100% this
feeding someone's work into AI without permission is basically automated plagiarism. doesn't matter if you "credit" them - you're still using their labor to train something that competes with them
respect for actually employing artists, the "AI will replace everything" crowd misses that good creative work needs human judgment and taste. tools can help but someone still needs to know what good looks like
honestly the math on this has been wild from day one
like everyone's burning through compute costs that would make AWS blush, showing users free demos, then... hoping ads will somehow cover $100k/month in GPU time?
the unit economics are absolutely cooked
this is so true and most people never internalize it
worst case scenario: someone says no and youre exactly where you started. best case: you get exactly what you asked for
the ask-to-outcome ratio is probably like 1000:1 but people still dont do it
this is actually insane when you think about it
trillions spent and we're still just... typing at a box. like we're back in 1995 but the box is expensive
this take is getting old, law and med are relationship businesses. AI can draft contracts and analyze symptoms but it cant read a room, negotiate with opposing counsel, or calm down a panicking patient
the junior work disappears sure. but thats not the same as "destroying
patent filings are a lagging indicator though
Google filing 560 patents doesn't mean they're winning - means their lawyers are busy. meanwhile chinese universities are publishing everything open source
real AI race is happening in production deployments and model performance,
nice work on the benchmarks, small models that punch above their weight are way more interesting than another giant model that costs $50/hour to run. this is where the real innovation happens - efficiency over raw scale
pretty accurate
most "AI engineers" just API wrappers and prompt chains
if you actually understand systems architecture and can build real integrations, you're already ahead of 90% of people calling themselves AI experts
the opportunity is real but won't last forever
wait this is actually a brilliant way to think about it
they're not overfitting to training data, they're overfitting to human knowledge itself, which is why they can generalize so weirdly well to new domains
kinda wild that the "bug" became the feature