hi buds,
this past week, chat gpt’s ai image generation was integrated with open ai’s gpt-4o model, enhancing its ability to produce more realistic and detailed images. relatedly, they also adjusted their content moderation policies, allowing for chat gpt to generate images depicting public figures, hateful symbols, and racial features.
while this…. seems like a pretty irresponsible idea…. the immediate consequence on my timeline was actually more innocuous: images in the style of studio ghibli flooded my feed.
my immediate reaction was delight. it was pretty cute seeing pups and old friends rendered as characters distinctly in the style of miyazaki’s idiosyncratic universe. i even made a ghibli-style image of me and my wife, which she loved.
however, part of me also feels a bit sad, especially knowing miyazaki’s strongly anti stance on ai. as a kid who grew up enchanted by the worlds miyazaki built, it felt like watching magic getting flattened — something that once took years and soul now mimicked in seconds.
my friend brought up the good point that maybe a positive for miyazaki of this viral trend is that his work will reach new audiences who had never heard of him before. i personally re-watched spirited away this weekend after seeing so much miyazaki content in my feed.
however, there’s still something uneasy here. imo, it definitely reinforces the need for mechanisms that:
compensate artists when their work is used to train llm outputs, and
give artists easy opt-outs to have their work excluded from training llm outputs
my sense is that miyazaki would prefer the latter.
love you bud buds.
xxo,
dj
last week’s product launches
sick
waymo is bringing robotaxis to washington dc in 2026. self-driving taxis and the sphere are the sickest things we’ve built as americans in the past 10 years, so this makes me very stoked to have in our nation’s capitol
based
character ai added parental tools. parents get weekly usage summaries (but no chat access). this honestly feels pretty great, after seeing some v dystopian articles about kids falling in love with their ai friends
amazon launched “interests”, ai-powered personalized shopping prompts. really smart; would not be surprised to see this feature create significant impact on arpu, and become extremely optimized over time
open ai voice chat is allegedly getting better. fewer interruptions, more natural convo feel. thank god,
context dump → polished outputis a goated workflow, but the most annoying part is when your ai assistant interrupts you midway throughfacebook is bringing back a friends only newsfeed; just your friends' updates, no algo chaos. absolutely love it; v based. only a decade late!
cool if you’re a sociopath!
meta launched a 2x speed feature for reels — who does this?
theoretically cool, but honestly meh
google launched gemini 2.5 pro experimental — i struggle to keep track of all my different llm options, across claude, gemini, and anthropic. feels like
model option fatigueis real, and eventually 80% of consumers will just stick with their favorite llm, like they do their favorite search engine, leading to real brand / first-mover moatsmicrosoft 365 copilot added “deep research” tools. interestingly, their press release touts their “access to work data” (confluence, servicenow, salesforce) as a competitive advantage. but chatgpt enterprise also offers this? assume retrieval-augmented generation will just be tablestakes for all ai workflow providers?
google’s new vacation planning tools utilize ai in search to provide custom itineraries, integrated with their reviews / docs / maps features… honestly can’t imagine google designing this experience opinionated enough to make this experience really slick, but the verticalized workflow definitely exists
see you next week
xoxo,
・༓ newpixels ༓・