patrick.app

was drinking my tea and noticed how cool the dual refraction was, here’s codex + blender going thru all the states of IOR for both the glass and the tea itself.

ai has this problem where people don't know that they want ai, but ai companies go around selling ai. i've seen this by traveling around the US for 5 months trying to convince people to use ai; they just don't want it!

i think we see the effects of this problem in the fact that people don't talk about ai agents like they should; in SV we all know how absolutely bananas this technology is, but others don't talk about it.

ai companies are currently doing old, saas-style distribution. but i suspect they will find, as i have, that the best way to convince someone to use ai is to show them the work that it could be doing for them. that might mean company-specific distribution. it might mean ads created per user.

well, there's three realities ahead for any given company. either:

1. ai companies figure out how to do marketing to convince your smallco to use ai. and all these old businesses continue to exist, but now they use ai
2. PE rollups eat you because you didn't use ai; you can't figure it out so you get destroyed by a business who's using AI to get it done
3. you're in a business where ai efficiency doesn't eat your margin, and you're fine, because you're a jewelry manufacturer with only big orders and ai will help you become 2% more efficient

i spent about three weeks working in the amazon fulfillment center in Belmont, California. i worked in the grocery division, picking and coordinating orders for Amazon Fresh orders. it was a warehouse that was sorta set up like a grocery store.

1. there is still a long way to go to robotics automation. this is obvious to most people but +1!
2. you can see that amazon has "shipped" the org chart by what's efficient and what's not. picking the order is near maximal efficient. you can smell the amount of engineering hours in it. but stuff that changes from fulfillment center to fulfillment center (walking paths, people coordination, etc) is very slow. stuff that's digital is near perfect though.
3. fun one: amazon puts different flavors of yogurt (e.g.) on completely different parts of the warehouse. the strawberry yogurt might be in A12 whereas the blueberry yogurt is in N45. why? i think to prevent mispicks. impossible to mispick when they're spread all the way apart. and when you pick a set of orders, you walk all the aisles anyways.

general principle emerging: humans are offended by something that is "directly" read by humans, that is not written with humans. this explains the following:

1. why people get mad receiving AI emails
2. why everyone's okay with running code run with AI, but not okay with a website that "looks" AI generated
3. why AI images are repulsive for people
4. why managing relationships with ChatGPT is revolting

i've noticed something funny: ai images are beautiful images by 2022 standards, but most people react very poorly to them in restaurants, calendar invites, or hinge profiles. isn't that funny?

it's because it's AI that people (including me!) hate the image. not because the image itself is ugly.

this is bad news for AI image companies, because it means that their role will be tricking people that the image isn't AI. sure, it's possible, but if that's really a signal people use to tell if they love an image, we might see some new things emerge.