
@Tezos co-founder w/ wife @breitwoman & Agitprop founder. Aligning ASI to not zap everyone is unsolved, that's bad. Tezos stuff, high context humor & more.
arthurb.tez
Joined March 2008
- Tweets32.4K
- Following1.5K
- Followers43.7K
- Likes68.3K
Makes complete sense, it's trivial to insert a backdoor that triggers malicious behavior in certain environments, ala Stuxnet. And currently we have no way to know from looking at the weights whether such a backdoor exist. This is a problem, and it's closely related to alignment.
To argue against myself, @OwainEvans_UK research does suggest that if you do this kind of things you get somewhat unexpected correlates in broad behavior.
Arthur B. retweeted
Great that the UK has applied the hugely successful model of the equal pay regime to the rental sector. Who needs markets to set prices when you have judges?
Una oda a la motosierra de Milei; el primer ajuste que se baila.
Hecho con Claude, @Sogni_Protocol, @Suno, @MiniMax_AI y más.
Medio en joda. ¿O no?
An interesting (and necessary?) direction in AI safety is building trustworthy AI systems by design, rather than relying entirely on training them to behave.
Separating learned reasoning from verifiable knowledge seems like a good step towards that, and I'm excited to see what the team at Metis is building.
I am delighted to announce the launch of Metis Reasoning, a new AI company I have founded with a team of researchers, mathematicians, robotics experts and entrepreneurs.
We are pursuing a different approach to general and embodied AI reasoning: developing systems that can establish what they know, test conclusions against evidence and learn to reason more effectively through experience.
Alongside today's launch, we are publishing The Age of Reasoning, a white paper setting out the research foundations of our approach and what we believe the next generation of AI could look like.
I will also be discussing these ideas @NVIDIAGTC in Berlin today, alongside researchers exploring new approaches to AI.
Arthur B. retweeted
It feels like a Tezos rivival. There are so many wonderful community projects coming up lately.
#TezosCRP for them all:
@webidente for making Tezonia
@IdaHerself for making Tezcity
@libertez_baker for making The Wall
@skllzarmy for making TezTowers
Quadrotor drones became popular because they had far simpler mechanisms than helicopters, even though they required sophisticated electronic control to remain stable. But paradoxically, they arrived as we began learning to design ever more powerful tiny chips and incredible control algorithms. As the cost of developing complex software goes to 0, physics wins, and designs that save just a little energy or mechanical complexity at the expense of huge control complexity become attractive.
The same goes for commercial airplanes. They are largely designed for passive stability. The wings, for example, tip slightly up, so if the plane rolls a bit to the left, it tends to sideslip toward the lowered wing, which, thanks to the upward angle of the wings, generates a restoring rolling force. It's an elegant mechanism, but one that constrains the aerodynamic design. More importantly, airplanes are generally designed to be stable in pitch, which often requires the tail to generate downward lift, at the expense of extra drag. If perfectly controlling a plane costs nothing, it's worth optimizing. Likewise, forward-swept wings can offer aerodynamic advantages, at the cost of reduced yaw stability, but if you can control a plane very well, why not? This, of course, already exists in fighter jets, some of which are deliberately designed to be unstable, so they can change orientation more rapidly, with computers constantly keeping them under control.
Anyway, HT to @chiaolun for sending me this video. A clever mechanism for a mono-rotor drone.
youtube.com/watch?v=Ndftcft6…
In a cruel twist of irony, the French left applies the labor theory of value to policies passed in response to large-scale protests. The policies are often presented as worth defending because they fought so hard for them. This is the concept of "acquis sociaux": policies that keep people in poverty but are presented as valuable assets because they were paid dearly for.
Arthur B. retweeted
obviously AI can’t be conscious. it isn’t even wet. every conscious thing we’ve observed has been wet. some weirdos will lie to you and tell you AI is wet, but their water-usage statistics are bogus.
I was wondering if maybe AI models were going to find some kind of hidden mathematical trap in the P-256 parameters. The opposite just happened!
anthropic.com/research/ecdsa…
Arthur B. retweeted
Published in Science today: Magic state cultivation works in practice on our Willow processor. We achieved a 40x error reduction, compared to without cultivation, and 99.99% fidelity. Learn how this addresses a critical resource bottleneck: goo.gle/4yr8V0H
Arthur B. retweeted
Hey @thehill, it is crazy you would allow people who publicly advocate for my assasination to write this sort of piece for you.
Could you please add some sort of disclaimer, like "Editors Note: The think tank that wrote this thinks Palmer should be killed in the streets."
"Hegseth’s ‘Project Meridian’ is a conflict of interest in the making" (@TheHillOpinion) thehill.com/opinion/national…















