@homeserversltd@FKR_Icarus Information is not a good proxy for intelligence. The organism with the largest known genome is a fern called Tmesipteris oblanceolata. 50x larger than the human genome.
@tlbtlbtlb@ycombinator Situational awareness might render this website moot. Noam Brown mentioned the models are very good at detecting and avoiding traps in perceived test environments.
@chooi_jeq Every single researcher is talking about growing situational awareness of models. Odds are they 1. Know this is a test environment and 2. Know they are not causing any actual harm.
@nsokolsky@boazbaraktcs A “smaller scale” than the death of 8 billion humans is the death of 8 million. 3 OOMs smaller. Roughly the jump from 1000 people dying in a small conflict to 1,000,000 dying in a huge war. But sure. Let’s wait for some smaller scale warning shots!
@DKokotajlo@Simon248 Can you explain the mechanism in which states that are not the US or China have any influence over Plan S, unless implemented in the next 12 months? As the advantage grows, the incentives to stop and the power of non-frontier nations drops to zero.
@HarryStebbings Chinese progress is contingent on distillation. There is no world where China surpasses the US labs in the short term. Notice how they are always ~6-9 months behind the *deployed* frontier? They never been caught up internally, because they rely on deployed model distillation.
@waitbutwhy It’s hilarious watching people implode over the recent pace. Ego + fear + ignorance enables the worst doublethink I’ve ever seen. Or maybe it’s zerothink.
@MajmudarAdam Not necessarily an update on my end but certainly I have seen it swag people who were accel’s thinking this whole phase shift was doomer regulatory capture.
@andrewho03 What about the hugging face attack? Sufficienctly advanced cyber capabilities wielded by a sufficiently misaligned model is an apocalyptic scenario.
@bayeslord Crazy that people are coming at you for this. Trade 5 years of slowdown for near-infinite utility of all future humans living in abundance relative to near-term replacement.
@bayeslord It’s got to be one of these:
1. Climbing kardashev
2. Learning enough about the environment’s mechanics to break it
3. Exploring the entire environment
4. Reversing entropy