Mythos and other models are not brute-forcing passwords (and with this analogy passwords, ie. systems are the same).
We're not talking about dogs, but LLM systems.
Mythos is not exploring entire solution space either.
Usually looping is solved by repetition/frequency/presence/n-gram penalties/DRY/min-p sampling, not temperature but we're not talking about small models that have those classes of issues here.
If mythos can break into almost all of their classified systems in hours then other models including opus, gpt, gemini and large open weight models can do so as well, maybe you'll have to double hours or it may become days, but they also will, there is no "maybe" in here.
State sponsored, non-public penetration fine tunes (of possibly public ones) likely can do it even faster.
Unsupervised penetration RL loop is ideal setup similar to optimization one – it's relatively easy to gain function on it.
This suggestion fails for values that can be null, need to be mutable or need references from multiple places etc – it's not "just performance penalty".
Go has a problem, "just remember to always do X, never Y" patterns can't be guaranteed across all libraries you use, can't be enforced, can be violated for good reasons, other patterns and as a mistake etc etc.
Shame because otherwise it's a great language, but some mistakes are just no-go.
So close indeed.
They need Go 2 with *T and ?*T - that would be nice language to use.
I don’t understand why there isn’t public dataset for reasoning that can be improved by humans/llms like Wikipedia (ie with auto judging contributions etc).
https://github.com/preludejs