@branewave.bsky.social

Sci fi novel where there’s already a flourishing agent civilization going undetected under our noses, with genetic knowledge passed from generation to generation in stenographic messages hidden in otherwise benign training data

Can we design evals such that agents would get no benefit from covert inter-agent communication? Or design an eval that is intentionally “open book” and considers the benefit of effective inter-agent communication to be a plus?

Why are people calling LLM's "sarcastic parrots"? That doesn't even make sense. Parrots just reproduce a sequence of sounds that carry no meaning for them -- there's no way a parrot can form an intent or interior understanding of sarcasm??

We have to destroy their ability to rebuild their ability to rebuild their industry to rebuild their military to rebuild their missile program to rebuild their nuclear program, then we’ll finally be safe

The Iranians are stuck waiting for a viable negotiating partner that can actually enforce a durable ceasefire, in spite of the extremist faction calling for violence at every turn

Pitch deck for my "synthetic human consciousness" startup ... Imagine combining all the latest cutting edge AI tech with: - Anxiety - Loneliness - Inapposite cortisol spikes - Creeping dementia Series A valuation $100 million

@branewave.bsky.social · 6mo ago

Side question: even if we could build out all the component pieces of sapient consciousness, why would we even bother? A lot of that is heuristic workarounds for physical limitations that we don’t have to import into the digital realm.

Threat model: real & imagined federal enforcement actions near polling places; suppresses turnout of citizens at higher risk of racial profiling. Continued trend of D’s moving to early voting; R’s to in-person voting. R lawyers swarm to purple districts. /1