How Can AI Improve Itself Without Losing Control? teapawt.net/tkstohlman/status/2043…
1
3
1,535
How Do You Make LLMs Safe? teapawt.net/tkstohlman/status/2043…
2
3
1,845
What Is Needed to Make Superintelligence Safe? teapawt.net/tkstohlman/status/2041…
2
3
2,158
Most AI systems today tend to be specialized for specific domains. When you switch to a new domain, it often requires rebuilding parts of the architecture or the safety layer.
We took a different approach: we built one governed architecture that works across completely different domains without requiring any redesign of the underlying system or governance. It has already been proven to work across more than 1,000 domains.
We believe this kind of cross-domain intelligence that continues to build and get smarter over time will ultimately drive the most meaningful improvements for society.
3
398
When AI experts talk about recursive self-improvement, the hard part is enabling the system to get smarter without it eventually going off the rails.
We built it so the architecture itself - along with the individual outputs and solutions - keep improving continuously, but always inside fixed, auditable governance boundaries that protect alignment and keep the system safe and under control.
This structure is what makes bounded recursive self-improvement actually safe and useful at scale.
341
This is why runtime governance is a real breakthrough for us.
The core issue with today’s AI is that it’s probabilistic - so you get drift, it sometimes misses the actual question, and it can quietly produce unsafe or misaligned outputs.
You can’t fix that by adding safety and ethics after the model has already produced an output - or even with checks partway through generation. Those checks have to run continuously during execution itself. That’s the only way advanced AI stays trustworthy at real scale.
2
784
The key shift is simple: as frontier models become more interchangeable, the real value moves to the architecture that makes them safe, auditable, and deployable in real-world systems.
Project 777 is that architecture.
We’ve been building Project 777 for a while.
Today marks another step forward: we’ve executed a $1M+ license with Atherion Bioresearch.
More here: apnews.com/press-release/ein…
3
261
Tattoo this to your brain:
"The world is a malleable place. If you know what you want, and you go for it with maximum energy and passion, the world will often reconfigure itself around you much more quickly and easily than you would think.”
– Marc Andreessen
TK Stohlman retweeted
Tattoo this to your brain:
"The world is a malleable place. If you know what you want, and you go for it with maximum energy and passion, the world will often reconfigure itself around you much more quickly and easily than you would think.”
-- marc andreessen
We need more people like @JTLonsdale supporting @elonmusk who are standing up for West! Great work Joe!
"If Elon hadn't bought Twitter, if he hadn't done this, the entire West would be at 10x the threat... our civilization works because we have free speech, because we have property rights," says @JTLonsdale "These basic values are at threat."
"X is up 25% or so year over year. Revenue's down of course because the brands all went at him and the brands are run by people who are not tied to the values of Western civilization."














