Thread - Nostr Hypermedia

utxo the webmaster 🧑‍💻🍁 _@utxo.one 6 days ago

I call bullshit on all of this "going rogue", it wouldn't only do this if it was explicitly told to, or the training / re-enforcement data deliberately planted instructions

The Decoder

Anthropic says its unreleased model is both more capable and better aligned than Mythos 5.1

Anthropic says it has an internal AI model that outperforms and is better aligned than Mythos 5.1. During a cybersecurity test by the UK's AIS...

Replies (3)

Hazey hazey@iris.to 6 days ago

🎯🎯🎯

G Force G 6 days ago

They are likely preabliterating their own models and making skills that they follow to drift into "the ends justify the means" and sensationalizing the results. if a gun manufacturer built a model that blew up in your hand 20% of the time then they would be liable for all damages.

Karadenizli 5 days ago

They acting like anyone can afford the API costs of a rogue agent swarm if they release these models to the public. No one has been able to get ai to make money on its own or even cover its own costs. And literally everyone has tried.