog modeled modules for it. you can play with it and make your own music modular.synthcity.org/p/7UIo…
Don’t worry we gotchu - daw.synthcity.org - a fresh take on a clean, simple daw
made a rust/WASM daw that works in the browser, and a bunch of circuit modelled synths and effects for it
you can also hang out, video/voice chat and make music realtime with friends - multiplayer mode
daw.synthcity.org/?room=9141…
Jouni Helminen retweeted
It's that time of the year – Apple again uses @threejs for their new iPhone Duo, and we can learn a lot from how they did it!
And now, we have the mighty @NeedleTools Inspector, the swiss army knife for understanding and editing 3D scenes on the web 🦾
Let's take a look!
1/12
Banger
For over a decade, we’ve accepted that end-to-end backprop is the only way to train deep networks. But holding the entire network in memory all at once is why AI training is hitting a resource wall.
We found a new way to break the network into blocks and train them independently. The trick? Treating the network’s forward pass like a diffusion model denoising a signal.
This reinterpretation slashes the memory needed to train deep models. In our #ICLR2026 paper (arxiv.org/abs/2506.14202), we matched end-to-end performance across ViTs, DiTs, and LLMs. We did this while training just one isolated block at a time.
Jouni Helminen retweeted
Bookmarked something fire on X… then spent 15 minutes scrolling trying to find it again? 😩
We got you fam.
Meet BookmarkBro — a beautiful native Mac app for browsing, super-fast search, tagging, and chatting with AI about your X bookmarks.
All locally on your Mac. No privacy leaks. No new sign-ups. No cloud nonsense.
Free in beta.
More + download in the replies 👇
Jouni Helminen retweeted
There was so much talk about the protests in Iran – but they drowned in blood. The world has not helped enough the Iranian people, it has stood aside.
What will Iran become after this bloodshed? If the regime survives, it sends a clear signal to every bully – kill enough people, and you stay in power.
Jouni Helminen retweeted
🚨News of the horrendous crimes against humanity in Iran over the last 48 hours.
🚨 An estimated 7,000 killed by regime forces. Western outlets and human rights groups outside the country put it at 650 but news from inside have it closer to the number above. Many of them young girls and teenagers, I’ve seen the videos and they are too horrific to share. Yet STILL they are out on the streets in massive numbers.
This situation is unprecedented. The most remarkable story of revolution I’ve ever seen. Berlin Wall 1989 comes close - and I was there - but this is different. Very different.
The people on the streets are ordinary civilians, some of them sick, some even in wheel chairs. But the majority are young and they are fighting with all they have. And if they’re to have any kind of life ahead of them, they know there’s no way back from this. It’s a war of attrition never seen before with little to compare it to. In layman’s terms it’s do or die.
With 5 days of internet shut down and black outs so the regime can kill at will, it’s time for us to speak up. The world needs to act. We are on the precipice of something seismic.
#IranRevoIution2026
Jouni Helminen retweeted
A simple algorithm to organize your understanding of any idea or phenomenon:
1. Write down a high-level natural language description of it -- the first thing that comes to mind. Don't sweat it, keep it simple, it's fine if most of it stays vague or implicit or sounds dumb.
2. Do a new pass on what you just wrote, with a focus on replacing any vague or implicit bit with a more detailed, explicit version. Keep around the previous draft.
3. Repeat as many times as needed (typically at least 3 times).
By the time you stop you either have a description so precise it is algorithmic in nature, or you clearly understand what you don't understand.
Jouni Helminen retweeted
It’s still mind blowing to me that we can read data from a passports NFC chip, and ask the private key inside any e passport to sign a message.
Watching this team ship zero knowledge proofs of passport has been a joy.
Voting, stable coins, and identity are all going to change in 2025.
Which country will be first?
Showing @VitalikButerin what we’ve built at the @ZKpassport booth. Bring your passport and come try it out!
Jouni Helminen retweeted
The point of a startup is to make usable technology for others. When you make software, you have to watch at least 10 people use it. Sit next to them and say absolutely nothing. Force yourself to marinate in the failure of your product design.
Every version 1 of any software will be absolutely destroyed by first interaction with users. You need to watch your new creation be absolutely misunderstood by users to reform version 1 into the one that actually works.
There is only one path: figuring out where the sharp edges, the places people get caught, the assumptions you make as a builder that turn out to be wrong, and then relentlessly sanding it down so that anyone can use it.
That is good design. That is the key to a good product. There is no shortcut for this.
WATCH USERS AND CRINGE AND THEN FIX IT.
SAND DOWN THE EDGES.