murat 🍥 @mayfer

i once put a space heater on craigslist and asked to be traded something open endedly and one guy offered making a walking stick / candle holder / wizards staff from rare enchanted wood and arctic fox fur and we enthusiastically talked about the design at length and when he showed up he was super awkward and i never understood what happened

1 0 3 671 0

View Details

murat 🍥 @mayfer

3 days ago

@aliceisplaying the regular way is still there nw

0 0 2 204 0

View Details

murat 🍥 @mayfer

3 days ago

@deepfates seems like a weak point of computational theories of conscioussness to me, i have yet to hear a formal theory for the computational boundary and would like to see if that's a me problem or them problem

0 0 1 212 0

View Details

murat 🍥 @mayfer

3 days ago

yeah, it's a feature of subjective experience that leads to clear questions and paradoxes that help navigate the topic imo attempted answers usually lead to clear follow up questions example; if experience is located in gpu does batched inference mean separate threads merge into one experience? if not, then why does/doesnt two LLM's talking to each other merge into one experience? is text input from the other LLM really that different from using a precomputed KV cache or using mmproj for multimodal? answers to boundary problem can usually be taken to the limit where you either start believing in donald hoffman style "every boundary you draw is a plausible conscious experience" or you start believing in some physicalist geometric topology and most in between stuff melt away which is why i like the line of thinking

1 0 2 229 0

View Details

murat 🍥 @mayfer

3 days ago

@normonics @alth0u exactly

0 0 1 23 0

View Details

murat 🍥 @mayfer

4 days ago

@alth0u i think we need to mirror civilization's nested jurisdictions online is all and i'm working on this very thing but it's hard to get it to a point where it sparks joy

2 0 5 325 0

View Details

murat 🍥 @mayfer

4 days ago

@Winterrose grey

0 0 1 98 0

View Details

murat 🍥 @mayfer

4 days ago

i (i mean gpt) ported the original Soldat to web soldatweb.com requires login with x so we know who is who. sorry the bots are too good i should nerf them

0 0 5 692 1

View Details

murat 🍥 @mayfer

4 days ago

@nuwandavek @CompleteSkeptic jev computer use is overhyped due to it only working for "Choose one from these available actions" and lacks typing or vision

1 0 1 76 0

View Details

murat 🍥 @mayfer

4 days ago

@mitsuhiko it's smarter to have a SOTA llm emit message IDs to keep than getting jev to boolean filter them

0 0 0 175 0

View Details

murat 🍥 @mayfer

4 days ago

@VictorTaelin nvm i think i get it: you have to define app code or data structures in bend for it to work. so it sounds like workflow is bend->prove->iterate->port to regular app code do i understand that correctly?

0 0 2 313 0

View Details

murat 🍥 @mayfer

4 days ago

@VictorTaelin great launch video congrats. still unclear on how Bend is aware of the board state or state of codebase wrt how it matches the laws...

1 0 3 880 0

View Details

murat 🍥 @mayfer

4 days ago

@tszzl pacing the frontier hq while waiting for the agents to finish

0 0 6 493 0

View Details

murat 🍥 @mayfer

5 days ago

@RobertHaisfield wait but jev would need matching context size then. it might be easier to just get the properly intelligent LLM to output message ID's for what to keep? timing difference won't be material to ux

0 0 7 526 0

View Details

murat 🍥 @mayfer

5 days ago

LLM's still have the completion engine soul.. in the rare occasion, a pivot token, (here likely "freed"), they double down on the wrong thing. increasingly rare with post training BUT ALWAYS THERE all you need for misalignment events is for one to start off a feedback loop (like the agents message boards) where it virally spreads to other contexts i really think the CORE of the problem hasn't changed since gpt-2 or whatever. just less common. when alignment is a "march of the .9's" statistical problem, sheer volume of use is pretty much ensured to create tail events