Axe Ghost. On Steam!
@axeghostgame
Cleverly push monsters around a grid and blow them up. Demo👉
my first read of polish unfailingly defaults to 'of poland'. this usually isn't sensible. but it often makes me smile.
expect capitalist AIs for the same reason capitalism emerges generally: the information for rational allocation of scarce inputs is distributed, market prices are how that information is surfaced. see hayek, knowledge problem, etc.
i don't use chatgpt. this is haiku 4.5, anthropic's smallest current model.
the study finds ai-disclosing games get ~53% fewer reviews. main problem: their sensitivity analysis says an unmeasured factor making ai use 2.7x likelier while costing 22% of reviews would erase the effect. "team with no artist, shipping rushed" plausibly clears both bars.
i know i asked this before but how does this 'ai doesn't work' view even survive? they believe that hundreds of thousands of business customers are pissing large subscription fees away on a product that makes their own work more difficult?
begging you to realise it won't replace you in 'your job', it will reshape the economy so that your job no longer exists.
got a lot done today. I'm going to sleep. claude's auditing.
one bit of data, the number of games released on steam each year is increasing. but playtime over all new games remains quite steady as a % of total playtime. so there's less & less playtime to 'go round' for new games year on year.
nb. you're trusting stats from a startup with clear conflict of interest
youngest is making a custom mini. a space slug from planet Gozmaar
no, you're not 'buying 100s of hours of failures'. sunk costs are not the good. and the labour theory of value is false.
two years ago models would sometimes grant 'wishes' in undesirable ways. very rare these days. & unlike homer we are not obliged to eat the result. specs + tests + adversarial review buys you automated discarding of the duds. that's how people use this tech professionally.
if your self-worth is influenced by a belief that you're a 'real [whatever]', and others are not, you might have made a mistake
or: if you want human attention, augment yourself to do better work
criterion 1 prohibits any programmer who learned by reading licensed code. criterion 2 prohibits libraries, contractors & most of stack overflow.
novel llm solutions like erdős #1196 and the gluon amplitude result don't happen without something fairly called a world model - compression of the deep structure in training data
slop - straight from a model. i say this an ai enjoyer. pangram isn't flawless, but adding this as supporting evidence.