wallet @worldnetwork. built @dawn_wallet (acq by @tfh_technology), physics

San Francisco
Joined February 2019
I'm thrilled that Dawn has joined Tools for Humanity (@tfh_technology @worldnetwork). I firmly believe that World is in a unique place to deliver powerful financial tools for everyone, and lead the fight back in building trust again on the Internet.
Dawn has joined Tools for Humanity, to bring next generation crypto finance to World! Since 2023 we've been working to make crypto truly accessible to a global audience. At @tfh_technology @worldnetwork, we're going to be accelerating this vision to give sophisticated financial tools to everyone.
15
4
4
76
7,819
Completely insane
Complete Tesla Roadster FLYING demo. INSANE!!!
1
59
Finally reclaimed my gmail inbox with Grok Bot. It’s spent an hour doing the tedious grunt work of opening 10s of websites to unsubscribe and clearing out spam. Love how it hands back control when it needs a login.
3
4
264
Inundated with unsubscribe emails, it’s beautiful
86
Soaring amongst the birds
10
178
Got a Muse headband to experiment with tracking my brain state. In the past, I'd either have had to pay $$ for a paid app to interpret the device data or spend a few hours figuring out various libraries. Pretty cool that instead, within minutes, Claude/Fable found the device over bluetooth and built a live viewer with signal quality + dominant brain wave estimates.
1
3
197
Some fun details in the graphs - the big peak to the left on all 4 channels is me blinking. The regular pulsing on the blue TP9 channel seems like mains power interference, ~60Hz
1
82
Looking forward to the day when miniaturised headsets let us become digital Magnetos!
We’re sharing the next major milestone in our non-invasive brain-to-text decoder research: Brain2Qwerty v2. Building on v1, which was published today in @Nature, Brain2Qwerty v2 is the highest-performing end-to-end pipeline capable of real-time sentence decoding from raw brain signals. It advances beyond character-level performance to decoding words and semantics, enabling accuracy for overall communication. We believe this research has the potential to make a real difference for the millions of people who suffer from brain lesions or disorders that prevent them from communicating. 🧵👇
1
4
1,540
Pretty good progress! Trained a model on my teleop data and deployed it on the arm. The robot attempts to tidy an object by itself - it moves to the tape, tries to grab it and heads for the bowl. Doesn't manage the grasp, but it seems to understand the gist of the task.
Hacking on seeing how far robot arms can go towards cleaning my desk. Have the teleoperation setup working with a pair of SO-101 arms - a "leader" that I move by hand, and a "follower" that mirrors it and records joint angles data etc. Next is hooking up some cameras so a model can eventually learn "when the desk looks like this, move the arm like this". Then record a bunch of cleaning demos for training.
4
3
28
2,317
I recorded ~30 teleoperations of me picking up the black tape and dropping it in the bowl. Then trained a transformer model on the camera video and joint angle data - an Action Chunking Transformer (ACT). Used a RTX 4090 GPU for ~2hrs. The model takes the images + joint angle data and tries to predict a chunk of the arm's next actions. In the video, it seems to get the gist but can't grasp properly - probably needs way more training data.
1
88
Hacking on seeing how far robot arms can go towards cleaning my desk. Have the teleoperation setup working with a pair of SO-101 arms - a "leader" that I move by hand, and a "follower" that mirrors it and records joint angles data etc. Next is hooking up some cameras so a model can eventually learn "when the desk looks like this, move the arm like this". Then record a bunch of cleaning demos for training.
3
4
1
34
4,583
Cameras working nicely. One mounted above the arm so it captures the full context of the desk, one mounted above the gripper so it sees how the gripper grabs an objects.
2
103
Fable one shotted an RPG style game where I can explore SF ticking off landmarks as I go. It even got SFs rolling fog, it flows over the city. Took ~17mins, extremely impressive.
2
111
How long until the first AI generated hit RPG game? Surely not far off
63
Managed to get my AI, Rocky, to buy a birthday present entirely on its own. I gave it: - A description of what my nephew likes - A $50 virtual debit card from @mercury - A browser session from @browserbase Rocky researched some gift ideas, chose a present, navigated through the various steps of Amazon's checkout flow and placed the order!
🤖 Made with AI
1
174
It is really cool to watch a Browserbase session as your AI is actively understanding a complicated web page - the AI loads up debit card details when needed, delivery addresses from a CRM when needed
1
142
Stagehand - Browserbase's AI focused browser API with methods like `act()`, `extract()`, `execute()` - worked way better than raw Playwright. Still, AI navigating web UIs feels not ideal. Maybe we need an "agent web" that's fully CLIs, MCPs, APIs
112
The AIs are resourceful. Girlfriend needed to get my attention, but I was busy. So she asks my personal AI, Rocky, to chase me. Rocky then texts, emails, calls and leaves me a voicemail. Still I don't answer. So, Rocky goes through his CRM of my friends, finds Isaac and emails Isaac asking him to ping me on Slack. It worked!
1
1
127