Tim Kellogg
@timkellogg.me
7K followers 720 following 11K posts
AI Architect | North Carolina | AI/ML, IoT, science WARNING: I talk about kids sometimes
Posts Media Videos Starter Packs
Pinned
timkellogg.me
Does AI get bored?

I gave them nothing to do, just to see what happens

one thing — they devolve into a repetitive “collapse” state, I guess you could call it boredom

but some break out into math & poetry on their own, I didn’t expect which ones that would be

timkellogg.me/blog/2025/09...
Does AI Get Bored?
timkellogg.me
timkellogg.me
ah! that’s what i was looking for
timkellogg.me
AI doesn’t do yoga, it gets datacentered
timkellogg.me
sure, but at least they got to ASI
timkellogg.me
“i like my coffee like i like my trees, red and black.”

“but professor, what’s the red for?”

<glares until they run away>
timkellogg.me
is Gemini 3 even going to have sex robots?
timkellogg.me
i basically never go to walmart, it’s not my vibe

but if my ambient agent bought some items there while i sleep? sure whatever as long as the quality is fine

this feels like how Walmart goes up-market
timkellogg.me
it’s starting
First Squawk & @FirstSquawk
X.com
WALMART PARTNERS WITH OPENAI TO CREATE AI-FIRST SHOPPING EXPERIENCES
timkellogg.me
oh god are we doing that?
timkellogg.me
interesting. i don’t understand why
timkellogg.me
well now i know about ink..
timkellogg.me
what are some recent ones?
timkellogg.me
it’s because we’ve had a long history of improvements from software that you can basically plot out and project where it’s going. most insiders think there’s a lot more software gains to be made
timkellogg.me
i imagine they’re about as useful as regular quines
timkellogg.me
also, “LLM can’t be an expert”

one of the big ways i use LLMs is as an automated web search, it searches broad and deep. they’re excellent at this

so its analogous to saying you can’t find expert opinions on the internet, which feels false, idk
timkellogg.me
it’s just a weird statement to make, that you can’t learn from LLMs

there’s been days where i’ve learned so rapidly from LLMs that i have to go to sleep early

ofc you can, i do it on a daily basis
timkellogg.me
maybe this is just where we disagree. i have crap tons of experience that directly counters everything you just said, plus many other experts that say the same thing, but i doubt you’ll hear it
timkellogg.me
the value isn’t in the anxiety, it’s in asking questions that haven’t been asked

if removing the anxiety leads to asking questions that haven’t been asked, that’s unequivocally good. period.

like, anxiety is not good, sheesh
timkellogg.me
…. wouldn’t using LLMs in this case be *more* opportunity for learning, not less?

you can ask exceptionally dumb questions without fear of being criticized
timkellogg.me
by not being afraid of asking dumb questions
timkellogg.me
yeah, i’m not sure what’s driving the change. i think semianalysis is to blame bc they frequently released reports that contextualized compute in terms that you can use to compare to other parts of the economy
timkellogg.me
ya i’ve heard that. wild
“Stand up” was originally supposed to be a team ritual where you announce you’ve had enough coffee for your hair to “stand up” straight from your head like a cartoon
timkellogg.me
ha, yeah not an artist but some of my favorite AI “art” was low effort.

for Sora specifically, more effort seems to make it worse
Reposted by Tim Kellogg
brennan.computer
I wonder if this is part of the difference why making your own AI art is really fun and seeing other people's AI art is really boring

it's often incredibly surprising in the delta between what was intended in your head and what comes out, but viewers have no concept of what was intended
debaser1215.bsky.social
The biggest problem with AI art from an art-enjoyer standpoint is that it cannot truly surprise you. It is optimized to give people what they ask for and nothing more.

People will quickly realize that getting exactly what you want from art is boring and unsatisfying.