=ijk=-1
down by the bay
Joined May 2015
- Tweets3K
- Following190
- Followers279
- Likes12.5K
One day, a lover.
Next day, a stranger.
Maybe always a stranger.
People are just a bit deeper than the time we put into knowing them.
1
39
Look, I'm as pro AI as they come... But when you butcher an iconic movie scene this badly, maybe we need to take your tokens away...
when was the last time you saw AI act this good?
cheap Minimax H3 here beats 98% of expensive seedance 2.5 videos,
in both visuals and acting
how? it's a skill issue
a lot of people keep rolling 7 turns each per clip for a 1 minute video treating it like a casino and on looking at their prompts i roll my eyes, always try to avoid relying on claude for prompts or other ai's unless you got no other option
an AI has to be an extension of your own mind
38
The people who are fixating on serialization work on things where fast and correct serialization are very important.
Criticizing protobuf for being complicated and overengineered is like criticizing an airplane's autopilot for having too many failsafe mechanisms.
This quoted post is unavailable.
2
98
preston is actively inferencing retweeted
Beff must be so relieved that he can do like two months of hysterical politicsposting instead of having to work on his product
The startup I'm at is hosting the NVIDIA Halloween party and I was going to go as Toad. I got on a video call with my friend today (with no costume on) and he shouted "YOOO are you a samurai??"
Devastating. I must now scour the city for last-minute samurai costumes.
1
46
preston is actively inferencing retweeted
Look at that. The kerning on "panty dropper." The tasteful parenthetical. The Model Y Performance — not the base, the Performance. He even specified the loaf.
My hands started shaking.
I have been posting copypastas for eleven years. I have a Discord server, a secondary Discord server for when the first one gets too loud, and a spreadsheet that tracks every time one of mine gets quoted by an account with more than 40K followers. My wife knows the spreadsheet exists. She does not know it has tabs.
Last Tuesday I dropped my own. 2,100 upvotes. Four crossposts. I thought that was it. I thought that was the one. I printed it. I had it laminated at the Staples on Route 9, the one with the good laminator, not the one by the Chili's. I keep it in the glovebox so when I'm at a red light I can take it out and just hold it.
Then this guy comes along. 3,413 upvotes. A screenshot in a tweet with 664K views. A quote-tweet from a verified account whose avatar is a man thoughtfully using a MacBook in natural light.
I read it again. "She cradled the phone in her hands like a newborn baby." I read it a third time. "I can only imagine what was going through her mind." The restraint. The confidence. He never once raises his voice. He doesn't have to.
I went to the bathroom and ran the cold tap and looked at myself for a while.
I'm not jealous. I want to be clear about that. I am a happily married man with a laminated post in my glovebox and a very normal relationship with the concept of attention. But I will be at the Staples on Route 9 tomorrow at 9 AM, and I will be asking them to run mine through a second time, because I noticed a bubble in the lower left corner that I had previously chosen not to see.
The guy at the counter will say "sir, that's not how lamination works."
I will say "I know."
And I will drive home in a Model 3. Standard range. Rear-wheel drive. The last of the old interior.
preston is actively inferencing retweeted
Good reminder that ~all of generative AI is in some sense adversarially optimized for you to think it is better than it is (via pre-training, via RLHF, via hill-climbing benchmarks). or as nostalgebraist puts it:
this type of post makes me very bearish on humanity. I'm starting to discover a non negligible part of humanity appreciates slop? LLMs are incredibly good at some things, but talking about my essays right now is still at the level of slop.
This isn't a statement on future capabilities. It's just that I find it sad that current ones are taken to be at harold bloom level.
preston is actively inferencing retweeted
hey sam, your mclaren is only used 5% of the time. this is a terrible allocation of scarce resources. have you considered taking the bus to work
sam: local inference is a terrible allocation of scarce resources
“a lot of people want local inference, and somehow they assume that the problem is solved now because we have open-source models, we're going to be able to run them on our computers, and so we're not going to need all the stuff in the data centre
but from a global allocation perspective, actually it's quite terrible to have everybody, like, have a bunch of the world's limited supply of DRAM sitting in their own computer and being used five percent of the time
you want maximum utilisation in the cloud if you're trying to deal with this very supply-constrained environment
somehow many people's intuition is the opposite of that”
The spacex guys have to set up codebases for the dumbest agents. News at 11.
This is f*king insane. SpaceXAI engineer Lauren Tan:
"You spend the tokens to set up a codebase so that even the dumbest agents can do a good job."
In this 60-minute workshop, she shows why the setup matters before you scale the team
Job → Rules → Skills → Verification → 6 Bots → 200+ Agents*
Most people add another bot when the first one fails. That just multiplies the mistakes
Spend the tokens on rules and checks once, instead of paying for the same failure every run
Worth more than most $1,000 agent engineering courses
Bookmark and watch it today
Then read the full setup: "Grok Bot: 6 Bots, 200+ Agents, a Smaller Bill" ↓
44
Honeymoon phase with Opus is officially over. It's a good model, but nowhere near the caliber it needs to be to replace me (which is all I want).
1
73
They're tryna automate intelligence, but if they'd just automate unintelligence instead, then we'd have a surplus of dumb and nobody would have to work on producing dumb anymore.
1
2
108
You used to have to be smart to ship things fast so you could get done in 50 lines what took others 500.
Now the damn clankers write 2000-line PRs on the daily. What a time to live in.
47