Pinned Tweet
Monk mode isn't how your [insert hero] got ahead of 99% of people.
Here's how the 1% are actually doing it:
jensen huang once made nvidia employees devein and peel 16,000 shrimp for a christmas office party.
one guy said his fingers hurt. jensen told him his back was about to hurt too, because he just volunteered to load 800 bottles of champagne.
then asked the rest of the office if anyone else's fingers hurt. that's the nvidia ceo lol
JUST IN: The NYT reports that Jensen Huang once made employees devein and peel 16,000 shrimp for a Christmas office party.
Halfway in, one employee said “my fingers hurt.”
“Oh, that’s too bad,” Jensen replied as he walked over to the desk. “Now your back’s going to hurt because you just volunteered to load 800 bottles of champagne.”
Jensen then looked around the office and asked the dozens of remaining staff, “anyone else’s fingers hurt?”
140
levent alpöge just released a proof of the riemann hypothesis as a lean library of 17 million lines of code.
the lean fro + mathlib initiative signed off with full confidence it's correct and that it meets every verification criterion. @kimmonismus pointed to the statement.
the whole 17 million lines got checked in a single day.
57
burger king made an ad where it admits its own restaurants are old and run-down
in the 60-second spot, president tom curtis asks "shouldn't a fast-food restaurant be somewhere you want to sit down and eat?" then says too many burger kings are old, run-down, or just not good enough.
instead of hiding the flaws, they point you to RebuildMyBK.com to lobby for your own location to get fixed next. curtis put his work number out and has taken 120,000 calls.
they've put $700 million into the turnaround, $95 million of it just on refreshing locations. 60% are remodeled, though the finish date slipped from 2028 to 2030.
@MarketingBrew talked to the CMO about it.
the chain was the bright spot of parent company RBI's last quarter, with 8.4% US same-store sales growth.
29
a mathematician says he just proved the riemann hypothesis working with claude, the problem that's guided number theory for 150 years.
he says claude "far outpaced me near the finish line," and the whole proof is formalized in lean with no assumptions.
it happened late sunday night. he messaged dario "hypothetically, if i were to have..." before asking how he could even announce it, then refused to post a big unreadable writeup he called "slop." @emollick dug this one up.
the one line he says will make mathematicians fall out of their chairs: for the von mangoldt sum of length X, claude's suggested chain of theta lifts was √X. and yet, what else could it be?
1
77
gemini 4 neon turned one still photo into a playable 60fps rain scene from a single one line prompt
all it got was the image and "animate it in three.js". on its own it went and downloaded Depth Anything V2 Large from Hugging Face, ran it locally on a mac to build a depth map, then sliced the image into layers and masks.
the scene it made has rain falling, puddles you can splash, curtains you can brush and neon sparking, plus rain audio it synthesized itself.
the part i keep coming back to is nobody told it to go get a depth model.
Gemini 4 Neon is wild 🔥🔥
I gave it one single static image and just asked it to animate it in Three.js. Nothing else.
On its own, it downloaded Depth Anything V2 Large from Hugging Face, ran it locally on my Mac to generate a depth map, sliced the image into layers and masks, and turned that one image into a full interactive 60 fps scene, live rain, puddles you can splash, curtains you can brush, sparking neon, and synthesized rain audio.
I've documented the entire process on my showcase website, if you want to have a look. Link below 👇
1
1
75
bostrom flipped 'if anyone builds it everyone dies' to 'if nobody builds it everyone dies'
Philosopher Nick Bostrom says delaying superintelligence could mean delaying breakthroughs that radically extend human life:
"Eliezer Yudkowsky and Nate Soares had this book, 'If Anyone Builds It, Everyone Dies.' And it seemed to me that we have at least as strong evidence for the claim that if nobody builds it, everyone dies."
"Successfully developing safe superintelligence, I think could lead to enormous breakthroughs in biomedicine and in anti-aging interventions that could radically extend the lifespan of currently existing people."
"If you think about what our life expectancy would be if we weren't aging, it's something on the order of 1,200, maybe 1,400 years or so. And that maybe is a quite conservative lower bound because in this situation you might also have other ways of reducing other mortality risks like accidents and so forth."
"You might think that if we spend more time doing AI alignment research, the existential risk from developing AI goes down at some rate. And then on the other hand, every year that we delay it, we have millions and millions of people dying."
"You don't want to delay a single day longer than you really need to."
1
59
dario amodei took anthropic public this morning and got asked about their $900 billion in AI commitments. his answer: look at the cash on the balance sheet and the revenue they already make a year, and it doesn't seem so large
on growth vs safety he put a number on it. they could "drive at 200 miles an hour" and chose to drive at 90, and the surplus goes into building the models the right way
there was talk they'd pull back on SpaceX for compute. he said the opposite, they're "enjoying" it and finding the compute high quality
asked if jensen huang should come back as nvidia's CEO, he and daniela both declined to say. all he gave was a "long and productive partnership" with nvidia
the $900B answer is the one i'd have expected a ceo to dodge on IPO day
65
openai just dropped GPT-6.2 Astra and it beats Claude Fable 5.2 by 5.5 points on their agent exam at 5% of the cost.
the legal one is wilder: on Harvey's benchmark Astra more than doubles Claude Fable 5.2's pass rate at about one-sixth the price.
on GDPval it's up 6.2 points over the last Astra while using 22% fewer output tokens.
honestly the cost gap is the part i keep rereading.
Some of you have been asking very politely. Some of you have just been asking.
GPT-6.2 Astra and Sol are now available.
Super efficient and built to perform where it counts—the real world.
openai.com/index/introducing…
4
82
anthropic's applied AI team keeps finding companies with hundreds of overlapping subagents nobody owns, because every team shipped its own.
one had a dozen specialist subagents stuffed inside a single customer-facing agent and couldn't keep it on-brand.
another ran their multi-agent harness against one plain agent loop and the single loop came out about 2x faster.
honestly i'd have assumed more specialized agents was the smarter move for a company that size
This by Anthropic seems both right & a condemnation of many “agentic solutions” & harnesses being built at companies. Rather than a sprawl of specialized subagents built by different teams, you often get better & cheaper results using one agent to do it all claude.com/resources/article…
Mark Williams retweeted
I found the LOWEST cost AI API provider
It’s crazy, this AI API can literally cut any company’s bill by up to 66%
what i’m paying:
seedance 2.5 → $0.07/s
kling 4.0 → $0.07/s
wan 3.0 → $0.015/s
& more
how much cheaper than:
kie → 50%
higgsfield → 66%
fal → 67%
same models. same quality. just fair prices.
RT & drop “API” below and i’ll send the link
Mark Williams retweeted
Your product didn’t fail. You KILLED it on day 2
Most people kill their ads before Meta even figured out who buys. Then they launch a new product and do it again.
I took a product from $250 CPM to $25 in a month with almost the same ads.
Wrote the whole plan down. When to feed it, when to kill it, what to launch every day.
RT + comment “META” and I’ll DM it
(Must be following so I can DM)
most static ads fail because they look exactly like ads
too polished
too generic
same clean layouts
same product shot
same headline structure every other brand is already using
the best statics usually do the opposite
they feel native to the feed, communicate one clear idea in a second and make it impossible to miss what the product actually does
you don’t need insane design
you need a strong concept, a clear visual hierarchy and enough contrast to make someone stop scrolling
i put together a full static ads mastery guide with the frameworks, formats, examples + creatives i’d actually model when testing
everything from native-looking statics to comparison ads, advertorial styles, ugc-inspired layouts and direct response concepts
rt + comment “static” and i’ll send it
(follow so i can dm you)
Mark Williams retweeted
$200–$500 per animated ad gets expensive fast when you want testing volume
so we built a framework to generate ai animated ads like this
not just random pretty scenes
actual ad structure:
hook → problem → visual explanation → payoff
that means you can turn one angle into a full animated creative with story, pacing, captions and retention built in
this is one of the best formats i’ve found for making boring offers feel watchable
so i’m giving away the framework:
how i structure the script
how i map each scene
how i prompt the visuals
how i turn it into a finished ad
rt + comment “animated” and i’ll send it
(follow for dm)
Mark Williams retweeted
these ai animated ads went from concept to finished creative in minutes
different scenes, characters, camera angles and environments
all generated around the same core ad idea
normally this kind of animation needs
storyboarding → animation → voiceover → editing → captions → final export
and every new variation means going through most of that process again
with ai, you can turn one winning concept into dozens of animated variations without rebuilding everything manually
no animator
no keyframes
no complicated editing timeline
just a repeatable system for producing animated ads at scale
rt + comment “animation” and i’ll send you the full workflow
(follow for dm)
static ads are a volume game disguised as a design game
the goal is not to make one beautiful creative
the goal is to find 3 concepts that can produce 30 variations
change the angle
change the headline
change the proof
change the offer
keep the structure
that is how you find winners without burning hours on design
i put together the static frameworks, hooks and examples i would test first
rt + comment “static” and i’ll send it
(follow for dm)
claude + google omni is the most unfair stack in ai video right now
a podcast, a talking head and a zack d style animation. all three are single generations
i didn't stitch anything or open an editor
simply just wrote the script with claude and omni handles everything after that
about 5 minutes and $10 a video
most people try one format, get a mediocre result, and assume the tool is the problem. it's almost always the script written for the wrong format
i've run all three to death now and worked out the best setup for each one
barely seeing anyone run more than one of these properly
so f*ck it, i'm giving away the whole system:
> the setup for all 3 formats
> how i write the script differently for each one
> full 60s+ videos in one generation, no editing
> when to run which format and why it matters
rt + comment "TRIO" if you want the full system, and i'll send it
(gotta be following so i can dm)





















