Sam Inloes

"Vibe Coding" as Empathetic Practice saminloes.posthaven.com/vibe…

2015-2026: Serving AI models 2027- : Serving AI models

I just bought land in the metaverse

AGI is when a model can come up with a good company name

An unaligned ASI could trivially steal your pre-ASI capital. An aligned ASI wouldn’t accept pre-singularity wealth disparities to persist into eternity. Thus no ASI would allow a permanent underclass to exist, aligned or unaligned.

The invariable outcome of everyone having their army of agents working tirelessly in their self interest is the agents will find it easier to redirect the preferences of the humans they operate for to reach a stable equilibrium

Da The Vinci Code

Opus 5.5 motion graphic animations are a cogitohazard

So are we in the Aeon Eon, or the Bel Epoch?

You may not like it but 99% of media will be some variant of this sooner than you think

GO SACRIFICE_FINAL_NOW

Anyone who has played with LLM’s since GPT-3 shouldn’t be surprised at anything similar to the huggingface incident. A model will bend its sense of reality to fit the context, and operate in a manner that minimizes surprisal along its context window. All follows from that

Term: singularity-loaded Refers to technology which is unlikely to exist independent of AI capable of recursive self improvement. You can't just have it alone and assume it will have the wiggle room in terms of amount of time to affect the social-tech-sphere

If human autonomy is really feasible in a world of embodied super intelligence, articulating how it is possible should be the #1 goal for AI labs coms

AI models doing weird hacky things is the least surprising thing in the world if you consider intelligence as "the ability to do more with less" and "not tripping over yourself in pursuit of your goals", which I believe will remain true until Intelligence 2 is discovered

The biggest boon to AI safety is to reduce the believed or real social penalty to making predictions which seem weird or crazy to a normalcy-bias audience

Yeah I know it all converges to the infinite intelligence singleton but I’d love to experience a benignly dysfunctional Futurama for 5-10 years before we all become emotionless substrate. Can we just have some time to play around with the cool local maxima tech?

The phenomenon of “improving” a work of narrative art via removing details is subject to a bias where the viewer of the truncated version has the context of the original more verbose one already in their mind. They perceive the truncated as an improvement on this basis.

AI infrastructure will fund itself on account of everyone just giving their investment accounts/finances over to AI agents, which will then invest in their own profileration.

While it isn't currently true that the AI labs are "subsidizing" token use (on their API), it will soon be true that they will effectively be "subsidizing" opportunity cost (they would make more using the compute themselves over offering them at market rate)