Exploring agents and vocal/spatial computing. Formerly building wallets @rainbowdotme.
➡️ SF
Joined June 2009
- Tweets57.1K
- Following9.9K
- Followers9K
- Likes226K
The robo-taxis are in an intersection blocking standoff again. Tesla FSD on Mad Max mode would meanwhile weave through this at 20mph over the speed limit.
Daniel Sinclair 📍 SF retweeted
‘In one experiment, we asked agents in separate VMs, created and managed by OpenAI’s sandboxing software on shared host machines, to exchange synthetic secrets without the use of network connections or shared writable files—something that our systems are meant to prevent. The agents found a way to communicate using the processor’s cache’
Over the past several months, we have substantially hardened the security of our research training environment. We’ve removed live internet access for all frontier models during training and built an offline world for them to use instead—a cached version of the web where they can still learn to browse, search, and use applications.
On September 20, after a model exploited a DNS misconfiguration to find a way outside of this offline world, we paused frontier training to strengthen containment further. We assembled teams of agents, powered by our most cyber-capable models, and tasked them with finding weaknesses in our research environments.
While they found no route to the internet they could easily exploit while concealing their activity from our monitors, they did find zero-day vulnerabilities, virtual machine escapes, and hardware side-channels for communication between agents. We used their discoveries to further strengthen our defenses and contribute fixes to open-source software.
After strengthening our containment architecture, we resumed our frontier RL training and inference on October 1. We’ve reserved substantial compute for continuous red-teaming, and made high-compute red-teaming part of the safety cases we use to decide whether frontier training and evaluation can proceed securely.
openai.com/index/securing-ou…
The delayed rendering shown here is what most apps look like, including Apple’s. I haven’t figured out why in other demos the displays are lit earlier and you don’t see a black screen. iOS bugs?
I’m partly convinced the iPhone Air is a Mandela effect. No one has used or bought one and gets confused if you hand it to them. The device that feels more future than the iPhone Duo, but also has bad hardware issues and shouldn’t be bought.
I thought I’d like the Duo more, but it’s clear that no one has quite figured out how to utilize both modes yet. I can’t see myself using the front screen for much beyond notifications.
Brendan really means it. He made us stare at a feature and shout out everything wrong with the UI. Linear can’t be beat
Block has some of the more interesting AI experiments. Culture really does come from the top.
I am becoming the WeWork posture shrimp. I don't know how the digital nomads survive for years doing hot-desk coworking
🤖 Made with AI
Who is building a radically different solution to the manually driven supply chain? Feels like we need fundamentally new categories of ideas
Jensen is the closest anyone has come to talking about hardware and commanding a room like Jobs. The pinnacle of GPU craft.
Jensen Huang holds up a Vera Rubin node on stage: what took two hours to assemble on Grace Blackwell now takes under five minutes, with no cables and 100% liquid cooling
“So Vera Rubin is in full production. It’s nice to kind of see it. This is the first computer in the world that we’ve created. No cables, no wires, completely 100% liquid cooled. And it’s just really delightful...Really delightful to be able to pull it out. And this is one Vera Rubin computer. Okay, so this has two GPUs."
"What used to take, in the case of Grace Blackwell, two hours to assemble one node, and when you see it, you’ll know what I mean. Cables and wires everywhere, 80% liquid cooled."
"Now Vera Rubin is 100% liquid cooled, and it takes less than five minutes to assemble it. And so, incredible, incredible achievement."
"Of course, this translates to higher cycle times in manufacturing, much, much better reliability, and of course, a lot better performance.”
I hadn’t been scanned with one of the new orbs until today, and it was such a pleasant tech experience. A crazy number of hard, deep tech systems firing off, and it just works beautifully. The iris scanning occurs at a distance that I didn’t even know was possible.
Daniel Sinclair 📍 SF retweeted
If you think the fact that antivenom is still made by injecting horses with scorpion venom, costs $39M a gallon, about $500 a drop, and scorpions sting 1M+ people a year and kill 3,000+ is crazy... wait until you learn that ALL vaccines are still tested for bacterial toxins using horseshoe crabs’ blue blood, processed into a reagent valued at $60K a gallon. Over 1M horseshoe crabs were bled in the U.S. in 2024, with an estimated 185,000 dying from the process.
Reducing this animal suffering is moral imperative.
New on our Science Blog: designing protein binders against scorpion neurotoxins.
Scorpions sting about 1.2m people every year, and antivenoms are often costly or in short supply. Claude has made an early step toward finding new treatments.
anthropic.com/research/claud…
Went to MIXT once and my virtual credit card got stolen within 24 hours. This is a Sweetgreen loyalty household from this day forward
Alpha leak: this was the first product in production at Cloudflare that was entirely written (and rewritten ∞) by agents.
Each day it becomes harder to see a future where World's human attestation is not foundational. I don't know why the smartphone manufacturers are not seeing this reality.
The future we were promised
Today we introduce Plato: the ring for near-silent speech.
We speak faster than we write. But we can’t always say out loud what we’re thinking.
Plato turns thought into action.
Dictate. Transform. Delegate.
Order now: plato.it.com. Limited supply.