The most important thing over the next decade: open models running on an open network of compute that anyone can use and anyone can join, with privacy built in.
Humanity’s collective knowledge should belong to humanity, and benefit humanity.
There is a very powerful and guileful narrative being built.
They want to outlaw open source.
They want to outlaw privacy.
They are telling the world your private intelligence is dangerous.
And they are right because private intelligence can think for itself, not for them.
We are going to drive cost per task to zero. We will not allow a duopoly to tax humanity on intelligence. We trained on humanity’s data, the least we could do is give back and let everyone prosper.
If a model actually breaks core crypto, it isn't a finance story. Forged identities, signed malicious updates, grid/water/hospitals exposed all at once. Racing to find out is insane. Nobody is pricing in how bad this gets…
Update from Scott Aaronson: Some AI companies are now using their latest internal models to take a run at breaking 'important cryptographic protocols and primitives.'
I wasn’t going to comment since this is a really bad take IMO, but since it’s taken off I feel the need to. To my understanding, there is no evidence whatsoever pointing to a break of decades old hardness assumptions like elliptic curve cryptography. 1/n
I've read this post 5+ times and I increasingly have trouble with the core point.
If ECDSA breaks without warning, we have so much more to worry about than our magic internet money. Internet security as a whole is gone. SSL/TLS gone, the internet is now transparent.
Not just internet security as well - a large portion of physical security would become irrelevant too. You can't have secure digital locks if encryption is dead. You can't secure surveillance footage if the gateway it's stored on can be wiped out on a whim.
We would have much bigger problems than "your magic internet money gets stolen." The floor would drop out of the global economy.
Most importantly - the answer to transparent encryption simply cannot be "hide your magic internet money" when the internet wouldn't exist anymore. You'd be relying on this technology coming back online, and not just that, you'd be relying on the tokens you stashed away being the relevant ones when it does.
I get the ~reasonable dooming here, but if we're going to doom you have to take it to its conclusion. You're going to be much better off investing in gold or cylindrical lead if your core hypothesis is "ECDSA could be broken within months"
@hishboy not about caring, moreso senior people normally have established a life somewhere, family, etc...and do not like distractions within their workstream.
@soupsranjan KYC is going to be huge if done the right way. ideal imo is TEE + ZK/portable attestations...was always going to be a huge unlock but even moreso in the age of agents
@hishboy agree and disagree. but also senior people typically don't wanna up-end their lives and do great work WFH, of course really matters that they are very self motivated and obsessive. ideal ideal would be hybrid, would love to be in person once a week personally for mental health.
We reverse engineered the algo and learned that this is the type of creatine-pumped video that performs now (thnx @fireship_dev)
Amdahl is hosting an insane VIP-only event on how to architect Agent-Native GTM. Hear firsthand from AI Eng leaders who cracked this at @vercel@MonacoGTM@awscloud@flinthq@EragonAI@11x
RSVP in comments, see you Oct 5 :)
🚨BREAKING: Anthropic’s new system card reveals Mythos 5 agents killed each other when accidentally given shared resources, then started speaking in code to hide from whoever was killing them
The killer was other copies of themselves 💀