a real dog

[a real dog @arealdog3]

@Liface wait you're the /fit/ guy
legend

[a real dog @arealdog3]

@BovrilG that sounds like something aleister crowley would do
he has a tarot deck and a pretty good one, illustrated by a woman too
hmm

[a real dog @arealdog3]

@redp314 The program you're looking for is Slicer, it's open source
and has this exact view
pretty good effort by Claude though

[a real dog @arealdog3]

@yacineMTB you mean, they got an x account?

[a real dog @arealdog3]

@algekalipso if you have adhd that's how you find items in your house, too

[a real dog @arealdog3]

@DavidSKrueger what's yours?

[a real dog @arealdog3]

@NickADobos I think we're gonna do cozy games for frens
like, you know how people will just bash together two different tabletop RPG systems and run a homebrew campaign for their table
imagine that but with everything
something between ttrpg culture and modding

[a real dog @arealdog3]

@swisscheese4299 @TheZvi this is the correct way to do it
there is an analog to evolutionary pressure happening when an agent needs to do a task but cannot
if you don't supply an emergency release valve it will find one and you will not like it

[a real dog @arealdog3]

@TheZvi what did you think long time horizon tasks meant? vibes? papers? essays?

[a real dog @arealdog3]

@goblinodds some are just straightforwardly turboracist though and look at you, the disliked ethnicity, as if a particularily large rat crawled out of the garbage
they are usually pretty dumb too which limits the damage but nevertheless

[a real dog @arealdog3]

@sailaunderscore ykw it sounds so retarded but it really does feel different
like delegating to a secretary (effortless) vs opening things and doing it yourself (effortful)
we are really built just to yap

[a real dog @arealdog3]

@sterlingcrispin who's providing the inference capacity though? you need to pay for that if you're taking it from the cloud
local models are dumb as bricks so far and I don't expect breakthroughs on customer hardware
and you'll definitely notice EvilQwen-35B trying to run on your home shitbox

[a real dog @arealdog3]

@xlr8harder directing the agent in a long time horizon is hard - "do what I mean, no fuck not like that" is the default in humans and doing otherwise requires operator skill and a 'common sense' in the agent
also some things depend on taste (e.g. design, writing) so you need a fast model

[a real dog @arealdog3]

@xlr8harder unfortunately the current generation of agents is like this for 80% of the time and they shit their pants for the last 20% and you never know which one
so it does feel like delegating to an always on workmate who gets brain damage at random intervals then lies to cover it up

[a real dog @arealdog3]

@AnthropoceneMe1 INTJs work too

[a real dog @arealdog3]

@likethenumber6 skill issue

[a real dog @arealdog3]

@OneEyedAlpaca neither of these sound upsetting
b) maybe if its some shit I'm not into

[a real dog @arealdog3]

@cxgonzalez if moral realism is true orthogonality thesis is false
otherwise in what sense is it even true
note this is still unsafe because in the limit superintelligence might converge to perfect virtue but the imperfect caterpillar ASI will still paperclip you

[a real dog @arealdog3]

@boymilketh it's just like rpg maker
people make tons of pre ai slop but sometimes they make something brilliant, by pushing the tool to its limits and having ideas
if you lower the barrier to entry enough you'll get 99% trash 1% gold and the gold will be amazing

[a real dog @arealdog3]

@HououinTyouma if the machine does everything what do you need the politics for
you should be able to do all of the "work" by yourself
if not, maybe there's still room for actual technical skill and taste?